BOOK RECOMMENDATION SYSTEM

	GROUP MEMBERS	GITHUB
1.	MAUREEN WAMBUGU	https://github.com/Mau-Wambugu
2.	STEPHEN WAWERU	https://github.com/stendewa
3.	LYDIA NJERI	https://github.com/lydiahsherry23
4.	LILIAN MULI	https://github.com/mwikali24
5.	PETER MAINA	https://github.com/Mr-PeterMaina

Business Understanding

• Recommendation systems are powerful tools that use machine learning algorithms to provide relevant suggestions to users based on behaviour patterns or user data.
• A Book Recommendation System is a recommendation system where we recommend similar books to readers based on their interests. Recommendation systems help drive engagement, increase sales, increases revenue and this in return brings in loyal clients as the customer experience is elevated promoting customer satisfaction. We have 2 main recommendation system models:
• 1. Collaborative filtering
• 2. Content-based Filtering

BUSINESS PROBLEM

• Over the past years, there has been rise in huge ecommerce and online services leading clients facing difficulty when searching for the right products.
• Clients looking to purchase books also face the same struggle when trying to match the right books with their taste and preferences.
• The Business Problem is to develop a recommendation system that recommends books that are tailored to our users preferences inorder to improve customer experience and engagement.

PROJECT OBJECTIVE

1. To build a book recommendation system that provides personalized suggestions to our users.
2. Improve sales by showcasing books a user is most likely to buy.
3. Offer relevant books to users inorder to improve customer retention.
4. Increase customer engagement

The analysis aims to achieve these objectives by considering the following analysis questions:

1. Which authors consistently receive higher ratings from users?

2. How does the year of publication affect the average ratings of books? Are newer books preferred over older ones?

3. How accurate are the recommendations generated by the collaborative filtering model especially in terms of precision?

4. How does class imbalance in ratings affect the performance of the recommendation model?

DATA SOURCE

• We used data obtained from Kaggle mined by Cai-Nicolas Ziegler
• It contains 3 CSV Files:
  1. Books.csv - contains information about books{ISBN;Title;Author;Year;Publisher}
  2. Ratings.csv - contains book ratings provided by users that range from 0 to 10. {{User-ID;ISBN;Rating}}
  3. Users.csv - contains information about the users {User-ID;Age}

STAKEHOLDERS

1. Customers

• As the end user, they expect accurate book suggestions based on personal interests.

2. Marketing team

• They would want to do targeted advertising on specific books and also promote personalized offers.

3. Data scientist

• Interested in ensuring the recommendation system models are accurate and scalable.

4. Book Authors

• They would be interested in knowing how their books are recommended inorder to learn and understand their readers taste and preferences.

5. Executive {CEO}

• They would want to understand how recommendation systems impact revenue and customer retention comparing it to the budget allocated to the project.

METHODOLOGY

• Our project will focus on the CRISP-DM:
  1. Business Understanding
  2. Data Understanding
  3. Data Preparation
  4. Modeling
  5. Evaluation
  6. Deployment

DATA SOURCE

• We used data obtained from Kaggle mined by Cai-Nicolas Ziegler
• It contains 3 CSV Files:
  1. Books.csv - contains information about books{ISBN;Title;Author;Year;Publisher}
  2. Ratings.csv - contains book ratings provided by users that range from 0 to 10. {{User-ID;ISBN;Rating}}
  3. Users.csv - contains information about the users {User-ID;Age}

STAKEHOLDERS

1. Customers

• As the end user, they expect accurate book suggestions based on personal interests.

2. Marketing team

• They would want to do targeted advertising on specific books and also promote personalized offers.

3. Data scientist

• Interested in ensuring the recommendation system models are accurate and scalable.

4. Book Authors

• They would be interested in knowing how their books are recommended inorder to learn and understand their readers taste and preferences.

5. Executive {CEO}

• They would want to understand how recommendation systems impact revenue and customer retention comparing it to the budget allocated to the project.

METHODOLOGY

• Our project will focus on the CRISP-DM:
  1. Business Understanding
  2. Data Understanding
  3. Data Preparation
  4. Modeling
  5. Evaluation
  6. Deployment

Explanatory Data Analysis

1. Univariate data analysis

2. Bivariate analysis

Modelling

In our analysis we used collaborative filtering to build a recommendation system.
In our evaluation we focused majorly on precision score,recall score,f1 score,accuracy score

Conclusion

1. Effectiveness of collaborative filtering

• Collaborative filtering using Singular Value Decomposition {SVD} can effectively predict book ratings for users based on their historical preferences and those of similar users.

2. Model performance evaluation

• The model has a resonable predictive accuracy score since RMSE= 3.4219 and MAE = 2.7470
• The model met our goal of having a precision score of atleast 75 % as it had a Precision of 79% measures the proportion of relevant recommendations that were actually correct i.e how many of the recommended books were truly relevant.
• However the model has a low recall score of 0.0814, this indicates most relevant books are not captured by the model.Thus it requires further attention to enhance the model quality.

3. User - Item based collaborative filtering

• User-User method effectively identifies similar users and their preferences.
• Item based approach using SVD provided personalized recommendations based on predicted rating for unseen books.
• Our analysis shows that the same user has different recommendations based on the two different approaches.

4. Item - based approach

• Based on this approach it is evident that highly rated books are preferred by most users.

Recommendations

1. Enhanced personalization and feedback mechanism.
2. Utilize data for targeted marketing.
3. Business should feature highly rated books.
4. Use top rated labels to highlight highly rated books in marketing campaigns.

FURTHER STUDIES

1. Explore Alternative Algorithms: Investigate the performance of other recommendation algorithms, such as neural collaborative filtering or hybrid approaches, to compare their effectiveness against the current SVD model.

2. Content-Based Recommendations: Explore the potential for a content-based recommendation system that utilizes book attributes, such as author, to complement collaborative filtering methods and provide more holistic recommendations.