Cross-platform rating prediction method based on review topic

摘要

Rating prediction is one of the research hotspots in intelligent recommendation. As the number of e-commerce platforms continues to increase, cross-platform user rating prediction has become an important prerequisite for cross-platform recommendations. For the same product, this paper constructs a Dynamic Cross-platform Information Network Model (DCINM) by using reviews. It integrates the information of the same product from multiple platforms. According to the relationship between network nodes, the DCINM can mine the hidden information of user reviews in different platforms. The user preference vector, constructed by combining the original and hidden information of user reviews, can more accurately reflect the relationship between user preferences and ratings. And it can also reduce the predicting ratings error. In the process of rating prediction, the paper proposes a method based on a binary classification algorithm to optimize the GBDT-LR (GABC-LR), which transforms the prediction process of GBDT-LR into the solution of a quadratic equation, such that the new predicted value from the solution is closer to the real values of the test data. The paper optimizes the rating prediction task from the perspectives of data processing and the prediction model, which reduces the error of the rating prediction while achieving cross-platform rating predictions. (C) 2019 Elsevier B.V. All rights reserved.