Proposal for Movie review system

Page 1 of 24 – Cover PageSubmission ID trn:oid:::1:2998930686
HELEN SAN NEI MAWI
ANL488_PPL01_HELEN003_HELENSANNEIMAWI.docx
PPL01
ANL488_JUL24_T10: BUSINESS ANALYTICS APPLIED PROJECT
Singapore University of Social Sciences (SUSS)
Document Details
Submission ID
trn:oid:::1:2998930686
22 Pages
Submission Date
4,349 Words
Sep 5, 2024, 9:08 PM GMT+8
25,839 Characters
Download Date
Sep 8, 2024, 4:18 PM GMT+8
File Name
ANL488_PPL01_HELEN003_HELENSANNEIMAWI.docx
File Size
167.6 KB
Page 1 of 24 – Cover Page
Submission ID trn:oid:::1:2998930686
Page 2 of 24 – AI Writing Overview
52% detected as AI
The percentage indicates the combined amount of likely AI-generated text as
well as likely AI-generated text that was also likely AI-paraphrased.
Submission ID trn:oid:::1:2998930686
Caution: Review required.
It is essential to understand the limitations of AI detection before making decisions
about a student’s work. We encourage you to learn more about Turnitin’s AI detection
capabilities before using the tool.
Detection Groups
1 AI-generated only 52%
Likely AI-generated text from a large-language model.
2 AI-generated text that was AI-paraphrased 0%
Likely AI-generated text that was likely revised using an AI-paraphrase tool
or word spinner.
Disclaimer
Our AI writing assessment is designed to help educators identify text that might be prepared by a generative AI tool. Our AI writing assessment may not always be accurate (it may misidentify
writing that is likely AI generated as AI generated and AI paraphrased or likely AI generated and AI paraphrased writing as only AI generated) so it should not be used as the sole basis for
adverse actions against a student. It takes further scrutiny and human judgment in conjunction with an organization’s application of its specific academic policies to determine whether any
academic misconduct has occurred.
Frequently Asked Questions
How should I interpret Turnitin’s AI writing percentage and false positives?
The percentage shown in the AI writing report is the amount of qualifying text within the submission that Turnitin’s AI writing
detection model determines was either likely AI-generated text from a large-language model or likely AI-generated text that was
likely revised using an AI-paraphrase tool or word spinner.
False positives (incorrectly flagging human-written text as AI-generated) are a possibility in AI models.
AI detection scores under 20%, which we do not surface in new reports, have a higher likelihood of false positives. To reduce the
likelihood of misinterpretation, no score or highlights are attributed and are indicated with an asterisk in the report (*%).
The AI writing percentage should not be the sole basis to determine whether misconduct has occurred. The reviewer/instructor
should use the percentage as a means to start a formative conversation with their student and/or use it to examine the submitted
assignment in accordance with their school’s policies.
What does ‘qualifying text’ mean?
Our model only processes qualifying text in the form of long-form writing. Long-form writing means individual sentences contained in paragraphs that make up a
longer piece of written work, such as an essay, a dissertation, or an article, etc. Qualifying text that has been determined to be likely AI-generated will be
highlighted in cyan in the submission, and likely AI-generated and then likely AI-paraphrased will be highlighted purple.
Non-qualifying text, such as bullet points, annotated bibliographies, etc., will not be processed and can create disparity between the submission highlights and the
percentage shown.
Page 2 of 24 – AI Writing Overview
Submission ID trn:oid:::1:2998930686
Page 3 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
ANL 488 PROJECT PROPOSAL
Emotion-Aware Movie Recommendation System
Submitted by
HELEN SAN NEI MAWI
PI No. : Z2010827
SCHOOL OF BUSINESS
Singapore University of Social Sciences
Presented to the Singapore University of Social Sciences
in partial fulfillment of the requirements for the
Degree of Bachelor of Science
in Business Analytics
2024
Page 3 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 4 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Table of Contents
Chapter One: Introduction ……………………………………………………………………………………………….. 1
Chapter Two: Literature Review ………………………………………………………………………………………. 6
Chapter Three: Data Understanding and Preparation …………………………………………………………. 10
Chapter Four: Proposed Modelling and Evaluation……………………………………………………………. 13
Chapter Five: Proposed Schedule ……………………………………………………………………………………. 16
References ……………………………………………………………………………………………………………………. 17
Page 1 of 22
Page 4 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 5 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Chapter One: Introduction
The rapid growth in the use of digital platforms for content consumption has fundamentally
altered how people engage with media, especially in the movie and television sectors. According
to the American Marketing Technology Laboratory (2021), this shift has been both profound and
widespread, with an alarming increase in content consumption through platforms such as Netflix,
Amazon Prime, and Disney Plus. These platforms have seamlessly integrated into the
entertainment industry, offering organized content with minimal human intervention. This
convenience, combined with the vast array of options available, has made these platforms
indispensable to modern media consumption.
To thrive in an environment abundant with options, digital streaming services rely heavily
on personalized content recommendations. These recommendations are primarily based on user
ratings, viewing histories, and basic profiles. The underlying technology typically uses algorithms
that analyze past behavior to predict future preferences. For instance, if a user has a history of
watching romantic comedies, the system will likely recommend similar movies. This approach has
proven effective in maintaining user engagement by delivering content that aligns with users’
tastes.
However, while these recommendation systems are effective to an extent, they have
significant limitations. The most notable shortcoming is that they fail to capture the richness and
dynamism of human emotions, which play a critical role in media consumption. People’s feelings
and moods are influenced by a myriad of factors, including the time of day, recent life events, and
even weather conditions (Oxford Academic, 2023). These emotional states significantly affect the
type of content a user might want to engage with at any given moment. For example, a person
1
Page 5 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 6 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
feeling sad might seek out motivational videos, while someone who is bored might prefer a
thrilling action movie. Unfortunately, most current recommendation systems do not account for
these emotional states, leading to suggestions that may be misaligned with the user’s immediate
needs.
Addressing the Gap: Integrating Sentiment Analysis and Emotion Detection
The primary goal of this project is to address this critical gap in current recommendation
systems by integrating sentiment analysis and emotion detection with traditional user ratings.
Sentiment analysis involves examining user-generated content, such as reviews or social media
posts, to determine the emotional tone behind the text. By classifying these opinions and
identifying the user’s emotional state, the proposed system aims to create a more suitable
recommendation environment that better satisfies user demands.
This approach is expected to lead to higher user loyalty. When users feel that the platform
truly understands their preferences, including their emotional needs, they are more likely to have
a positive experience and return to the platform more frequently. This personalized and
emotionally resonant interaction with the platform not only enhances the user experience but also
fosters a deeper connection between the user and the service.
The project is guided by three specific objectives:
1) Personalized Recommendations Based on Historical Preferences
The first objective is to provide personalized movie recommendations based on the user’s
historical preferences using collaborative filtering techniques. Collaborative filtering leverages
Page 2 of 22
Page 6 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 7 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
large datasets of user interactions such as ratings, likes, or viewing histories to identify patterns
and make predictions about what a user might like next (IBM, n.d., para. 2).
Collaborative filtering operates on the principle that users who have agreed in the past will
continue to agree in the future. For example, if User 1 and User 2 both rated several movies
similarly, and User 1 has seen and liked a movie that User 2 has not seen, the system might
recommend that movie to User 2. This technique is highly effective in capturing the general
preferences of users and ensuring that the recommendations align with their established tastes.
