«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

期数:

2025年03期

页码:

172-178

栏目:

新型计算应用系统

出版日期:

2025-03-10

文章信息/Info

Title:

Sequence Recommendation Algorithm Based on One-dimensional Convolutional Neural Network

文章编号:

1673-629X(2025)03-0172-07

作者:

黄康鹏; 冯锋

宁夏大学 信息工程学院,宁夏 银川 750021

Author(s):

HUANG Kang-peng; FENG Feng

School of Information Engineering,Ningxia University,Yinchuan 750021,China

关键词:

推荐算法; 序列推荐; 卷积神经网络; 前馈网络; 用户特征

Keywords:

recommendation algorithm; sequential recommendation; convolutional neural network; feedforward network; user feature

分类号:

TP391.3

DOI:

10.20165/j.cnki.ISSN1673-629X.2024.0335

摘要:

近年来,针对信息过载问题,推荐算法已成为关键解决方法之一,能够有效地提供用户个性化内容。 在序列推荐研究中,卷积神经网络因其能有效提取序列信息中的局部特征而受到广泛关注。 然而,卷积神经网络在捕捉时序信息方面存在局限性。 为了解决这一问题,提出了一种基于一维卷积的序列推荐算法。 该算法首先通过卷积操作提取序列的局部特征,然后通过池化操作提取序列的长期特征,并将两者进行加权相加获得用户特征信息,使得其能够综合提取局部特征和长期特征。 随后,将用户特征信息与线性变化的序列信息进行点乘,以引入时序信息。 此外,将用户特征信息通过前馈网络,以实现非线性转换和跨维度交互增强。最后,对用户特征向量与项目特征向量进行计算获得相关性,并以此进行推荐。 实验结果表明,在 MovieLens 电影数据集和 KuaiRand 短视频数据集上的测试中,该算法在推荐命中率和归一化折损累计增益等指标上相比四种基线算法均有显著提升,表明该算法能够更有效地进行推荐。

Abstract:

In recent years,recommendation algorithms have become a key solution to the problem of information overload,providing users with personalized content effectively. In the field of sequential recommendation,convolutional neural networks have gained widespread attention for their ability to extract local features from sequential information. However,convolutional neural networks have limitations in capturing temporal information. To address this issue,we propose a temporal recommendation algorithm based on one-dimensional con-volution. The proposed algorithm first extracts local features from the sequence through convolution operations,then extracts long-term features through pooling operations, and combines the two using weighted addition to effectively capture both local and long - term features. Next,the extracted information is multiplied pointwise with linearly transformed sequential information to introduce temporal in-formation. Additionally,a feedforward network is used to achieve nonlinear transformation and enhance cross-dimensional interactions.Finally,the algorithm calculates the correlation between user feature vectors and item feature vectors to make recommendations.Experimental results show that in tests on the MovieLens movie dataset and the KuaiRand short video dataset,the proposed algorithm sig-nificantly improves metrics such as hit rate and normalized discounted cumulative gain compared to four baseline algorithms. It is demon-strated that the proposed algorithm is more effective in making recommendations.

常用功能

更新日期/Last Update: 2025-03-10