摘要
个性化推荐已成为现代智能化服务中的重要技术,时间信息是影响个性化推荐效果的重要因素。然而现有基于网络表示学习的个性化推荐方法,有的将事物及其关系抽象为同质网络,忽视了固有的异质性;有的虽将事物及其关系建模为异质网络,但没有较好地利用时间信息。因此,文中提出一种基于时间信息表示学习的个性化推荐方法(Time Representation Learning Incorporating into User Collaborative Filtering, TRLIUCF),TRLIUCF方法综合利用了评论的时间信息、文本信息、评分信息,获得了较好的推荐结果。首先,根据评论文本提取评论情绪特征,并根据时间信息提取评论时间特征,基于二者提出评论综合情绪-贡献值及其计算方法。然后,基于评论综合情绪-贡献值和用户评论数据构建用户-商品-评论异质网络,并采用网络表示学习方法学习节点嵌入向量。最后,通过用户节点嵌入向量计算用户相似性,并采用基于用户的协同过滤进行TOP-N推荐。在两个不同规模的真实数据集上的实验表明,与基准方法相比,TRLIUCF方法提高了推荐精确率和召回率。
Personal recommendations have become an important technology in modern intelligent services, and time information is an important factor affecting the effectiveness of personalized recommendations. However, some existing personalized recommendation methods based on network representation learning abstract things and their relationships as homogeneous networks, ignoring the inherent heterogeneity;some model things and their relationships as heterogeneous networks, but do not make use of time information. Accordingly, we propose a personalized recommendation method based on Time Representation Learning Incorporating into User Collaborative Filtering(TRLIUCF),which integrates temporal information, textual information, and rating information of reviews to obtain better results. Firstly, we extract the comment sentiment features based on the comment text, and extract the comment time features based on the time information, and propose the comment integrated sentiment-contribution value and its calculation method. Then, a user-item-review heterogeneous network is created based on the comment integrated sentiment-contribution value and user review data, and a network representation learning method is used to learn the node embedding vector. Finally, user similarity is calculated by the embedding vector of the user node and TOP-N recommendation is executed using user-based collaborative filtering. Experiments on two real datasets of different sizes show that the TRLIUCF method improves recommendation precision and recall compared to the benchmark method.
作者
任秋臻
陈红梅
周丽华
REN Qiu-zhen;CHEN Hong-mei;ZHOU Li-hua(School of Information Science and Engineering,Yunnan University,Kunming 650500,China)
出处
《计算机技术与发展》
2023年第1期34-41,共8页
Computer Technology and Development
基金
国家自然科学基金(62062066,61966036)
云南省中青年学术和技术带头人后备人才项目(202105AC160067)
云南省应用基础研究计划重点项目(202101AS070056)。
关键词
时间信息
网络表示学习
异质网络
协同过滤
个性化推荐
time information
network representation learning
heterogeneous networks
collaborative filtering
personalized recommendation
作者简介
任秋臻(1996-),女,硕士,研究方向为数据挖掘、推荐系统;通信作者:陈红梅(1976-),女,博士,副教授,CCF会员(49450M),研究方向为数据挖掘、空间数据挖掘。
