文献类型：期刊文献

英文题名：The recommendation of satisfactory product for new users in social commerce website

作者：Li, Shugang[1];Wang, Ru[1];Lu, Hanyu[1];Yu, Zhaoxu[2]

机构：[1]Shanghai Univ, Sch Management, Shanghai 200444, Peoples R China;[2]East China Univ Sci & Technol, Dept Automat, Shanghai 200237, Peoples R China

年份：2022

卷号：81

期号：12

起止页码：16219

外文期刊名：MULTIMEDIA TOOLS AND APPLICATIONS

收录：;EI(收录号：20221011741417);WOS:【SSCI(收录号:WOS:000763256600012)，SCI-EXPANDED(收录号:WOS:000763256600012)】；

基金：This work was supported by the Chinese National Natural Science Foundation (No. 71871135).

语种：英文

外文关键词：New user; Satisfactory product recommendation; Homogeneity; Consensus prediction; Social commerce

摘要：Recommending the post-use satisfactory product for new users can allow more effectively escalating the sale of products and expanding the new customers. When recommending a post-use satisfactory product for new users, the existing methods often suffer cold-start problems, which have tremendous consequences for their recommendation accuracy. To fill this gap, this study develops the consensus interest prediction model (CIPM) to predict new user's post-use satisfaction for the recommended products, which overcomes the cold-start problems by improving the robustness of the prediction model and deep extraction of user information based on the homogeneity in social commerce website. Specifically, we extract the five types of homogeneity characteristics between the active users and new users from the aspects of product homogeneity and user social relationship homogeneity. Then the multidimensional homogeneity user interest prediction models are designed with direct linear fusion, direct nonlinear fusion, and indirect fusion of multiple homogeneity indices. Subsequently, in order to improve the robustness of the model, the suitable interest prediction model is nominated for a given new user satisfactory product recommendation (NUSPR) scenario. The upper and lower bounds of the prediction error of each interest prediction model are calculated to estimate the similarity between it and other models in predicting results. The weight voting algorithm selects the model with the highest consensus as the final prediction model. Accordingly, the NUSPR strategies are proposed according to the model predicted result. Relying on Yelp's social commerce website, we verify that our CIPM can accurately predict the new users' satisfaction level.