A Collaborative Neural Model For Rating Prediction By Leveraging User Reviews And Product Images
INFORMATION RETRIEVAL TECHNOLOGY, AIRS 2017(2017)
摘要
Product images and user reviews are two types of important side information to improve recommender systems. Product images capture users' appearance preference, while user reviews reflect customers' opinions on product properties that might not be directly visible. They can complement each other to jointly improve the recommendation accuracy. In this paper, we present a novel collaborative neural model for rating prediction by jointly utilizing user reviews and product images. First, product images are leveraged to enhance the item representation. Furthermore, in order to utilize user reviews, we couple the processes of rating prediction and review generation via a deep neural network. Similar to the multi-task learning, the extracted hidden features from the neural network are shared to predict the rating using the softmax function and generate the review content using LSTM-based model respectively. To our knowledge, it is the first time that both product images and user reviews are jointly utilized in a unified neural network model for rating prediction, which can combine the benefits from both kinds of information. Extensive experiments on four real-world datasets demonstrate the superiority of our proposed model over several competitive baselines.
更多查看译文
关键词
User Reviews, Prediction Rate, Token Representation, Appearance Orientation, Side Information
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络