1 Introduction

Artificial Intelligence is the technology that enables computers and machines to think,
learn, understand, solve problems, make decisions, create, and manage on their own.
Applications involving artificial intelligence include expert systems, NLP, speech
recognition, and machine vision. With the increasing buzz about AI, several vendors have
jumped onto the bandwagon, claiming how their offerings incorporate AI. But many of
what they call "AI" could be simply the use of machine learning algorithms. To develop AI,
there are special hardware and software to design and train machine learning algorithms.
There is no single programming language which dominates in the development of AI;
although, the popular ones that developers use include Python, R, Java, C++, and Julia.
AI systems work mostly by processing large volumes of labeled training data, looking for
correlations and patterns within the data, and using the latter to predict future outcomes.
(Cole Stryker, 2024)
(What is AI? Artificial Intelligence explained, 2024)

1.1 Explanation of the topic/AI concepts used

1.1.1 Recommendation system
Recommendation systems are a machine-learning-based algorithm, normally powered
by AI, which, with the help of Big Data, recommends products or services to customers.
These recommendations may be based on their purchase history, search history,
demographic information, among other factors. Recommendation systems are good tools
in exposing users to items that they would not have been exposed to otherwise. These
systems learn the preferences of users, their past behavior, and the characteristics of
products through interaction data in the form of clicks, views, likes, and purchases. The
personalized suggestions of recommendation systems are done by predicting what
consumers may be interested in. They find wide usage by content and product providers
aimed at helping consumers make choices over everything from books and videos to
fitness classes and clothing. (Recommendation System, 2024)
Figure 1: recommendation system

1.2 Explanation/Introduction of the chosen problem domain/topic

Modern technologies have turned the scale, diversity, and velocity of data production
upside down, hence ushering in the big data era by digitizing even day-to-day activities.
With this explosion in data growth, a challenge called information overload has come into
being, a condition wherein a person becomes incapable of handling such a great amount
of data to process it for decision-making. Data mining techniques are crucial in resolving
this problem by extracting and processing data relevant to reducing information overload.
Various data mining techniques are employed, and one of the most utilized tools is a
recommender system. In recommender systems, the various available information is
analyzed to identify patterns in user behavior, after which personalized suggestions are
offered. Recommender systems recommend finding the most appropriate products or
services matching users' preferences or feature needs on the basis of very poor
information, thus simplifying their search process. (Movie Recommender Systems:
Concepts, Methods, Challenges, and Future Directions, 2022)
Recommendations concerning a product preferred by customers are highly effective for
improving user experience, thus driving business outcomes. Given that this task should
be automated and optimized, machine learning will come in handy. Among them, the
movie recommendation system is one of the useful tools for movie buffs: it allows users
to choose movies according to their taste without wasting too much time on irrelevant
films. The recommendation of products according to customer preferences plays a crucial
role in the business strategy, and it has found an effective solution in machine learning. A
movie recommendation system aims to provide movie enthusiasts with a time-saving way
of finding films they would enjoy, eliminating much time spent on unnecessary browsing.
This project focuses on integrating cosine similarity and sentiment analysis to build such
a system. Cosine similarity measures the amount of similarity between two items, while
sentiment analysis checks for the emotional tone of movie reviews to classify them as
positive or negative, thus giving a general film rating. This automates the process of
classification of reviews as favorable and unfavorable based on training and testing data.
In that respect, comparing various content-based approaches over time would result in
increasingly more precise results, hence making recommendations more exact. (Raja
Marappan, 2022)
2 Background
2.1 Research work done on the chosen topic/problem domain
2.1.1 Movie recommendation system
The "movie recommender systems" are intelligent algorithms that make film
recommendations based on users' interests in viewing and previous viewing history. In
developing personalized recommendations, the systems take into consideration data
such as user ratings, reviews, and preferences of movies. Movie recommendation
systems have totally changed how people find and watch movies; it's much easier for
consumers to browse through enormous movie collections.
The two main types of recommendation systems are collaborative and content-based
filtering. While collaborative filtering utilizes other customers' overlapping movie ratings,
content-based movie recommendation system algorithms use the similarities between
movies to recommend new films to customers. All things said, the movie recommender
system has evolved into a crucial tool for movie watchers to discover new movies.
(Sharma, 2024)

