融合命题逻辑与神经网络的隐式深度协同推荐模型  

Implicit deep collaborative recommendation model based on propositional logic and neural network

作　　者：陈邦 吴茂念[1,2] 朱绍军 郑博 彭蔚 Chen Bang;Wu Maonian;Zhu Shaojun;Zheng Bo;Peng Wei(School of Information Engineering,Huzhou University,Huzhou Zhejiang 313000,China;Zhejiang Province Key Laboratory of Smart Management&Application of Modern Agricultural Resources,Huzhou Zhejiang 313000,China)

机构地区：[1]湖州师范学院信息工程学院,浙江湖州313000 [2]浙江省现代农业资源智慧管理与应用研究重点实验室,浙江湖州313000

出　　处：《计算机应用研究》2022年第9期2689-2694,共6页Application Research of Computers

基　　金：国家自然科学基金资助项目(61906066);浙江省自然科学基金面上项目(LY20F02006);湖州师范学院研究生科研创新基金资助项目(2022KYCX43)。

摘　　要：以增强推荐算法模型认知推理能力,克服传统推荐算法高度依赖数据质量致使性能受限的现状为目标,提出一种将命题逻辑与神经网络融合的隐式深度协同推荐模型。首先,构建一个隐式逻辑表征模块,辅助联结实际问题中复杂变量与逻辑变量的隔阂,并将推荐问题转换为一个逻辑表达式。随后,利用神经网络拟合逻辑符号对逻辑表达式进行求解并完成推荐。在具有不同特点的三个经典数据集MovieLens、Book-Crossing、Amazon-E上的实验表明该方法展现了更好的推荐性能。With the goal of enhancing the cognitive reasoning ability of the recommendation algorithm model and overcoming the performance limitation of the traditional recommendation algorithm which is highly dependent on the data quality,this paper presented a deep collaborative recommendation model which fusion of propositional logic and neural networks.Firstly,it built an implicit logic representation module to linking complex variables in real-world problems with logical variables,and transformed the recommendation problem into a logical expression.Subsequently,it used neural networks to fit logical symbols to solve logical expressions and complete recommendations.Experiments on three classical datasets with different characteristics,MovieLens,Book-Crossing,and Amazon-E,show that the propesed model exhibits better recommendation performance.

关 键 词：推荐系统 协同过滤 神经网络 认知推理 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]