燕山大学学报2024，Vol.48Issue(4)：349-355,376,8.DOI:10.3969/j.issn.1007-791X.2024.04.007
基于知识图谱的多任务推荐算法
Multi-task recommendation algorithm based on knowledge graph
摘要
Abstract
The knowledge graph-based recommendation algorithm can enrich the characteristics of items,explore user interests,and effectively solve the cold start and data sparsity issues of traditional recommendation algorithms.However,existing knowledge graph-based recommendation algorithms often overlook the positive effect of collaborative information in user interactions on graph training,making it difficult to explore the deep features of items when the graph has a high degree of missingness.Therefore,in this article,a multi-task recommendation method called MRGC based on the knowledge graph is proposed,which jointly trains the recommendation task and the graph completion task.Firstly,the algorithm constructs a user-item connectivity graph and an item-related knowledge graph.It utilizes graph convolutional neural networks to expand the interaction representation of users and items and the structural representation of entity relationships,propagating collaborative information and graph information.At the same time,a two-layer attention structure is used to model the importance differences of same-order neighborhoods and the information decay of different-order neighborhoods,adaptively aggregating information.Finally,high-order representations of items and entities are cross-shared to learn knowledge from the other task.This algorithm fully characterizes item and entity representations,improving recommendation efficiency based on improving graph completeness.Comparative experiments are conducted with benchmark algorithms on three public datasets and one self-built dataset,the results show that the MRGC algorithm significantly improves metrics such as AUC and F1.关键词
推荐算法/知识图谱/图卷积神经网络/多任务学习/图谱补全Key words
recommendation algorithm/knowledge graph/graph convolutional neural network/multi-task learning/knowledge graph completion分类
信息技术与安全科学引用本文复制引用
柳啸峰,林广艳,于九阳,谭火彬..基于知识图谱的多任务推荐算法[J].燕山大学学报,2024,48(4):349-355,376,8.基金项目
国家自然科学基金资助项目(62276015) （62276015）