2.1.2 Research done in content-based filtering

Content-based filtering only depends on attributes and metadata of the movies to
generate recommendations of similar characteristics. The example of such a movie
recommendation system includes filtering data about genre, director, cast, or plot to offer
movies that represent similar genres, actors, or themes. The very critical power of
content-based filtering lies in its reliability for producing effective recommendations
despite having minimal or no data about a certain user. But if the movie metadata is
improperly labeled, misleading, or just too scant in detail, its effectiveness is
compromised. (Sharma, 2024)
Example of content-based filtering: If I am a fan of the Harry Potter series and have only
watched fantasy movies online. The platforms like Google or Wikipedia will analyze my
data like previous searches and recognize my preference for the fantasy genre. Based
on this information, my recommendations will mostly be fantasy films. For instance, if two
movies are considered “Fantastic Beasts” and “Shawshank Redemption”, because of my
interest in fantasy films, I would be more likely to receive a recommendation for “Fantastic
Beasts”.
(Mishra, 2021)

Figure 2: content based filtering

2.1.3 Research done in collaborative filtering

This technique of collaborative filtering, it finds out users with similar behavior patterns
and recommends new items based on shared characteristics of their group. It relies on
the analysis of the manner in which other users with similar preferences and actions have
interacted with certain items. In other words, collaborative filtering algorithms group users
based on their behavior and make personalized recommendations using the overall traits
of the group. The idea behind every collaborative recommender system is based upon
the supposition that a user will be interested in those things in which others with similar
behaviors are interested. (Jacob Murel Ph.D., 2024)
Figure 3: collaborative filtering
There are two types of collaborative filtering. They are as follows:
i. User based collaborative filtering
Collaborative filtering can gather information that users might not otherwise supply
because it obtains its results from implicit data. The user-based approach is the first
category of collaborative filtering. With the aid of collaborative filtering, this method
reduces the number of individuals that share contacts, comparable consumer
behaviors, and similar demographics. This method is used by social networking sites
to suggest people to other users based on their behavioral patterns. Additionally, this
strategy is used for targeted advertisements and product recommendations based on
the tastes and choices of other users. Among the different collaborative filtering
techniques, user-based collaborative filtering was the first to be developed. An
example of this approach is Facebook's "Suggested Friends" feature, which
recommends people users may know by analyzing their virtual connections and
shared preferences. (Rawat, 2021)

ii. Item based collaborative filtering

Item-based collaborative filtering is a subset of collaborative filtering techniques that
focuses on recommending items or products by considering users' past interactions
and historical data. This works by assessing the similarities between products and
predicting ratings for each, using the history of user interactions to suggest relevant
items. It was first implemented by Amazon in 1998 and finds huge applications in e-
commerce sites like Amazon and Flipkart for recommending similar products to users
based on their previous activity. This approach applies the Nearest Neighbor statistical
technique, with effective delivery of accurate recommendations that have been proven
to drive increased consumption. (Rawat, 2021)
2.1.4 Research done in hybrid recommendation system
A hybrid recommendation framework makes a combination of CBRS and CFRS to
improve the power and mitigate the limitations that existed in traditional recommendation
approaches. This approach integrates the user side information such as comments,
ratings, connections, attributes, reviews together with the rating matrix, which enhances
the accuracy of the recommendations and boosts the power of recommendations. Hybrid
system comes with several advantages. (Krupa Patel, 2020). They help solve the cold
start problem, where there is not much data available. These systems also alleviate over-
specialization since a wide variety of recommendations are offered, ensuring that users
will not be recommended items only because they are too similar to their previously
interacted items. Hybrid systems combine the best of collaborative and content-based
approaches for better recommendations. In addition, they will help one know more about
users' preference and the relationship among items and then have better results with
more personal recommendations. (Kramin, 2024)

The differences between the approaches are as follows:

2.1.5 Dataset
In machine learning and artificial intelligence, a dataset is any source of data that is used
in training and testing algorithms or models. These datasets constitute a very important
aspect of creating effective machine learning and AI systems since they provide both input
and output data the algorithms need to learn from. Examples of datasets include
structured and unstructured data. Structured data is basically the information that is
organized for computer analysis in a tabular form, which could be shown in a database
table or spreadsheet. It becomes simpler to analyze and use this type of data since it
already comes in a well-organized format. Conversely, unstructured data involves
information not preformatted to a specific structure, such as text or picture forms. This
kind of data takes more processing and analysis before it could be used in machine
learning and AI applications. (Datasets, 2024)
2.2 Review and analysis of existing work in the problem domain
2.2.1 Movie recommendation using collaborative filtering
2.2.2 Similar case study