However, while collaborative filtering is a robust method for tailoring recommendations based on
past behavior, it does not consider the user’s current emotional state. This is where the second
objective comes into play.
2) Emotionally Relevant Recommendations
While collaborative filtering is effective in tailoring recommendations based on historical
preferences, it does not consider the user’s current emotional state, which can significantly
influence their content choices (Ricci, Rokach, & Shapira, 2011). To address this gap, the proposed
system integrates emotion-based filtering, which adjusts recommendations according to the user’s
real-time emotional state whether they are feeling happy, sad, excited, or something else (Cambria
& Hussain, 2015).
Emotion-based filtering works by extracting emotion tags from user reviews through
sentiment analysis (Liu, 2015). For instance, reviews that frequently express joy, excitement, or
laughter can tag a movie as “uplifting” or “fun,” while reviews mentioning sadness, tension, or fear
can tag a movie as “somber” or “intense.” When a user interacts with the recommendation system,
they can input their current mood, perhaps through an emoji-based interface (Picard, 1997).
Page 3 of 22
Page 7 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 8 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
For example, if a user indicates that they are feeling sad, the system might prioritize recommending
movies that are comforting or uplifting. Conversely, if the user is feeling excited, the system might
suggest fast-paced action films or comedies. By taking into account the user’s emotional state, the
system ensures that the content it recommends is not only relevant to their general preferences but
also resonates with their current mood, providing a more satisfying and engaging experience.
This dual approach of combining collaborative filtering with emotion-based filtering allows the
recommendation system to be both reflective of the user’s long-term preferences and responsive to
their immediate emotional needs.
3) Improving User Satisfaction
The final objective is to demonstrate that incorporating emotional awareness into
recommendation systems can significantly enhance the quality and relevance of the suggestions,
thereby improving user satisfaction. Traditional systems often overlook the user’s current
emotional state, leading to recommendations that may align with past preferences but miss the
user’s immediate needs.
Integrating emotional context makes the system more adaptive, delivering content that
resonates personally with the user (Cambria & Hussain, 2015). Research indicates that systems
capable of recognizing and responding to user emotions tend to foster greater engagement, trust,
and loyalty, which are vital in the competitive digital content market (Picard, 1997).For example,
a user who regularly watches comedies might not always be in the mood for light-hearted content.
There could be times when they feel like watching a drama or a thriller based on their current
emotional state. A recommendation system that understands and adapts to these fluctuations in
mood is likely to provide more relevant suggestions, leading to a more satisfying user experience.
Page 4 of 22
Page 8 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 9 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Furthermore, emotionally aware systems can increase user loyalty. When users feel
understood and catered to on an emotional level, they are more likely to trust the platform and
continue using it. This trust can translate into higher user retention rates and more positive wordof-mouth, both of which are crucial for the platform’s long-term success.
In conclusion, the proposed project seeks to fill a critical gap in current recommendation
systems by combining sentiment analysis and emotion detection with traditional user ratings. By
classifying user opinions and identifying their emotional states, the system will create a more
suitable recommendation environment that better satisfies user demands. This approach is
expected to lead to higher user loyalty, as users will experience a more personalized and
emotionally resonant interaction with the platform.
The specific objectives of providing personalized recommendations based on historical
preferences, offering emotionally relevant recommendations, and improving overall user
satisfaction are designed to enhance the effectiveness of the recommendation system. By
integrating collaborative filtering with emotion-based filtering, the system offers a more nuanced
and personalized approach to content recommendations, ensuring that the suggestions are not only
aligned with the user’s historical preferences but also adapt to their current emotional state. This
dual approach ultimately leads to a more satisfying and enriching user experience, positioning the
platform as a leader in the competitive digital content market.
Page 5 of 22
Page 9 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 10 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Chapter Two: Literature Review
The combination of sentiment analysis and traditional user rating in recommender systems
constitutes a major advancement in the field of personalized content delivery. Conventional
recommender systems are mostly based on fixed parameters or user characteristics and their
history of viewing or their profile. Even though such approaches proved useful to a certain extent,
they do not take into account the dynamic shifts in emotions, which play a significant role in users’
decisions.
Modern development in NLP and advances in machine learning allow to exploration of
new possibilities for the integration of emotional context within recommendation systems. For
example, Ezaldeen et al (2022) proposed a HERS which stands for a hybrid e-learning
recommendation system that uses both an adaptive user profiling technique and sentiment analysis.
Lee and Thein seized on the volatility of users’ attitudes to content and insisted on the necessity of
immediate feedback to improve the learning results and content selection.
In the same context, Bhaskaran & Marappan (2023) carried out a study on generalized
modeling in the context of personalized recommendation systems. They underlined the necessity
of introducing models that can cope with the abundance of datasets and interactions performed by
users that indicate that global solutions fail very often. They discuss how it is important to adapt
algorithms to the forms of the data and this is important when working with vital data and that
includes movie reviews.
The technical approach by Kazmaier and van Vuuren (2020) introduced a general
framework of sentiment analysis that operates on opinion-containing data to realize decisionmaking. The definition they have presented is general enough to be used in any application domain
among which recommendation systems. Because the framework operates on data gathered from
Page 6 of 22
Page 10 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 11 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
users directly, it can also identify their sentiments and analysis from reviews and posts in social
media to enhance recommendations given.
Chen et al. (2021) – in turn – decided to further pursue this notion by exploring emojis as
a means of sentiment and emotion classification. A study conducted on communication data of
software developers in which they noted that emojis could act as robust signals of users’ feelings.
This finding is directly applicable to the proposed project as it states that emojis which are quite
popular in informal digital communication can be employed to capture the real-time emotional
context in the context of a recommendation system.
These two works combined give a firm starting point for addressing the idea of
incorporating sentiment analysis and emotion detection into recommendations. They reveal the
trends towards shifting from measuring static sets of preferences to designing engaging dynamic
systems that adapt to the emotional decision-making processes.
Gap Analysis
Of all the areas, the use of real-time emotion detectors in a system of sentiment analysis
along with recommendations has not achieved much ground. A majority of researchers have used
static data to forecast the preferences of users and, in the process, overlooked the dynamism in the
emotional state of users which may significantly influence their actions. This gap is especially
significant in providing contexts such as suggesting films because the preferences of a specific
user can change with her/his mood.
Most importantly, there is little knowledge available on how the opinion mining and
analysis Affect Meter can be integrated to propose real-time emotional data such as emojis to the
recommendation system. Thus, in the studies by Ezaldeen et al. (2022) and Chen et al. (2021), they
found that attributes such as real-time feedback and emotion detection can enhance the application
Page 7 of 22
Page 11 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 12 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
of the recommendations but these features are not incorporated into a widespread recommendation
system.
This project aims to fill these gaps by designing a recommender system that must not only
consider the historic choices of a user but also his/ her current emotional state. Therefore, apart
from user review, the proposed system is to offer more accurate and recommendations based on
the detection of users’ emotions.This approach will provide a great contribution to the literature by
making it easier to develop a reactive recommendation system that will better suit the users.
