Personalized Review Recommendation without User Interactive Data

Reviews from other raters play an important role in the user's decision making (e.g., purchasing a product). However, it is difficult for users to accurately find useful information from a large number of user reviews. The process of browsing these reviews is often very boring and time-consuming. Therefore, this paper aims to propose a personalized review recommendation method to help users solve this problem. Meanwhile, in most service communities, such as e-commerce services, the review reply is not available, which makes most of the current studies losing the efficacy. Therefore, realizing an accurate personalized review recommendation based on non-interactive user review data is easier to be promoted, but more challenging. In addition, most of the personalized recommendation related algorithms introduce complex large models, which require greater computational power and bandwidth. Our method proposed in this paper is an unsupervised lightweight neural network model, which can be trained only with ratings and non-interactive reviews. This method can be deployed hierarchically in the cloud and devices with less computational consuming. This method has been tested on five different mobile devices with four datasets from Amazon. The experiments illustrate that our approach can achieve a good personalized review recommendation result. Besides that, the resulting device memory usage and energy consumption are within the acceptable range.