Alleviating the User Cold-Start Problem in Recommendation Systems Based on Textual Reviews Using Deep Learning

Abstract

The main objective of recommender systems is to assist users in overcoming the issue of information overload by providing them with a carefully selected list of items that they are likely to find useful or relevant. Recommender systems may face many limitations and challenges, such as the cold start problem, which occurs when there is insufficient or no information about a new user or item. This leads to a decline in the performance of the recommender system. In this paper, we propose a recommendation system based on textual reviews and the deep learning method (RS-TRDL) to alleviate the user cold-start problem. Our RS-TRDL model can extract the important aspects and underlying sentiment polarity classification from the review text using NLP techniques and a deep learning method. These are then fused into collaborative filtering techniques to improve the RS and alleviate the user cold-start problem. The proposed method consists of two components: (i) An aspect-based sentiment analysis module that aims to extract aspects from the review text with its polarity; (ii) A recommendation generation component that uses the aspects as additional information with the numeric ratings. It also employs an important feature in the dataset, namely, the helpfulness to finally infer the overall rating prediction. Extensive experiments were conducted by the proposed system on two Amazon datasets. The experimental results show that the proposed RS-TRDL model exceeded all literature-reviewed comparison methods in the cold-start problem alleviation task.