Theoretical Framework
Thus, based on the literature review, the theoretical foundation of this project is based on
two concepts which are sentiment analysis with emotion detection and traditional recommendation
system. As such the following are the valid perspectives for sentiment analysis; reference is made
to Kazmaier & van Vuuren (2020)’s generic framework for sentiment analysis. Their framework
focuses on the evaluation of opinion-containing data which is particularly relevant to the problem
of determining users’ attitudes toward movies from reviews. This framework will be adjusted
depending on the overall feasibility of the project regarding how sentiment data should be applied
in conjunction with the real-time detection of emotion towards improving the recommendations of
products or services.
As a result, the project will be based on the methodologies outlined in Chen et al. (2021),
where the authors successfully tested emojis as the means of emotion recognition. Based on their
research, emojis can probably be used as barometers of users’ emotions in digital communication.
In as much as performing the task in this project, emojis will help to collect instant emotion from
the user, which will be combined with sentiment data from reviews to create a detailed user profile.
Page 8 of 22
Page 12 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 13 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
The project will also incorporate the ideas of Bhaskaran & Marappan (2023) specifically
the fact that the author focused on the necessity of the generalization of modeling in the
recommendation systems. This perspective will guide the construction of machine learning models
that have the capability of learning from the given data as well as the user actions hence making
the system responsive to the various datasets as well as contexts.
By integrating these theories into the project, the project hopes to come up with a new
recommendation system that incorporates both emotional sentimental score and traditinal
recommendation system based on rating to make better recommendations of a movie that is most
likely the user will like given the environment the user is in.
Relevance to Project
The literature reviewed offers basic information that plays an important role in building the
recommendation system depicted in this paper. According to Kazmaier and van Vuuren (2020),
sentiment analysis will be integrated, whereas, Chen et al (2021) have highlighted the inclusion of
real-time emotion detection to be at the heart of the proposed solution. All these studies in
combination point to the possibilities of integrating affective context into recommendation
systems, which is the focus of this project.
This project follows the approaches and results generated by these studies in developing
an enhanced recommendation system that is not just more customized but also addresses the user’s
current mood. This kind of recommendation system is not limited to movie recommendation, more
generally, this recommendation system can be adapted for any other domains where understanding
the emotions of users is pivotal like e-commerce, music streaming, or any kind of social media
platform.
Page 9 of 22
Page 13 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 14 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Hence, this research forms part of the background of the project in terms of offering the
theoretical foundation and the practical approaches to build a recommendation system that
incorporates sentiment analysis and emotion detection components. This literature review will
inform the development, deployment, and assessment of the system to fill the knowledge gap as
spotted in the research to contribute to the enhancement of the recommendation algorithm.
Chapter Three: Data Understanding and Preparation
The data for this project is acquired from the Kaggle website, namely Rotten Tomatoes movies
and critic reviews dataset.
Dataset Description
The dataset contains 1,130,017 entries and 8 columns, which include:
1. rotten_tomatoes_link : Links to the Rotten Tomatoes page of the movie.
2. critic_name
: The name of the critic who reviewed the movie.
3. top_critic
: A boolean indicating whether the critic is a top critic.
4. publisher_name
: The name of the publication where the review was published.
5. review_type
: Indicates whether the review was “Fresh” or “Rotten”.
6. review_score
: The score given by the critic, though it has many missing values.
7. review_date
: The date when the review was published.
8. review_content
: The text of the review.
Data Exploration
1) Distribution Analysis
review_type  Fresh
63.734439 % & Rotten 36.265561%
Page 10 of 22
Page 14 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 15 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Next, the distribution of rating which will be used in the analysis is examined and the review
score which are non numeric are will be converted to numeric value as below.
There are several missing values in the dataset are as follows:

critic_name: 18,529 missing values.

review_score: 305,936 missing values.

review_content: 65,806 missing values.
Page 11 of 22
Page 15 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 16 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
The missing values will be carefully investigated to determine the best course of action. One option
is to impute the missing values using the mean, median, or mode, depending on which method is
most appropriate. Alternatively, rows with missing values may be excluded from the analysis,
especially since a large amount of data is available. However, a thorough evaluation will be
conducted to ensure that the dataset remains feasible for analysis if all missing values are removed.
The review score column is checked and there are several format were seen as below.
The above scores will be standardized into a common format, such as converting all ratings to
a 0-10 scale. By converting all review scores to a uniform 0-10 scale, the system can ensure
consistent comparisons across all reviews, which is essential for accurate recommendation
calculations.
Text Preprocessing such as tokenization will be done to split the review text into individual
word. Followed by removing stopwords where common words such as a, the, etc will be removed.
For more cleaning punctuation removal, lowercasing and removing punctuation will be done too
in order to get a clean data.
From there word frequency will be identify and tool like TEXIBLOB , VADER basic text
analysis to identify the sentiment (positive, negative, neutral) . Next is to extract the emotion
Page 12 of 22
Page 16 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 17 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
specific tags such as joy, anger, etc from review content to enable emotion based filtering. Each
movie will be tagged with one or more emotions based on the aggregated sentiment of reviews.
This metadata will be used for emotion-aware filtering.
In addition, the rule-based system that links the user’s current emotional state (selected via
emoji) to the emotion tags of movies extracted from reviews. For instant, if the user selects
happy, prioritize movies tagged as happy, exciting, or uplifting. If the user selects sad, prioritize
movies tagged as emotional, heartwarming, or comforting.
The integration of text preprocessing, sentiment analysis, emotion extraction, and emotionaware filtering forms a robust framework for creating a highly personalized movie
recommendation system. By standardizing scores, cleaning the text data, analyzing sentiments,
and extracting emotional content, the system can offer recommendations that are not only relevant
to the user’s tastes but also to their current mood. This approach represents a significant
advancement in recommendation system technology, aligning closely with recent research in
affective computing and user experience design (Picard, 1997; Cambria et al., 2015). Such a
system is expected to enhance user satisfaction, increase engagement, and ultimately build stronger
user loyalty.
Chapter Four: Proposed Modelling and Evaluation
This chapter outlines the modeling approach and the performance measures necessary for
designing a proper movie recommendation model. The integration of Python will therefore
enhance the creation of a broad and scalable analytical environment.
Modelling
Matrix factorization technique will be used train collaborative filtering model. This
involves optimizing the latent factor matrices to minimize prediction error. Once trained, the model
Page 13 of 22
Page 17 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 18 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
can predict ratings for movies that a user hasn’t interacted with yet. These predictions form the
basis for generating personalized recommendations. The output is a list of recommended movies
tailored to the user’s preferences based on their historical ratings and interactions. These
recommendations form the foundation of the recommendation system before any emotion-based
adjustments are made.
Sentiment Analysis will then be performed to analyze the text of movie reviews and extract
emotional tags (e.g., happy, sad, exciting) that describe the emotional tone or sentiment of each
movie. These tags will be used to adjust recommendations based on the user’s current emotional
state. In this analysis, Text Preprocessing to clean the review text by performing tokenization,
stopword removal, and stemming/lemmatization. This ensures that the text is standardized and free
from noise. VADER or TextBlob will be used to determine the overall sentiment of each review
(positive, negative, neutral) and to extract specific emotional content.
The Rule-Based System will be used to adjust the list of recommended movies generated
by collaborative filtering based on the user’s current emotional state. This step ensures that the
recommendations are not only personalized but also emotionally relevant to the user at the moment
of interaction. A set of predefined rules that link the user’s selected emotional state to the emotion
tags extracted from movie reviews will be created. The rules define how recommendations should
be adjusted based on the user’s mood.
