© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.

org (ISSN-2349-5162)

COURSE RECOMMENDATION SYSTEM USING

CONTENT BASED FILTERING AND COSINE
SIMILARITY
1
Vedula Lakshmi Spandana, 2Joga Rohit Abhinav, 3Pothu Naga Sampath,
4
Mohammed Musaiyab Ahamed
1 2 3,4
, , Student
1
Department of CSE, GITAM School of Technology, Visakhapatnam-530045, Andhra Pradesh, India

Abstract: Recommender systems compile a list of all currently accessible content, filter it in compliance with content moderation
standards, and then narrow the list down to the items that users are most likely to be interested in. The objective is to create a system
for recommending courses that takes user needs into account. A recommender system is helpful because it only shows consumers
the courses they want from a huge selection of online courses. In this project, streamlit is used to create a course recommendation
system that uses the course name and desired amount of suggestions as inputs to suggest all relevant courses on Udemy. Cosine
similarity and content filtering are used to accomplish this. Although there are other strategies, such as clustering models and matrix
factorization, Cosine similarity guarantees superior recommendation accuracy. Cosine Similarity calculates how similar two vectors
are to one another.It goes from -1 to +1, with +1 denoting similar items and -1 denoting differences between the compared items. A
csv file with 3672 rows is used to train the machine learning model. The streamlitcreated web page is then attached to this trained
model. The result would be a list of relevant courses including information like the course's title, url, rating, and other details that are
desired.
Keywords– ML (Machine Learning), Content Based Filtering and Cosine Similarity

1. INTRODUCTION

This project aims to build a course recommender system that recommends courses based on the user’s choice using content-based
filtering which takes the feature “course title” and uses cosine similarity to recommend the courses that are similar to the user’s input.
It is very much essential to remember that recommendations will be similar but not accurate. In this example we can see that one or two
of the requirements that are matching with the user requirement are being recommended to the user. Hence our model would be working
in the similar fashion.

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i210
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)

Figure 1: Example of working of a recommender system

2. LITERATURE SURVEY

In the paper titled "Methods for building course recommendation systems" KNN-COLLABORATIVE FILTERING is used. Many
techniques for creating course recommendation systems are presented in this book. A genuine data collection is used to analyze and
evaluate these methods. We show the foundation for creating the course recommendation system after choosing the best approach. This
kind of method might assist students in creating their study plan as well as their advisers. Future work will involve updating the models
to account for limitations like required coursework requirements or nonnegative matrix contents in order to improve prediction
accuracy[1]. In the paper titled "A course recommendation system based on grades" KNN-COLLABORATIVE FILTERING(non-
negative matrix) is used .Using experiments, we examined three different approaches and discovered that the suggested strategy is more
appropriate, efficient, and advantageous for learners. Implementing a recommendation system is frequently done to boost user pleasure
and the financial success of the platform. Additionally, to produce suggestions from a knowledge base for developing future
neighborhoods [2]. In the paper titled "Yield Prediction & Reaping Recommendation System Using Machine Learning" K-CLOSEST
NEIGHBORS DECISION TREES is used. The Crop Recommender system aids farmers in making decisions regarding which crop to
cultivate as well as yield predictions for certain crops. Also, it instructs the user when to apply the fertilizer. The fully automated system
envisioned by The Future Work will accomplish the same thing. The provision of the appropriate fertilizer for the specified crop and
area is another service that we are attempting to develop. It is necessary to put this into practice by researching fertilizers and how they
interact with soil and climate [3]. In the paper titled "Song Recommendation System Based on Real-Time Facial Expressions"
SVM(support vector machine) is used. In emotion recognition facial expressions are very important so facial detection is done using
SVM technique and for future scope of current project CNN can be used which is more efficient in facial detection [4]. In the paper
titled "Emotion Based Music Recommendation System" CNN-VGG is used. This project is designed for the purpose of making better
interaction between the music system and the user because music is helpful in mood changing of the user and developing the music
recommendation for the mood[5]. In the paper titled "Movie Recommendation System " KNN is used. The algorithm being developed
for this project will be used in a collaborative filtering recommendation system for movies. The user-based co-coin similarity algorithm
and the singular value decomposition algorithm are both used in this recommendation system [6].