Figure 4: Netflix

Summary
During the early days, Netflix heavily relied on collaborative filtering method for the
recommendation of its services to the users. Collaborative filtering, this method learns the
behavior of users who engage in similar or same contents in order to recommend new
items. Collaborative filtering has two types viz. user-based and item-based collaborative
filtering. User-based filtering approach works on the basis of the users having similar
patterns which include having viewing histories and so on. While on the other hand, item-
based filtering approach reviews the similarities between the content items themselves.
But as the user base and popularity of the Netflix grew, it became more important to have
a more accurate, precise and curated recommendation system. Therefore, now Netflix’s
current recommendation algorithm has become a combination of multiple models and
techniques like collaborative filtering, content-based filtering and deep learning. Hence,
Netflix now uses a hybrid recommendation system to harness the powers of multiple
models to provide the most accurate and curated items to the users. (Krysik, 2024)
Research no. 1

Figure 5: Research no. 1

Introduction
The present paper is based on collaborative filtering for recommending movies. In a world
where online content providers are faced with increasing data, there is an important
question of what exactly is relevant for each user. This is where recommendation systems
come into play. Recommendation systems will filter and suggest the right kind of content
to the users, depending on their preferences and past behavior. It is focused on
collaborative filtering-a method that predicts the preferences of active users based on the
history of similar users' actions, investigating how it can enhance the recommendation
system. For this study, the MovieLens dataset was used; hence, different collaborative
filtering techniques, such as user-based and item-based filtering, have been implemented
to provide recommendations for movies.
Findings
This research developed an approach to collaborative filtering, comparing the historical
ratings made by users and offering effective movie recommendations. The paper
discussed mainly two approaches:
i. User-based filtering – This is where recommendations are done based on the
similarity between users, such as recommending items liked by users of similar
preferences.
ii. Item-based filtering – This approach recommends items similar to those that a
user has rated highly in the past. In this study, the item-based collaborative
filtering may be more reliable since user preferences might change with time,
while items remain constant.

Conclusion
In conclusion, collaborative filtering is a powerful tool for creating recommendation
systems that can help users discover content they may enjoy based on similar users'
preferences. However, the system still faces challenges like data sparsity, cold starts, and
scalability. The study suggested that combining collaborative filtering with other
techniques, like hybrid models, can enhance performance and overcome these
challenges. Although collaborative filtering is effective, it can be improved further,
especially by refining algorithms to handle large amounts of data more efficiently. Future
research could focus on optimizing these systems and improving user interface
interactions for a better user experience. (Pradhan, 2021)
2.2.3 Research no. 2

Figure 6: Research no. 2

Introduction
Because technology is rapidly evolving, much advanced platforms such as Machine
Learning, Deep Learning, Data Mining, and the Internet of Things are introduced. These
technologies have been integral to our lives in easing several tasks that happen within an
industry or even your daily activities. Among many use cases, one of the widest-ranging
uses has to be recommendation systems. These systems facilitate decision-making by
giving filtered recommendations, enhancing the overall experience for any type of user
on any e-commerce, streaming service, or educational tool. From a leisure perspective,
there's also an overexposure in available choices-for example, movies that run into
millions found in libraries-where people end up not knowing which to choose. It's out of
this difficulty that movie recommendation systems began: the enabling of users to
discover those films that best match their preferences. Modern recommendation systems
harness machine learning algorithms, mostly collaborative and hybrid filtering techniques,
which improve user satisfaction and grow business. In this regard, the work described
below presents a movie recommendation system based on a machine learning technique
called K-Nearest Neighbors or KNN-algorithmically calculating similarity between users
of the system and suggesting personalized options.