The list of recommended movies from the collaborative filtering model and the emotionally
adjusted recommendations from the rule-based system will be integrated. This involves balancing
the weight of long-term preferences with the user’s current mood. The movie will then be ranked
based on both the predicted user preference score from collaborative filtering and the relevance of
the emotion tags from the rule-based system.
Page 14 of 22
Page 18 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 19 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
This process outlines a comprehensive approach to building an Emotion-Aware Movie
Recommendation System. By training a collaborative filtering model, performing sentiment
analysis, applying a rule-based system, and combining the results, the system delivers
recommendations that are not only personalized based on user preferences but also dynamically
adjusted to reflect the user’s current emotional state. This multi-layered approach is designed to
enhance user experience by offering content that resonates both with their tastes and their
emotions.
Evaluation
Evaluation and measurement are crucial aspects of assessing the effectiveness of an
Emotion-Aware Movie Recommendation System. The model will be evaluated using below
methods.
RMSE (Root Mean Square Error): A lower RMSE indicates that the model’s predictions are
close to the actual user ratings, which is crucial for ensuring the relevance of the recommendations
(Ricci, Rokach, & Shapira, 2011).
Emotion-Specific Metrics: A higher Emotion Alignment Score indicates that the system
successfully tailors recommendations to match the user’s mood, which is a key differentiator of
emotion-aware systems from traditional recommendation systems (Cambria & Hussain, 2015).
Comparison of Baseline vs. Emotion-Aware Model
To assess the added value of incorporating emotion-awareness into the recommendation system,
it is essential to compare the performance of the emotion-aware model against a baseline model
that uses traditional collaborative filtering without emotion integration. Baseline Model
collaborative filtering model will be evaluated using traditional metrics such as RMSE and
compared with the proposed model.
Page 15 of 22
Page 19 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 20 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Incorporating a comprehensive evaluation framework that includes both traditional
recommendation metrics and emotion-specific measures ensures that the Emotion-Aware Movie
Recommendation System is robustly tested. This multi-faceted approach not only verifies the
technical accuracy of the model but also gauges its practical effectiveness in enhancing user
experience. By comparing the performance of the baseline and emotion-aware models, developers
can continuously refine the system to better meet user needs and preferences, ultimately leading
to greater user satisfaction and engagement.
Chapter Five: Proposed Schedule
Project Timeline
Timeline
Duration Tasks
1/7/24
8/7/24
8 days
Project topic and supervisor identification
19/7/24
20/8/24
30 days
Download and explore the dataset
21/8/24
6/9/24
16 days
Work on proposal
7/9/24
28/9/24
21 days
Develop the model
28/9/24
30/9/24
2 days
Prepare for oral presentation
4/10/24
18/10/24
15 days
Prepare project based on feedback from oral presentation
19/10/24 3/11/24
15 days
Preparation of final report and submission.ß
Word count : ~3800
Page 16 of 22
Page 20 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 21 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
References
Bhaskaran, S., & Marappan, R. (2023). Enhanced personalized recommendation system for
machine learning public datasets: generalized modeling, simulation, significant results, and
analysis.
International
Journal
of
Information
Technology, 15(3),
1583-1595.
https://doi.org/10.1007/s41870-023-01165-2
Chen, Z., Cao, Y., Yao, H., Lu, X., Peng, X., Mei, H., & Liu, X. (2021). Emoji-powered sentiment
and emotion detection from software developers’ communication data. ACM Transactions
on
Software
Engineering
and
Methodology
(TOSEM), 30(2),
1-48.
https://doi.org/10.1145/3424308
Ezaldeen, H., Misra, R., Bisoy, S. K., Alatrash, R., & Priyadarshini, R. (2022). A hybrid E-learning
recommendation integrating adaptive profiling and sentiment analysis. Journal of Web
Semantics, 72, 100700. https://doi.org/10.1016/j.websem.2021.100700
Kazmaier, J., & van Vuuren, J. H. (2020). A generic framework for sentiment analysis: Leveraging
opinion-bearing data to inform decision making. Decision Support Systems, 135, 113304.
https://doi.org/10.1016/j.dss.2020.113304
American Marketing Technology Laboratory. (2021, August 12). Algorithms in streaming
services. AMT Lab @ CMU. https://amt-lab.org/blog/2021/8/algorithms-in-streamingservices
Page 17 of 22
Page 21 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 22 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Oxford Academic. (2023). The influence of external factors on mood and media consumption.
Oxford
University
Press.
https://academic.oup.com/edited-volume/28365/chapter-
abstract/215245107
Steven Bird, Ewan Klein, and Edward Loper (2009). Natural Language Processing with Python.
O’Reilly Media Inc. https://www.nltk.org/book/
TextBlob.
(n.d.).
TextBlob
documentation.
Retrieved
September
1,
2024,
from
https://textblob.readthedocs.io/en/dev/
Leone, S. (2020). Rotten Tomatoes movies and critic reviews dataset [Data set]. Kaggle.
https://www.kaggle.com/datasets/stefanoleone992/rotten-tomatoes-movies-and-criticreviews-dataset
IBM. (n.d.). Collaborative filtering. https://www.ibm.com/topics/collaborative-filtering
Ricci, F., Rokach, L., & Shapira, B. (2011). Introduction to recommender systems handbook.
Springer.
Cambria, E., & Hussain, A. (2015). Sentic computing: A common-sense-based framework for
concept-level sentiment analysis. Springer.
Page 18 of 22
Page 22 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 23 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Liu, B. (2015). Sentiment analysis: Mining opinions, sentiments, and emotions. Cambridge
University Press.
Picard, R. W. (1997). Affective computing. MIT Press.
Bird, S., Klein, E., & Loper, E. (2009). Natural Language Processing with Python. O’Reilly Media.
Hutto, C. J., & Gilbert, E. (2014). VADER: A Parsimonious Rule-Based Model for Sentiment
Analysis of Social Media Text. Proceedings of the International AAAI Conference on Web
and Social Media, 8(1).
Loria,
S.
(2018).
TextBlob:
Simplified
Text
Processing.
Available
at
https://textblob.readthedocs.io/en/dev/.
Frost, J. (n.d.). Root mean square error (RMSE): How to calculate it and interpret it. Statistics by
Jim. https://statisticsbyjim.com/regression/root-mean-square-error-rmse/
Cambria, E., & White, B. (2014). Jumping NLP Curves: A Review of Natural Language
Processing Research. IEEE Computational Intelligence Magazine, 9(2), 48-57.
Cremonesi, P., Koren, Y., & Turrin, R. (2010). Performance of Recommender Algorithms on TopN Recommendation Tasks. Proceedings of the 4th ACM Conference on Recommender
Systems, 39-46.
Page 19 of 22
Page 23 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 24 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Shani, G., & Gunawardana, A. (2011). Evaluating Recommendation Systems. In Recommender
Systems Handbook (pp. 257-297). Springer.
Kim TY, Ko H, Kim SH, Kim HD. Modeling of Recommendation System Based on Emotional
Information and Collaborative Filtering. Sensors (Basel). 2021 Mar 12;21(6):1997. doi:
10.3390/s21061997. PMID: 33808989; PMCID: PMC7999638.