3. DATASET

The dataset contains 3672 rows where all of the courses fall under any one of the domain among Web Development, Business Finance,
Musical Instruments and Graphic design [8].

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i211
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)

Figure 2: Dataset Description [8]

We have 1202 courses from Web Development, 1190 courses from Business Finance, 679 courses from Musical Instruments and 601
courses from Graphic Design. The dataset has 13 columns for each course namely course id, course title, url, price, Number of
subscribers, number of reviews, number of lectures, level, dup rating, Rating, Content duration, published timestamp and subject.

4. PROBLEM IDENTIFICATION AND OBJECTIVES

The goal of this project is to build a recommendation system that measures similarity between the available courses and the course that
is desired by the user. In case the user enters a course that is not available in the dataset then the system finds keywords from the input
and recommends the courses that contain those keywords.

4.1 EXISTING SYSTEM

4.1.1 COLLABORATIVE FILTERING

We find users who are similar to them through collaborative filtering [7], and we recommend activities they would like. Here, we cluster
users into groups of people who have similar attributes and then make suggestions for each one based on the preferences of its cluster,
as opposed to using the characteristics of the item to recommend.

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i212
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)

Figure 3: Example of collaborative filtering [7]

Predicting a user's rating based on their reviews of other movies and the overall ratings of other users is an example of collaborative
filtering. It is common practice to employ this idea when recommending books, articles, software, and a variety of other products.
4.1.2 DRAWBACK
One definition of collaborative filtering is "a strategy commonly employed across social networking, retail, and streaming services."
Cold start, or the lack of user history, is a shortcoming of collaborative filtering. User History plays a crucial role in collaborative
Filtering.

4.2 PROPOSED SYSTEM

4.2.1 CONTENT BASED FILTERING

Figure 4: Example of content based filtering [7]

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i213
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)

Content-based filtering [7] is one of the frequently used techniques in recommendation or recommender systems. The qualities or content
of the things you like are referred to as their “content.". In order to make recommendations for relevant products, content-based filtering
groups products according to specific keywords, determines the customer's preferences, looks up those terms in the database.

This form of recommender system heavily relies on user input; common examples include Google, Wikipedia, etc. For instance, Google
displays all the results that contain a user's specified selection of keywords. The answer to the question why are we proposing content
based filtering is it does not require data or history of other users.

The problem of cold start can be solved using content-based filtering because even though we do not have user history for a newly added
item we can still recommend using the attributes of the newly added item.

4. PROCESS FLOW OF THE MODEL

Figure 5 depicts the process flow of the proposed system. The available courses that are similar to the course desired by the user are
recommended to the user based on the similarity score.

Figure 5: Workflow of the model

5. OVER VIEW OF TECHNOLOGIES

5.1 PYTHON

Python [7] is a high-level programming language. It is employed because it is easy for the users to understand the written code and make
modifications if necessary. It supports structured programming, object-oriented programming etc and codes written in python can be
executed anywhere because python is platform independent.

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i214
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)
5.2 MACHINE LEARNING

Machine Learning [7] is subset of AI. It is defined as the capability of a machine or system to behave as a human. It is used in search
engines, recommendation systems, voice or text recognition etc. Since our project is a recommendation system the domain, we will be
dealing with will be Machine Learning.

5.3 CONTENT BASED FILTERING

This kind of recommendation system seeks to make recommendations for products (food, films, songs, anime, etc.) that are pertinent to
the user's chosen interests. It more closely resembles a user-content-based method where it determines how similar various products are
to one another. We must first transform our data into a matrix form, which consists of feature vectors, before we can begin to compute
similarities.
These similarities can be calculated using a variety of techniques:
Cosine similarity
K-Nearest Neighbor Distance (KNN)
Euclidean Distance
Using content-based filtering [7], a movie recommendation engine, for instance, may propose comparable films based on their genre,
director, and cast. Using content-based filtering, a news recommendation system might make suggestions for articles on related subjects
based on the keywords and tags assigned to each item.

