面向逻辑神经推理的无偏推荐算法研究  

Research on Unbiased Recommendation Algorithm for Logical Neural Inference

作　　者：王建芳[1] 刘玢 WANG Jianfang;LIU Fen(School of Computing,Henan Polytechnic University,Jiaozuo Henan 454150)

机构地区：[1]河南理工大学计算机学院,河南焦作454150

出　　处：《软件》2024年第6期130-132,共3页Software

摘　　要：在推荐过程中,逻辑神经推理展现了强有力的性能。然而,现有方法在推荐时会引入用户和项目的额外信息。本文提出了一种面向逻辑推理的无偏个性化推荐算法LNR-UR,该方法能够通过简单的逻辑运算构造推理过程以及通过判断逻辑空间中向量的相似性构造神经推理模型。在不增加额外信息的条件下,通过基于交叉成对排序的负反馈采样提取用户偏好以减小预测偏差。实验表明,与经典方法相比,所提算法具有更高的推荐精度。During the recommendation process,logical neural reasoning demonstrated strong performance.However,existing methods introduce additional information about users and projects when recommending.This article proposes an unbiased personalized recommendation algorithm LNR-UR for logical reasoning,this method can construct inference processes through simple logical operations and construct neural inference models by determining the similarity of vectors in the logical space.Without adding additional information,extracting user preferences to reduce prediction bias through negative feedback sampling based on cross paired sorting.Experiments have shown that,compared with classical methods,the proposed algorithm has higher recommendation accuracy.

关 键 词：推荐 逻辑推理 采样 相似度 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]