Findings
This report shows that the best recommendation systems are supported by collaborative
filtering techniques. The challenges that are in traditional methods are addressed in this
system:
i. Sparse rating data: This system uses various ways to mandate ratings so that
it can have a more vast and richer dataset for precise and accurate predictions.
ii. Cold start problem: This system tackles this problem by requiring users to put
ratings in at least six movies. This helps the model to generate reliable and
accurate predictions.
iii. Over-specialization: The system widens and introduces more diverse
environment by incorporating neighborhood-based collaborative methods,
which expand user options by factoring in the preferences of similar users.
Conclusion

This system recommends movies to the users using a collaborative approach. Hence,it
will give progressively explicit outcomes contrasted with different systems that are based
on the content-based approach. Content-based recommendation systems are limited to
people, hence these systems don't prescribe things out of the box. These systems work
on individual users’ ratings, hence limiting your choice to explore more. While, this system
which is based on a collaborative approach harnesses the connection between multiple
similar users and recommends movies to others relying upon the ratings the previous
similar users have given to the movies they’ve watched. This subsequently allows users
to explore more.
(Singh F. &., 2020)

3 Solution
3.1 Explanation of the proposed solution/approach to solving the problem
3.2 Explanation of the AI algorithm/algorithms used
3.2.1 Collaborative filtering
This technique of collaborative filtering, it finds out users with similar behavior patterns
and recommends new items based on shared characteristics of their group. It relies on
the analysis of the manner in which other users with similar preferences and actions have
interacted with certain items. In other words, collaborative filtering algorithms group users
based on their behavior and make personalized recommendations using the overall traits
of the group. The idea behind every collaborative recommender system is based upon
the supposition that a user will be interested in those things in which others with similar
behaviors are interested. (Jacob Murel Ph.D., 2024)

3.2.2 Content-based filtering

Content-based filtering only depends on attributes and metadata of the movies to
generate recommendations of similar characteristics. The example of such a movie
recommendation system includes filtering data about genre, director, cast, or plot to offer
movies that represent similar genres, actors, or themes. The very critical power of
content-based filtering lies in its reliability for producing effective recommendations
despite having minimal or no data about a certain user. But if the movie metadata is
improperly labeled, misleading, or just too scant in detail, its effectiveness is
compromised. (Sharma, 2024)

3.2.3 K-means clustering

An unsupervised learning approach called K-Means Clustering divides the unlabeled
dataset into many clusters. In this case, K indicates how many pre-defined clusters must
be formed during the process; for example, if K=2, there will be two clusters; if K=3, there
will be three clusters; and so on.

The unlabeled dataset is split into ‘k’ distinct clusters using an iterative technique so that
each dataset belongs to a single group with comparable characteristics. It enables us to
group the data into distinct categories and provides a practical method for automatically
identifying the group categories in the unlabeled dataset without the requirement for
training. Each cluster is linked to a centroid in this centroid-based approach. This
algorithm's primary goal is to reduce the total distance between each data point and the
clusters that belong to it. (k-means-clustering-algorithm-in-machine-learning, 2024)

3.3 Pseudocode of the solution

3.4 Diagrammatical representations of the solution

4 Conclusion
4.1 Analysis of the work done
This is the proposal document of the AI project that we need to create for the 40%
coursework of our entire AI module. In this document, I have completed the tasks of
choosing the three algorithms for my movie recommendation system. Not only that, I have
also completed a thorough research on this system and the filtering methods that are best
suited for movie recommendation system. Along with this, we also had to create
pseudocode and flow charts for all the three algorithms that we will be training and testing
our data with. I have also completed a small research on how the recommendation system
of Netflix works, or rather what Netflix used to use and uses now for a similar case study.
Hence, this is the analysis of the work that has been done for now.

4.2 How the solution addresses real world problems

A movie recommendation system is a very important aspect of the entertainment industry.
The more curated items are provided by movie service providers, the more it would help
the businesses bloom. It also helps with customer satisfaction which leads to positive
feedbacks. These recommendation systems help to provide users with movies according
to their preferences without digging through large sets of movies. This keeps users
engaged which eventually makes them come back for more. This creates an environment
where both service providers and users are benefited, satisfied and happy. Hence, movie
recommendation systems can play a crucial role in improving quality of entertainment and
satisfaction which eventually leads to the improvement of overall quality of life for
everyone.

4.3 Further work

There are multiple works that could be incorporated into the movie recommendation
system in order to enhance the quality of services. For example, it would be a better
option if user input could be incorporated where they can provide reviews about the
recommendations they receive. By doing this, the system can learn even more and be
able to provide really precise movies to the users.
Further, sentiment analysis could also be incorporated in order to recommend more
preferential movies to the users. Movies that effectively resonate with the emotional
nature of users tend to provide more satisfactory user experience. This creates a feeling
of safer area where users can feel like they belong. (Movie recommendation based on
emotion in Python, 2024) Hence, these further work are in considerations for the movie
recommendation system.