/DL-based-Collaborative-Filtering-Model-with-Content-based-Support

A book recommendation system that harnesses the power of Deeplearning based Collaborative Filtering complemented by content-based filtering to tackle the cold-start problem. Additionally, this model has the capability to recommend books based on external text queries, enhancing the versatility of the recommendations.

Primary LanguageJupyter NotebookMIT LicenseMIT

DL based Collaborative Filtering Model with Content-based Support

Overview

A book recommendation system that harnesses the power of Neural Collaborative Filtering (NCF) complemented by content-based filtering to tackle the cold-start problem. Additionally, this model has the capability to recommend books based on external text queries, enhancing the versatility of the recommendations.

Features

  1. Neural Collaborative Filtering (NCF): Predicts user-book ratings by learning from implicit interactions.
  2. Content-based Filtering: Provides recommendations based on content similarity, helping to handle scenarios where collaborative data might be sparse.
  3. Cold-start Solution: For new users or items, the system smartly leverages content-based recommendations ensuring a seamless experience.
  4. Text Query Recommendations: Users can input textual queries to get book recommendations that align with their immediate interests.

Implementation

  • The NCF model captures the latent factors from implicit user-book interactions.
  • Content-based filtering uses keyBert, TF-IDF and cosine similarity to recommend books similar to a given book.
  • Hybrid recommendation combines both the NCF and content-based increase reliability.

Citing

The Neural Collaborative Filtering (NCF) model is based on the following research paper:

He, X., Liao, L., Zhang, H., Nie, L., Hu, X., & Chua, T. S. (2017, April). Neural collaborative filtering. In Proceedings of the 26th international conference on world wide web (pp. 173-182).