5.4 COSINE SIMILARITY

Cosine similarity [7] is nothing but calculating the cosine angle between two vectors

Figure 6: Cosine similarity

For calculating similarity firstly, we will vectorize the text as follows.

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i215
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)

Figure 7: Vectorization of text

Basically, the vector representation is nothing but the count of each word encountered in each item. Then using the formula, we will
calculate the similarity score as follows.

Figure 8: Calculation of cosine similarity

5.5 MODULES USED

5.5.1 STREAMLIT
An open-source framework that helps in building websites that are required by ML engineers and it is so useful as it reduces thousands
of lines of code that should be written using HTML in order to build a website.[7]

5.5.2 PANDAS
Pandas [7] is used to manipulate data sets. It has tools that support analysis, and manipulation of data.

5.5.3 SKLEARN
It is a Python ML library called Scikit-learn (Sklearn) [7]. It has tools for modelling including classification, regression, clustering etc.

5.5.3.1 CountVectorizer
It is used to transform a group of text documents into a token count matrix.[7]

5.5.3.2 Cosinesimilarity
Cosine similarity [7] is designed to find similarity of the samples in the given input 2-D arrays.
JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i216
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)

6. CONCLUSION

Sub: Web Development

Figure 9: Output of the model

Hence the outcome of this project is required number of courses with its details.

7. FUTURE SCOPE

Recommendation systems are increasingly being integrated into the internet by more and more businesses in an effort to enhance
internal knowledge management and communications. Future Recommendation Systems trends are expected to be driven by the
following sectors:

• Systems for Making Recommendations in Businesses

• Recommendations in Mobile Social Networks

This project can be further improvised by adding user id to the dataset so that recommendations can be precise and an user profile can
be created to store users search history such that similarity score based on user’s interest can be increased and recommendat ions can
be made accurately.

8. REFERENCES:

[1] Huynh-Ly Thanh-Nhan, Huu-Hao Nguyen, Nguyen Thai-Nghe, Methods for building course recommendation systems, research
gate conference.
[2] Bhaskar Mondal , Om Patra ,Sanket Mishra, Priyadarshan Patra, A course recommendation system based on grades, research
gate conference
[3] E. Nirosha, Dr. J. Sreeram Babu , M. Mohammed Riyaz, Yield Prediction and Reaping Recommendation System Using Machine
Learning, International Journal for Research in Applied Science & Engineering Technology (IJRASET),Volume 10 (Issue
VIII),page no:932-935,[ISSN: 2321-9653]

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i217
© 2023 JETIR March 2023, Volume 10, Issue 3 www.jetir.org (ISSN-2349-5162)
[4] Himanshi Agrawal, Ramesh Patil , Prathmesh Abitkar , Nikhil Vartak , Shatayush Thakare, Song Recommendation System
Based on Real-Time Facial Expression, International Journal for Research in Applied Science & Engineering Technology
(IJRASET),Volume 10 (Issue XI),page no:1302-1306,[ISSN: 2321-9653]
[5] Aryan Shirwadkar , Pratham Shinde , Sahil Desai , Samuel Jacob, Emotion Based Music Recommendation System, International
Journal for Research in Applied Science & Engineering Technology (IJRASET),Volume 10 (Issue XII),page no:690-694,[ISSN:
2321-9653]
[6] Jose Immanuvel. J, Shilavati. A , Priyadarshan. M , Vignesh. S ,Elango. K, Movie Recommendation System, International
Journal for Research in Applied Science Engineering Technology (IJRASET),Volume 10( Issue VI),page no:2611-2615,[ISSN:
2321-9653]
[7] Algorithms knowledge (https://www.javatpoint.com ), (www.geeksforgeeks.com ) etc.
[8] Dataset knowledge (https://www.kaggle.com/datasets/thedevastator/udemy-courses-revenue-generation-and-course-
anal?select=3.1-data-sheet-udemy-courses-web-development.csv ).

JETIR2303833 Journal of Emerging Technologies and Innovative Research (JETIR) www.jetir.org i218