Page 20 of 22
Page 24 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 1 of 24 – Cover Page
Submission ID trn:oid:::1:2998930686
HELEN SAN NEI MAWI
ANL488_PPL01_HELEN003_HELENSANNEIMAWI.docx
PPL01
ANL488_JUL24_T10: BUSINESS ANALYTICS APPLIED PROJECT
Singapore University of Social Sciences (SUSS)
Document Details
Submission ID
trn:oid:::1:2998930686
22 Pages
Submission Date
4,349 Words
Sep 5, 2024, 9:08 PM GMT+8
25,839 Characters
Download Date
Sep 8, 2024, 4:18 PM GMT+8
File Name
ANL488_PPL01_HELEN003_HELENSANNEIMAWI.docx
File Size
167.6 KB
Page 1 of 24 – Cover Page
Submission ID trn:oid:::1:2998930686
Page 2 of 24 – AI Writing Overview
52% detected as AI
The percentage indicates the combined amount of likely AI-generated text as
well as likely AI-generated text that was also likely AI-paraphrased.
Submission ID trn:oid:::1:2998930686
Caution: Review required.
It is essential to understand the limitations of AI detection before making decisions
about a student’s work. We encourage you to learn more about Turnitin’s AI detection
capabilities before using the tool.
Detection Groups
1 AI-generated only 52%
Likely AI-generated text from a large-language model.
2 AI-generated text that was AI-paraphrased 0%
Likely AI-generated text that was likely revised using an AI-paraphrase tool
or word spinner.
Disclaimer
Our AI writing assessment is designed to help educators identify text that might be prepared by a generative AI tool. Our AI writing assessment may not always be accurate (it may misidentify
writing that is likely AI generated as AI generated and AI paraphrased or likely AI generated and AI paraphrased writing as only AI generated) so it should not be used as the sole basis for
adverse actions against a student. It takes further scrutiny and human judgment in conjunction with an organization’s application of its specific academic policies to determine whether any
academic misconduct has occurred.
Frequently Asked Questions
How should I interpret Turnitin’s AI writing percentage and false positives?
The percentage shown in the AI writing report is the amount of qualifying text within the submission that Turnitin’s AI writing
detection model determines was either likely AI-generated text from a large-language model or likely AI-generated text that was
likely revised using an AI-paraphrase tool or word spinner.
False positives (incorrectly flagging human-written text as AI-generated) are a possibility in AI models.
AI detection scores under 20%, which we do not surface in new reports, have a higher likelihood of false positives. To reduce the
likelihood of misinterpretation, no score or highlights are attributed and are indicated with an asterisk in the report (*%).
The AI writing percentage should not be the sole basis to determine whether misconduct has occurred. The reviewer/instructor
should use the percentage as a means to start a formative conversation with their student and/or use it to examine the submitted
assignment in accordance with their school’s policies.
What does ‘qualifying text’ mean?
Our model only processes qualifying text in the form of long-form writing. Long-form writing means individual sentences contained in paragraphs that make up a
longer piece of written work, such as an essay, a dissertation, or an article, etc. Qualifying text that has been determined to be likely AI-generated will be
highlighted in cyan in the submission, and likely AI-generated and then likely AI-paraphrased will be highlighted purple.
Non-qualifying text, such as bullet points, annotated bibliographies, etc., will not be processed and can create disparity between the submission highlights and the
percentage shown.
Page 2 of 24 – AI Writing Overview
Submission ID trn:oid:::1:2998930686
Page 3 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
ANL 488 PROJECT PROPOSAL
Emotion-Aware Movie Recommendation System
Submitted by
HELEN SAN NEI MAWI
PI No. : Z2010827
SCHOOL OF BUSINESS
Singapore University of Social Sciences
Presented to the Singapore University of Social Sciences
in partial fulfillment of the requirements for the
Degree of Bachelor of Science
in Business Analytics
2024
Page 3 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 4 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Table of Contents
Chapter One: Introduction ……………………………………………………………………………………………….. 1
Chapter Two: Literature Review ………………………………………………………………………………………. 6
Chapter Three: Data Understanding and Preparation …………………………………………………………. 10
Chapter Four: Proposed Modelling and Evaluation……………………………………………………………. 13
Chapter Five: Proposed Schedule ……………………………………………………………………………………. 16
References ……………………………………………………………………………………………………………………. 17
Page 1 of 22
Page 4 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 5 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Chapter One: Introduction
The rapid growth in the use of digital platforms for content consumption has fundamentally
altered how people engage with media, especially in the movie and television sectors. According
to the American Marketing Technology Laboratory (2021), this shift has been both profound and
widespread, with an alarming increase in content consumption through platforms such as Netflix,
Amazon Prime, and Disney Plus. These platforms have seamlessly integrated into the
entertainment industry, offering organized content with minimal human intervention. This
convenience, combined with the vast array of options available, has made these platforms
indispensable to modern media consumption.
To thrive in an environment abundant with options, digital streaming services rely heavily
on personalized content recommendations. These recommendations are primarily based on user
ratings, viewing histories, and basic profiles. The underlying technology typically uses algorithms
that analyze past behavior to predict future preferences. For instance, if a user has a history of
watching romantic comedies, the system will likely recommend similar movies. This approach has
proven effective in maintaining user engagement by delivering content that aligns with users’
tastes.
However, while these recommendation systems are effective to an extent, they have
significant limitations. The most notable shortcoming is that they fail to capture the richness and
dynamism of human emotions, which play a critical role in media consumption. People’s feelings
and moods are influenced by a myriad of factors, including the time of day, recent life events, and
even weather conditions (Oxford Academic, 2023). These emotional states significantly affect the
type of content a user might want to engage with at any given moment. For example, a person
1
Page 5 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 6 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
feeling sad might seek out motivational videos, while someone who is bored might prefer a
thrilling action movie. Unfortunately, most current recommendation systems do not account for
these emotional states, leading to suggestions that may be misaligned with the user’s immediate
needs.
Addressing the Gap: Integrating Sentiment Analysis and Emotion Detection
The primary goal of this project is to address this critical gap in current recommendation
systems by integrating sentiment analysis and emotion detection with traditional user ratings.
Sentiment analysis involves examining user-generated content, such as reviews or social media
posts, to determine the emotional tone behind the text. By classifying these opinions and
identifying the user’s emotional state, the proposed system aims to create a more suitable
recommendation environment that better satisfies user demands.
This approach is expected to lead to higher user loyalty. When users feel that the platform
truly understands their preferences, including their emotional needs, they are more likely to have
a positive experience and return to the platform more frequently. This personalized and
emotionally resonant interaction with the platform not only enhances the user experience but also
fosters a deeper connection between the user and the service.
The project is guided by three specific objectives:
1) Personalized Recommendations Based on Historical Preferences
The first objective is to provide personalized movie recommendations based on the user’s
historical preferences using collaborative filtering techniques. Collaborative filtering leverages
Page 2 of 22
Page 6 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 7 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
large datasets of user interactions such as ratings, likes, or viewing histories to identify patterns
and make predictions about what a user might like next (IBM, n.d., para. 2).
Collaborative filtering operates on the principle that users who have agreed in the past will
continue to agree in the future. For example, if User 1 and User 2 both rated several movies
similarly, and User 1 has seen and liked a movie that User 2 has not seen, the system might
recommend that movie to User 2. This technique is highly effective in capturing the general
preferences of users and ensuring that the recommendations align with their established tastes.
