Computer Sciences and Information Technology
Title : | Exploiting Side Information in Neural Collaborative Filtering Models for Improving Recommendation Quality |
Area of research : | Computer Sciences and Information Technology |
Focus area : | Machine Learning |
Principal Investigator : | Dr. Dilip Singh Sisodia, National Institute Of Technology Raipur, Chhattisgarh |
Timeline Start Year : | 2024 |
Timeline End Year : | 2027 |
Contact info : | dssisodia.cs@nitrr.ac.in |
Details
Executive Summary : | The deep learning-based neural collaborative filtering models are extensively used in real-world recommendation engines deployed by e-commerce giants. This proposal aims to integrate the additional user and item features (side information) with available product/item rating datasets in the neural collaborative filtering models to improve the recommendation quality. In the proposed work, the investigation will be performed to integrate side information into the personal recommender system and group recommendation systems with neural network-based factorization techniques. Integrate Side Information in Personal Recommendation Systems: Background: Matrix Factorization (MF) is one of the most explored techniques in collaborative filtering (CF) models. Several enhancements have been made to MF since its induction to recommender systems (RS). One of the significant improvements towards MF is its generalization to neural collaborative filtering (NCF) models. Xue et al. [1] proposed a deep matrix factorization model that combines the advantages of matrix factorization with deep learning approaches. The authors used explicit and implicit feedback on the user-item rating matrix. The user?item interaction matrix (binary) is obtained by simply checking whether a user consumed a particular item or not. Instead of using implicit feedback information, we can pass the side information of items to the system by applying binary representation. For example, in the case of movies, the genre information can be represented in a binary vector format for the items. Integrate Side Information in Group Recommendation Systems: Background: Cao et al. [2] used the neural collaborative filtering [3] to learn the group preferences by using the user and item latent feature vectors. The preference of an item for a group of users using the user and item latent feature vectors are modeled as a function. Instead of using existing weight matrices of the attention network, we can use the item features among the items as the weights for item embedding and the users' features as the user embedding. The latent feature matrices for users and items can be represented in binary indicator matrices and any vectorization procedure. These latent feature matrices can be given as input to the equation given by Cao et al. [2] in the motivation section for providing side information in the group recommendation process |
Total Budget (INR): | 6,60,000 |
Organizations involved