Item Recommendation Using User Feedback Data and Item Profile

TL;DR

This paper explores content recommendation in a CMS using user feedback and content data, proposing a hybrid matrix factorization model that improves accuracy by incorporating content similarity and multiple feedback categories.

Contribution

It introduces a hybrid matrix factorization model that integrates content similarity and various user feedback categories for improved recommendation accuracy.

Findings

Different feedback categories impact recommendation accuracy differently.

Using all feedback categories yields the best performance.

The hybrid model outperforms traditional matrix factorization models.

Abstract

Matrix factorization (MS) is a collaborative filtering (CF) based approach, which is widely used for recommendation systems (RS). In this research work, we deal with the content recommendation problem for users in a content management system (CMS) based on users' feedback data. The CMS is applied for publishing and pushing curated content to the employees of a company or an organization. Here, we have used the user's feedback data and content data to solve the content recommendation problem. We prepare individual user profiles and then generate recommendation results based on different categories, including Direct Interaction, Social Share, and Reading Statistics, of user's feedback data. Subsequently, we analyze the effect of the different categories on the recommendation results. The results have shown that different categories of feedback data have different impacts on recommendation accuracy. The best performance achieves if we include all types of data for the recommendation task. We also incorporate content similarity as a regularization term into an MF model for designing a hybrid model. Experimental results have shown that the proposed hybrid model demonstrates better performance compared with the traditional MF-based models.