However, while collaborative filtering is a robust method for tailoring recommendations based on
past behavior, it does not consider the user’s current emotional state. This is where the second
objective comes into play.
2) Emotionally Relevant Recommendations
While collaborative filtering is effective in tailoring recommendations based on historical
preferences, it does not consider the user’s current emotional state, which can significantly
influence their content choices (Ricci, Rokach, & Shapira, 2011). To address this gap, the proposed
system integrates emotion-based filtering, which adjusts recommendations according to the user’s
real-time emotional state whether they are feeling happy, sad, excited, or something else (Cambria
& Hussain, 2015).
Emotion-based filtering works by extracting emotion tags from user reviews through
sentiment analysis (Liu, 2015). For instance, reviews that frequently express joy, excitement, or
laughter can tag a movie as “uplifting” or “fun,” while reviews mentioning sadness, tension, or fear
can tag a movie as “somber” or “intense.” When a user interacts with the recommendation system,
they can input their current mood, perhaps through an emoji-based interface (Picard, 1997).
Page 3 of 22
Page 7 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 8 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
For example, if a user indicates that they are feeling sad, the system might prioritize recommending
movies that are comforting or uplifting. Conversely, if the user is feeling excited, the system might
suggest fast-paced action films or comedies. By taking into account the user’s emotional state, the
system ensures that the content it recommends is not only relevant to their general preferences but
also resonates with their current mood, providing a more satisfying and engaging experience.
This dual approach of combining collaborative filtering with emotion-based filtering allows the
recommendation system to be both reflective of the user’s long-term preferences and responsive to
their immediate emotional needs.
3) Improving User Satisfaction
The final objective is to demonstrate that incorporating emotional awareness into
recommendation systems can significantly enhance the quality and relevance of the suggestions,
thereby improving user satisfaction. Traditional systems often overlook the user’s current
emotional state, leading to recommendations that may align with past preferences but miss the
user’s immediate needs.
Integrating emotional context makes the system more adaptive, delivering content that
resonates personally with the user (Cambria & Hussain, 2015). Research indicates that systems
capable of recognizing and responding to user emotions tend to foster greater engagement, trust,
and loyalty, which are vital in the competitive digital content market (Picard, 1997).For example,
a user who regularly watches comedies might not always be in the mood for light-hearted content.
There could be times when they feel like watching a drama or a thriller based on their current
emotional state. A recommendation system that understands and adapts to these fluctuations in
mood is likely to provide more relevant suggestions, leading to a more satisfying user experience.
Page 4 of 22
Page 8 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 9 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Furthermore, emotionally aware systems can increase user loyalty. When users feel
understood and catered to on an emotional level, they are more likely to trust the platform and
continue using it. This trust can translate into higher user retention rates and more positive wordof-mouth, both of which are crucial for the platform’s long-term success.
In conclusion, the proposed project seeks to fill a critical gap in current recommendation
systems by combining sentiment analysis and emotion detection with traditional user ratings. By
classifying user opinions and identifying their emotional states, the system will create a more
suitable recommendation environment that better satisfies user demands. This approach is
expected to lead to higher user loyalty, as users will experience a more personalized and
emotionally resonant interaction with the platform.
The specific objectives of providing personalized recommendations based on historical
preferences, offering emotionally relevant recommendations, and improving overall user
satisfaction are designed to enhance the effectiveness of the recommendation system. By
integrating collaborative filtering with emotion-based filtering, the system offers a more nuanced
and personalized approach to content recommendations, ensuring that the suggestions are not only
aligned with the user’s historical preferences but also adapt to their current emotional state. This
dual approach ultimately leads to a more satisfying and enriching user experience, positioning the
platform as a leader in the competitive digital content market.
Page 5 of 22
Page 9 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 10 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Chapter Two: Literature Review
The combination of sentiment analysis and traditional user rating in recommender systems
constitutes a major advancement in the field of personalized content delivery. Conventional
recommender systems are mostly based on fixed parameters or user characteristics and their
history of viewing or their profile. Even though such approaches proved useful to a certain extent,
they do not take into account the dynamic shifts in emotions, which play a significant role in users’
decisions.
Modern development in NLP and advances in machine learning allow to exploration of
new possibilities for the integration of emotional context within recommendation systems. For
example, Ezaldeen et al (2022) proposed a HERS which stands for a hybrid e-learning
recommendation system that uses both an adaptive user profiling technique and sentiment analysis.
Lee and Thein seized on the volatility of users’ attitudes to content and insisted on the necessity of
immediate feedback to improve the learning results and content selection.
In the same context, Bhaskaran & Marappan (2023) carried out a study on generalized
modeling in the context of personalized recommendation systems. They underlined the necessity
of introducing models that can cope with the abundance of datasets and interactions performed by
users that indicate that global solutions fail very often. They discuss how it is important to adapt
algorithms to the forms of the data and this is important when working with vital data and that
includes movie reviews.
The technical approach by Kazmaier and van Vuuren (2020) introduced a general
framework of sentiment analysis that operates on opinion-containing data to realize decisionmaking. The definition they have presented is general enough to be used in any application domain
among which recommendation systems. Because the framework operates on data gathered from
Page 6 of 22
Page 10 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 11 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
users directly, it can also identify their sentiments and analysis from reviews and posts in social
media to enhance recommendations given.
Chen et al. (2021) – in turn – decided to further pursue this notion by exploring emojis as
a means of sentiment and emotion classification. A study conducted on communication data of
software developers in which they noted that emojis could act as robust signals of users’ feelings.
This finding is directly applicable to the proposed project as it states that emojis which are quite
popular in informal digital communication can be employed to capture the real-time emotional
context in the context of a recommendation system.
These two works combined give a firm starting point for addressing the idea of
incorporating sentiment analysis and emotion detection into recommendations. They reveal the
trends towards shifting from measuring static sets of preferences to designing engaging dynamic
systems that adapt to the emotional decision-making processes.
Gap Analysis
Of all the areas, the use of real-time emotion detectors in a system of sentiment analysis
along with recommendations has not achieved much ground. A majority of researchers have used
static data to forecast the preferences of users and, in the process, overlooked the dynamism in the
emotional state of users which may significantly influence their actions. This gap is especially
significant in providing contexts such as suggesting films because the preferences of a specific
user can change with her/his mood.
Most importantly, there is little knowledge available on how the opinion mining and
analysis Affect Meter can be integrated to propose real-time emotional data such as emojis to the
recommendation system. Thus, in the studies by Ezaldeen et al. (2022) and Chen et al. (2021), they
found that attributes such as real-time feedback and emotion detection can enhance the application
Page 7 of 22
Page 11 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 12 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
of the recommendations but these features are not incorporated into a widespread recommendation
system.
This project aims to fill these gaps by designing a recommender system that must not only
consider the historic choices of a user but also his/ her current emotional state. Therefore, apart
from user review, the proposed system is to offer more accurate and recommendations based on
the detection of users’ emotions.This approach will provide a great contribution to the literature by
making it easier to develop a reactive recommendation system that will better suit the users.
