基于用户会话的TF-Ranking推荐方法  

TF-Ranking Recommendation Method Based on User Session

作　　者：贾丹 孙静宇[2] JIA Dan;SUN Jingyu(College of Information and Computer,Taiyuan University of Technology,Taiyuan 030024,Shanxi,China;College of Software,Taiyuan University of Technology,Taiyuan 030024,Shanxi,China)

机构地区：[1]太原理工大学信息与计算机学院,山西太原030024 [2]太原理工大学软件学院,山西太原030024

出　　处：《应用科学学报》2021年第3期495-507,共13页Journal of Applied Sciences

基　　金：山西省科技厅重点研发计划项目基金(No.201803D31226);山西省研究生教育创新项目基金(No.2019SY117)资助。

摘　　要：基于用户会话日志,提出了一种融合XGBoost和门控循环单元网络的TF-Ranking推荐方法。该方法利用门控循环单元学习用户行为。首先,利用XGBoost进行特征提取,克服了传统方法中数据模型复杂的缺陷,使数据模型在保持原始属性的基础上大大降低了复杂度;其次,利用改进后的Dropout网络对数据进行处理,使得召回率提高了1.32%;最后,基于Learning to Rank与Pairwise方法训练用户会话数据,尽可能为用户提供一个与查询内容相关性较强的正向排序推荐清单。实验在Trivago RecSys Challenge 2019数据集上进行。结果表明,所提出的推荐算法在召回率和平均倒数排名上均有提高,而且可以应用于大规模数据推荐。Based on users’ session logs, a TF-Ranking recommendation method that integrated XGBoost and gated recurrent unit was proposed. This method used gated recurrent unit to learn user session data. Firstly, XGBoost was used to extract features, and experimental results showed that XGBoost could overcome the defects of traditional data model and greatly reduced the complexity of data models while maintaining their original attributes. Secondly, an improved Dropout network was used to process data, leading to a recall rate improved by 1.32%. Finally, by training data based on Learning to Rank method in combination with pairwise method, a positive sort recommendation list with strong relevance to query contents was provided for users. Experiments were conducted on the data set of Trivago Recsys Challenge 2019. The results show that the proposed algorithm could improve the recall rate and the average reciprocal ranking, and could be applied to large-scale data recommendation.

关 键 词：推荐系统 TF-Ranking推荐 门控循环单元 提升树模型 Dropout网络 停留时间 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]