A Novel Customized FARS Recommendation Algorithm based on Search

2018 2nd International Conference on Artificial Intelligence Applications and Technologies

计算机科学

Yao, L.^1 ; Yin, Q.Q.^1 ; Zhong, Y.^2

College of Mathematics and Econometrics, Hunan University, Changsha

410082, China^1

College of Information Science and Engineering, Hunan University, Changsha

410082, China^2

关键词: Baseline models;    Bayesian structure;    Gibbs sampling methods;    Information overloads;    Model analysis;    Personalized recommendation;    Recommendation algorithms;    User interests;   

Others  :  https://iopscience.iop.org/article/10.1088/1757-899X/435/1/012035/pdf
DOI  :  10.1088/1757-899X/435/1/012035

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

Search engines and recommendation systems work in different ways, but they are designed to handle the same problem: information overload. This paper combines the advantages of the two and proposes a new search-based recommendation algorithm: focused analysis recommendation system (FARS). Inspired by LDA, the FARS model was designed as a three-layer (user-interest-item) Bayesian structure. Based on user-specified keywords, FARS can only perform modeling analysis on relevant users, and return the distribution of interest of these users, which are estimated by the Gibbs sampling method. We treat the top-ranking items in each interest vector as related, and then sort them according to the probability that an item is selected by this user group. Finally, a specified number of recommended items is returned based on the above sorting result. The experimental results shows that the proposed model can outperform on Precision@n in comparison with the other baseline models. To some extent, it provides some inspiration for combining search and personalized recommendations.