Theoretical Framework
Thus, based on the literature review, the theoretical foundation of this project is based on
two concepts which are sentiment analysis with emotion detection and traditional recommendation
system. As such the following are the valid perspectives for sentiment analysis; reference is made
to Kazmaier & van Vuuren (2020)’s generic framework for sentiment analysis. Their framework
focuses on the evaluation of opinion-containing data which is particularly relevant to the problem
of determining users’ attitudes toward movies from reviews. This framework will be adjusted
depending on the overall feasibility of the project regarding how sentiment data should be applied
in conjunction with the real-time detection of emotion towards improving the recommendations of
products or services.
As a result, the project will be based on the methodologies outlined in Chen et al. (2021),
where the authors successfully tested emojis as the means of emotion recognition. Based on their
research, emojis can probably be used as barometers of users’ emotions in digital communication.
In as much as performing the task in this project, emojis will help to collect instant emotion from
the user, which will be combined with sentiment data from reviews to create a detailed user profile.
Page 8 of 22
Page 12 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 13 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
The project will also incorporate the ideas of Bhaskaran & Marappan (2023) specifically
the fact that the author focused on the necessity of the generalization of modeling in the
recommendation systems. This perspective will guide the construction of machine learning models
that have the capability of learning from the given data as well as the user actions hence making
the system responsive to the various datasets as well as contexts.
By integrating these theories into the project, the project hopes to come up with a new
recommendation system that incorporates both emotional sentimental score and traditinal
recommendation system based on rating to make better recommendations of a movie that is most
likely the user will like given the environment the user is in.
Relevance to Project
The literature reviewed offers basic information that plays an important role in building the
recommendation system depicted in this paper. According to Kazmaier and van Vuuren (2020),
sentiment analysis will be integrated, whereas, Chen et al (2021) have highlighted the inclusion of
real-time emotion detection to be at the heart of the proposed solution. All these studies in
combination point to the possibilities of integrating affective context into recommendation
systems, which is the focus of this project.
This project follows the approaches and results generated by these studies in developing
an enhanced recommendation system that is not just more customized but also addresses the user’s
current mood. This kind of recommendation system is not limited to movie recommendation, more
generally, this recommendation system can be adapted for any other domains where understanding
the emotions of users is pivotal like e-commerce, music streaming, or any kind of social media
platform.
Page 9 of 22
Page 13 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 14 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Hence, this research forms part of the background of the project in terms of offering the
theoretical foundation and the practical approaches to build a recommendation system that
incorporates sentiment analysis and emotion detection components. This literature review will
inform the development, deployment, and assessment of the system to fill the knowledge gap as
spotted in the research to contribute to the enhancement of the recommendation algorithm.
Chapter Three: Data Understanding and Preparation
The data for this project is acquired from the Kaggle website, namely Rotten Tomatoes movies
and critic reviews dataset.
Dataset Description
The dataset contains 1,130,017 entries and 8 columns, which include:
1. rotten_tomatoes_link : Links to the Rotten Tomatoes page of the movie.
2. critic_name
: The name of the critic who reviewed the movie.
3. top_critic
: A boolean indicating whether the critic is a top critic.
4. publisher_name
: The name of the publication where the review was published.
5. review_type
: Indicates whether the review was “Fresh” or “Rotten”.
6. review_score
: The score given by the critic, though it has many missing values.
7. review_date
: The date when the review was published.
8. review_content
: The text of the review.
Data Exploration
1) Distribution Analysis
review_type  Fresh
63.734439 % & Rotten 36.265561%
Page 10 of 22
Page 14 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 15 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Next, the distribution of rating which will be used in the analysis is examined and the review
score which are non numeric are will be converted to numeric value as below.
There are several missing values in the dataset are as follows:

critic_name: 18,529 missing values.

review_score: 305,936 missing values.

review_content: 65,806 missing values.
Page 11 of 22
Page 15 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 16 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
The missing values will be carefully investigated to determine the best course of action. One option
is to impute the missing values using the mean, median, or mode, depending on which method is
most appropriate. Alternatively, rows with missing values may be excluded from the analysis,
especially since a large amount of data is available. However, a thorough evaluation will be
conducted to ensure that the dataset remains feasible for analysis if all missing values are removed.
The review score column is checked and there are several format were seen as below.
The above scores will be standardized into a common format, such as converting all ratings to
a 0-10 scale. By converting all review scores to a uniform 0-10 scale, the system can ensure
consistent comparisons across all reviews, which is essential for accurate recommendation
calculations.
Text Preprocessing such as tokenization will be done to split the review text into individual
word. Followed by removing stopwords where common words such as a, the, etc will be removed.
For more cleaning punctuation removal, lowercasing and removing punctuation will be done too
in order to get a clean data.
From there word frequency will be identify and tool like TEXIBLOB , VADER basic text
analysis to identify the sentiment (positive, negative, neutral) . Next is to extract the emotion
Page 12 of 22
Page 16 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 17 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
specific tags such as joy, anger, etc from review content to enable emotion based filtering. Each
movie will be tagged with one or more emotions based on the aggregated sentiment of reviews.
This metadata will be used for emotion-aware filtering.
In addition, the rule-based system that links the user’s current emotional state (selected via
emoji) to the emotion tags of movies extracted from reviews. For instant, if the user selects
happy, prioritize movies tagged as happy, exciting, or uplifting. If the user selects sad, prioritize
movies tagged as emotional, heartwarming, or comforting.
The integration of text preprocessing, sentiment analysis, emotion extraction, and emotionaware filtering forms a robust framework for creating a highly personalized movie
recommendation system. By standardizing scores, cleaning the text data, analyzing sentiments,
and extracting emotional content, the system can offer recommendations that are not only relevant
to the user’s tastes but also to their current mood. This approach represents a significant
advancement in recommendation system technology, aligning closely with recent research in
affective computing and user experience design (Picard, 1997; Cambria et al., 2015). Such a
system is expected to enhance user satisfaction, increase engagement, and ultimately build stronger
user loyalty.
Chapter Four: Proposed Modelling and Evaluation
This chapter outlines the modeling approach and the performance measures necessary for
designing a proper movie recommendation model. The integration of Python will therefore
enhance the creation of a broad and scalable analytical environment.
Modelling
Matrix factorization technique will be used train collaborative filtering model. This
involves optimizing the latent factor matrices to minimize prediction error. Once trained, the model
Page 13 of 22
Page 17 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 18 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
can predict ratings for movies that a user hasn’t interacted with yet. These predictions form the
basis for generating personalized recommendations. The output is a list of recommended movies
tailored to the user’s preferences based on their historical ratings and interactions. These
recommendations form the foundation of the recommendation system before any emotion-based
adjustments are made.
Sentiment Analysis will then be performed to analyze the text of movie reviews and extract
emotional tags (e.g., happy, sad, exciting) that describe the emotional tone or sentiment of each
movie. These tags will be used to adjust recommendations based on the user’s current emotional
state. In this analysis, Text Preprocessing to clean the review text by performing tokenization,
stopword removal, and stemming/lemmatization. This ensures that the text is standardized and free
from noise. VADER or TextBlob will be used to determine the overall sentiment of each review
(positive, negative, neutral) and to extract specific emotional content.
The Rule-Based System will be used to adjust the list of recommended movies generated
by collaborative filtering based on the user’s current emotional state. This step ensures that the
recommendations are not only personalized but also emotionally relevant to the user at the moment
of interaction. A set of predefined rules that link the user’s selected emotional state to the emotion
tags extracted from movie reviews will be created. The rules define how recommendations should
be adjusted based on the user’s mood.
