基于长短期偏好注意力网络的兴趣点推荐  

Recommendation for Point-Of-Interest Based on Long-and Short-Term Preferences Attention Networks

作　　者：廉小亲[1] 米嘉晨 高超 关文洋 LIAN Xiao-Qin;MI Jia-Chen;CAO Chao;CUAN Wen-Yang(School of Artificial Intelligence,Beijing Technology and Business University,Beijing 100048,China;Key Laboratory of Industrial Internet and Big Data,China National Light Industry,Beijing Technology and Business University,Beijing 100048,China)

机构地区：[1]北京工商大学人工智能学院,北京100048 [2]北京工商大学中国轻工业工业互联网与大数据重点实验室,北京100048

出　　处：《计算机仿真》2024年第3期399-405,共7页Computer Simulation

基　　金：北京市自然科学基金资助项目(6214034)。

摘　　要：兴趣点(Point-Of-Interest,POI)推荐是基于位置的社交网络(Location-based Social Networks,LBSNs)研究中最重要的任务之一。为了解决POI推荐中的空间稀疏性问题,提出一种用于位置推荐的长短期偏好时空注意力网络(LSAN)。首先,构建了签到序列的时空关系矩阵,使用多头注意力机制从中提取非连续签到和非相邻位置中的时空相关性,缓解签到数据稀疏所带来的困难。其次,在模型中设置用户短期偏好和长期偏好提取模块,自适应的将二者结合在一起,考虑了用户偏好对用户决策影响。最后,在Foursquare数据集上进行测试,并与其它模型结果进行对比,证实了提出的LSAN模型结果最优。研究表明LSAN模型能够获得最佳的推荐效果,为POI推荐提供新思路。Point-Of-Interest(POI)recommendation is one of the most important tasks in Location-based Social Networks(LBSNs)research.In order to solve the spatial sparsity problem in POI recommendation,this study proposes a long and short-term preference attention network(LSAN)for location recommendation.Firstly,a relationship matrix of check-in sequences is constructed,which extracts the correlations in non-consecutive check-ins and non-adjacent positions using a multi-headed attention mechanism to alleviate the difficulties caused by sparse check-in data.Moreover,the short-term preference and long-term preference extraction modules are set in the model,which adaptively combines the two and considers the influence of user preferences on user decisions.Lastly,testing on the Foursquare dataset and comparing the results with other models confirmed the optimal results of the LSAN model proposed in this paper.The study indicates that the LSAN model can obtain the best recommendation effect and provide new ideas for POI recommendation.

关 键 词：兴趣点推荐 用户偏好 注意力网络 时空间隔 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]