Online-rating prediction based on an improved opinion spreading approach
Jun Ai, Linzhi Li, Zhan Su, Chunxue Wu · 2017
Recommender systems are significantly useful to reduce information explosion nowadays. In order to design an optimized algorithm, we develop an improved opinion spreading approach to predict online rating of recommender systems. The proposed method provides a solution to zero-value problems of similarity results, which is ignored in existing publications. The proposed method could produce a more precise rating prediction for each unrated user-item pair. In our work, the similarity of items is defined as the number of corresponding reviews which a user has given and the differences between those viewpoints spreading in network model. Using Movie-lens data set, experiments confirm that the presented algorithm has better performance than both collaborative filtering algorithm based on Pearson correlation coefficient and the original opinion spreading approach. Our approach produces a mean absolute error (MAE) 2.1% lower and root mean square error (RMSE) 2.5% lower than existing algorithms, which indicates a higher accuracy score.