Anime Recommendation System

Anime Recommendation System is a content-based recommendation system that provides personalized anime recommendations based on user preferences. It utilizes a dataset scraped from Anime Planet and employs various techniques such as data cleaning, text preprocessing, and feature extraction to create a content representation for each anime.
Using the content representation of the selected anime, the recommendation system calculates the similarity between the selected anime and other anime in the dataset. It then generates a list of the top 5 recommended anime based on their similarity scores.
The recommendation system focuses on leveraging the characteristics and features of individual anime to make recommendations. It does not consider collaborative filtering or user preferences. Instead, it identifies similar anime based on the content attributes such as genres, themes, and descriptions.

Table of Contents

• Data Cleaning
• Text Preprocessing
• Feature Extraction
• Similarity Calculation
• Web App
• Usage
• Key Learnings
• Future Goals

Data Cleaning

The data was cleaned using the NumPy and Pandas libraries. Preprocessing steps, such as handling missing values and removing duplicates, were performed to ensure the data quality. Here is the link of Jupyter Notebook.

Text Preprocessing

To handle different word tenses and variations, the nltk library's PorterStemmer was used. This helped in standardizing the text data and improving the accuracy of the recommendation system.

Feature Extraction

The sklearn.feature_extraction.text module's CountVectorizer was employed to extract relevant features from the text data. It identifies the most frequently used words and creates a matrix representing the frequency of each word in a given row.

Similarity Calculation

Using the cosine_similarity function from sklearn.metrics.pairwise, the similarity between two anime rows was calculated. Since the recommendation system operates in a multidimensional space, cosine similarity is preferred over Euclidean distance. The resulting similarity scores range between 0 and 1.

Web App

The project includes a web app created with the help of the pickle library and Streamlit framework. The serialized data and similarity matrix were loaded using pickle to enable real-time recommendations based on user input.

working.mp4

Usage

To use this Anime Recommendation System, follow these steps:

1. Clone the repository: git clone https://github.com/atharv-patil/anime-recommendation-system.git
2. Navigate to the project directory: cd app
3. Install the required dependencies: pip install -r requirements.txt
4. Run the web app: streamlit run app.py
5. Access the web app by opening the provided URL in your web browser.
6. In the web app, select one anime from the provided options.
7. Click the "Show Recommendations" button to generate the top 5 recommended anime based on your selection.

Key Learnings

Through this project, I gained valuable insights into the following concepts:

• Bag of Words
• CountVectorizer for feature extraction
• Cosine distance and similarity
• Text preprocessing using PorterStemmer

Future Goals

Future goals for this project:

• Incorporate additional features and attributes from the dataset to enhance the recommendation model's accuracy and coverage.
• Enhance the web app's user interface and interactivity to provide a more engaging and immersive experience for users.
• Explore advanced recommendation algorithms, such as collaborative filtering or deep learning approaches, to improve the quality of recommendations.
• Deploy the web app to a cloud platform to make it accessible globally.