The list of recommended movies from the collaborative filtering model and the emotionally
adjusted recommendations from the rule-based system will be integrated. This involves balancing
the weight of long-term preferences with the user’s current mood. The movie will then be ranked
based on both the predicted user preference score from collaborative filtering and the relevance of
the emotion tags from the rule-based system.
Page 14 of 22
Page 18 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 19 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
This process outlines a comprehensive approach to building an Emotion-Aware Movie
Recommendation System. By training a collaborative filtering model, performing sentiment
analysis, applying a rule-based system, and combining the results, the system delivers
recommendations that are not only personalized based on user preferences but also dynamically
adjusted to reflect the user’s current emotional state. This multi-layered approach is designed to
enhance user experience by offering content that resonates both with their tastes and their
emotions.
Evaluation
Evaluation and measurement are crucial aspects of assessing the effectiveness of an
Emotion-Aware Movie Recommendation System. The model will be evaluated using below
methods.
RMSE (Root Mean Square Error): A lower RMSE indicates that the model’s predictions are
close to the actual user ratings, which is crucial for ensuring the relevance of the recommendations
(Ricci, Rokach, & Shapira, 2011).
Emotion-Specific Metrics: A higher Emotion Alignment Score indicates that the system
successfully tailors recommendations to match the user’s mood, which is a key differentiator of
emotion-aware systems from traditional recommendation systems (Cambria & Hussain, 2015).
Comparison of Baseline vs. Emotion-Aware Model
To assess the added value of incorporating emotion-awareness into the recommendation system,
it is essential to compare the performance of the emotion-aware model against a baseline model
that uses traditional collaborative filtering without emotion integration. Baseline Model
collaborative filtering model will be evaluated using traditional metrics such as RMSE and
compared with the proposed model.
Page 15 of 22
Page 19 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 20 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Incorporating a comprehensive evaluation framework that includes both traditional
recommendation metrics and emotion-specific measures ensures that the Emotion-Aware Movie
Recommendation System is robustly tested. This multi-faceted approach not only verifies the
technical accuracy of the model but also gauges its practical effectiveness in enhancing user
experience. By comparing the performance of the baseline and emotion-aware models, developers
can continuously refine the system to better meet user needs and preferences, ultimately leading
to greater user satisfaction and engagement.
Chapter Five: Proposed Schedule
Project Timeline
Timeline
Duration Tasks
1/7/24
8/7/24
8 days
Project topic and supervisor identification
19/7/24
20/8/24
30 days
Download and explore the dataset
21/8/24
6/9/24
16 days
Work on proposal
7/9/24
28/9/24
21 days
Develop the model
28/9/24
30/9/24
2 days
Prepare for oral presentation
4/10/24
18/10/24
15 days
Prepare project based on feedback from oral presentation
19/10/24 3/11/24
15 days
Preparation of final report and submission.ß
Word count : ~3800
Page 16 of 22
Page 20 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 21 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
References
Bhaskaran, S., & Marappan, R. (2023). Enhanced personalized recommendation system for
machine learning public datasets: generalized modeling, simulation, significant results, and
analysis.
International
Journal
of
Information
Technology, 15(3),
1583-1595.
https://doi.org/10.1007/s41870-023-01165-2
Chen, Z., Cao, Y., Yao, H., Lu, X., Peng, X., Mei, H., & Liu, X. (2021). Emoji-powered sentiment
and emotion detection from software developers’ communication data. ACM Transactions
on
Software
Engineering
and
Methodology
(TOSEM), 30(2),
1-48.
https://doi.org/10.1145/3424308
Ezaldeen, H., Misra, R., Bisoy, S. K., Alatrash, R., & Priyadarshini, R. (2022). A hybrid E-learning
recommendation integrating adaptive profiling and sentiment analysis. Journal of Web
Semantics, 72, 100700. https://doi.org/10.1016/j.websem.2021.100700
Kazmaier, J., & van Vuuren, J. H. (2020). A generic framework for sentiment analysis: Leveraging
opinion-bearing data to inform decision making. Decision Support Systems, 135, 113304.
https://doi.org/10.1016/j.dss.2020.113304
American Marketing Technology Laboratory. (2021, August 12). Algorithms in streaming
services. AMT Lab @ CMU. https://amt-lab.org/blog/2021/8/algorithms-in-streamingservices
Page 17 of 22
Page 21 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 22 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Oxford Academic. (2023). The influence of external factors on mood and media consumption.
Oxford
University
Press.
https://academic.oup.com/edited-volume/28365/chapter-
abstract/215245107
Steven Bird, Ewan Klein, and Edward Loper (2009). Natural Language Processing with Python.
O’Reilly Media Inc. https://www.nltk.org/book/
TextBlob.
(n.d.).
TextBlob
documentation.
Retrieved
September
1,
2024,
from
https://textblob.readthedocs.io/en/dev/
Leone, S. (2020). Rotten Tomatoes movies and critic reviews dataset [Data set]. Kaggle.
https://www.kaggle.com/datasets/stefanoleone992/rotten-tomatoes-movies-and-criticreviews-dataset
IBM. (n.d.). Collaborative filtering. https://www.ibm.com/topics/collaborative-filtering
Ricci, F., Rokach, L., & Shapira, B. (2011). Introduction to recommender systems handbook.
Springer.
Cambria, E., & Hussain, A. (2015). Sentic computing: A common-sense-based framework for
concept-level sentiment analysis. Springer.
Page 18 of 22
Page 22 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 23 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Liu, B. (2015). Sentiment analysis: Mining opinions, sentiments, and emotions. Cambridge
University Press.
Picard, R. W. (1997). Affective computing. MIT Press.
Bird, S., Klein, E., & Loper, E. (2009). Natural Language Processing with Python. O’Reilly Media.
Hutto, C. J., & Gilbert, E. (2014). VADER: A Parsimonious Rule-Based Model for Sentiment
Analysis of Social Media Text. Proceedings of the International AAAI Conference on Web
and Social Media, 8(1).
Loria,
S.
(2018).
TextBlob:
Simplified
Text
Processing.
Available
at
https://textblob.readthedocs.io/en/dev/.
Frost, J. (n.d.). Root mean square error (RMSE): How to calculate it and interpret it. Statistics by
Jim. https://statisticsbyjim.com/regression/root-mean-square-error-rmse/
Cambria, E., & White, B. (2014). Jumping NLP Curves: A Review of Natural Language
Processing Research. IEEE Computational Intelligence Magazine, 9(2), 48-57.
Cremonesi, P., Koren, Y., & Turrin, R. (2010). Performance of Recommender Algorithms on TopN Recommendation Tasks. Proceedings of the 4th ACM Conference on Recommender
Systems, 39-46.
Page 19 of 22
Page 23 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Page 24 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686
Shani, G., & Gunawardana, A. (2011). Evaluating Recommendation Systems. In Recommender
Systems Handbook (pp. 257-297). Springer.
Kim TY, Ko H, Kim SH, Kim HD. Modeling of Recommendation System Based on Emotional
Information and Collaborative Filtering. Sensors (Basel). 2021 Mar 12;21(6):1997. doi:
10.3390/s21061997. PMID: 33808989; PMCID: PMC7999638.
Page 20 of 22
Page 24 of 24 – AI Writing Submission
Submission ID trn:oid:::1:2998930686