|国家科技期刊平台
首页|期刊导航|电子科技大学学报|基于兴趣注意力网络的会话推荐算法

基于兴趣注意力网络的会话推荐算法OACSTPCD

Session-Based Recommender Algorithm Based on Interest Attention Network

中文摘要英文摘要

针对现有基于图神经网络的会话推荐算法对用户主要兴趣偏好提取不充分的问题,提出了一种基于兴趣注意力网络的会话推荐算法(Session-Based Recommender Method Based on Interest Attention Network,SR-IAN).首先,使用图神经网络捕获物品之间的上下文转换关系,得到物品的图嵌入向量;其次,将图嵌入向量输入兴趣注意力网络中,提取用户的主要兴趣偏好;然后通过注意力层对物品的图嵌入向量进行加权区分;最后,通过预测层得到候选物品的点击概率值并对其进行排序.算法模型在 3个公开数据集Diginetica、Retailrocket和Tmall上进行了实验验证,相比基准模型在MRR@20指标上分别有0.942%、1.183%和2.977%的提升,同时降低了模型时间复杂度,验证了该方法的有效性和高效性.

Aiming at the problems of insufficient extraction of users'main interest preferences in session-based recommender algorithms based on graph neural networks,a Session-Based Recommender Method Based on Interest Attention Network(SR-IAN)is proposed.First,the graph neural network is used to obtain the context transformation relationships between the items,and the graph embedding vectors of the items are obtained;Secondly,the graph embedding vector input into the interest attention network to extract the user's main interest preferences;Then the graph embedding vectors of the items are weighted by the attention layer;Finally,the click probability values of the candidate items are obtained through the prediction layer and sorted.The proposed algorithm model was verified by experiments on three public datasets Diginetica,Retailrocket and Tmall,which showed an improvement of 0.942%,1.183% and 2.977% compared with the baseline model on MRR@20.Besides,the time complexity of the model is reduced,which verifies the effectiveness and high efficiency of the proposed method.

崔少国;独潇;张宜浩

重庆师范大学计算机与信息科学学院,重庆 401331重庆理工大学两江人工智能学院,重庆 400054

计算机与自动化

注意力机制图神经网络推荐算法自注意力网络会话推荐

attention mechanismgraph neural networkrecommender algorithmself-attention networksession-based recommendation

《电子科技大学学报》 2024 (001)

67-75 / 9

重庆市自然科学基金面上项目(CSTB2022NSCQ-MSX1206);重庆市教委重点项目(KJZD-K202200510);重庆市科技局技术预见与制度创新项目(CSTB2022TFII-OFX0042)

10.12178/1001-0548.2022307

评论