Research Article
Research on Personalized Recommendation of Mobile Social Network Products Based on User Characteristics
@INPROCEEDINGS{10.1007/978-3-031-50549-2_14, author={Min Zhou and Wei Xu and Xinwei Li}, title={Research on Personalized Recommendation of Mobile Social Network Products Based on User Characteristics}, proceedings={Advanced Hybrid Information Processing. 7th EAI International Conference, ADHIP 2023, Harbin, China, September 22-24, 2023, Proceedings, Part III}, proceedings_a={ADHIP PART 3}, year={2024}, month={3}, keywords={User characteristics Mobile social network goods Personalized recommendation Attribute characteristics MetaEE Losses}, doi={10.1007/978-3-031-50549-2_14} }- Min Zhou
Wei Xu
Xinwei Li
Year: 2024
Research on Personalized Recommendation of Mobile Social Network Products Based on User Characteristics
ADHIP PART 3
Springer
DOI: 10.1007/978-3-031-50549-2_14
Abstract
In the process of conducting online product recommendations, the lack of comprehensive user profiling in constructing user personas has led to low Top-10 hit rate, average reciprocal rank, and normalized discounted cumulative gain of recommended products. To effectively address this issue, a user feature-based personalized recommendation method for mobile social networks (MSN) is proposed. By analyzing the basic attributes, interaction attributes, feedback attributes, and interest attributes of MSN users, user attribute features are extracted to build user personas. Based on these user personas, personalized recommendations for mobile social network products are achieved using MetaEE. This involves updating the recommended products based on the collection of user interactions with historical items until there is overlap between the support set and the query set of the personalized recommendation meta-learning samples. The corresponding products are then considered as the final recommended results. Experimental results demonstrate that the designed recommendation method outperforms the comparison methods in terms of Top-10 hit rate, average reciprocal rank, and normalized discounted cumulative gain across multiple experimental scenarios, indicating a promising recommendation performance.
