计算机工程与应用 ›› 2023, Vol. 59 ›› Issue (2): 110-119.DOI: 10.3778/j.issn.1002-8331.2106-0083

• 模式识别与人工智能 • 上一篇    下一篇

多视图融合TextRCNN的论文自动推荐算法

杨秀璋,武帅,杨琪,项美玉,李娜,周既松,赵小明   

  1. 1.贵州财经大学 信息学院,贵阳 550025
    2.荥经县政务服务和大数据中心,四川 雅安 625200
    3.贵州财经大学 大数据应用与经济学院(贵阳大数据金融学院),贵阳 550025
    4.中国船舶工业系统工程研究院,北京 100094
  • 出版日期:2023-01-15 发布日期:2023-01-15

Automatic Paper Recommendation Algorithm Based on Multi-View Fusion TextRCNN

YANG Xiuzhang, WU Shuai, YANG Qi, XIANG Meiyu, LI Na, ZHOU Jisong, ZHAO Xiaoming   

  1. 1.School of Information, Guizhou University of Finance and Economics, Guiyang 550025, China
    2.Yingjing County Government Services and Big Data Center, Ya’an, Sichuan 625200, China
    3.Guiyang School of Big Data and Finance, School of Big Data Application and Economics, Guizhou University of Finance and Economics, Guiyang 550025, China
    4.Systems Engineering Research Institute, China State Shipbuilding Corporation, Beijing 100094, China
  • Online:2023-01-15 Published:2023-01-15

摘要: 传统论文自动推荐算法仅从单视图角度实现分类,缺乏特征融合及多视图语义知识,上下文信息和长距离依赖利用不明显,较难挖掘到深层次文本特征,从而限制学术论文推荐的准确度。针对这些问题,提出了一种基于多视图融合TextRCNN的论文自动推荐模型,该模型融合论文标题、关键词和摘要三个视图特征,利用卷积神经网络(CNN)、双向长短时记忆网络(BiLSTM)和注意力机制构建模型,实现对不同学科方向论文的自动分类及推荐。实验结果表明,设计的论文推荐模型在精确率、召回率和[F1]值上均有所提升,比机器学习方法平均提高3.40%、3.57%和3.49%,也优于单视图和已有经典的深度学习方法。该方法有效利用多视图知识和上下文语义信息,提高论文推荐的准确率,进而节约科研工作者检索所需论文所花费时间和精力,进一步提高科研人员的效率,推荐符合其研究需求的学术论文,具有良好的学术价值和应用扩展。

关键词: 论文推荐, 多视图融合, TextRCNN, 注意力机制, 深度学习

Abstract: Traditional paper automatic recommendation algorithms only implement classification from the single-view perspectives, lacking feature fusion and multi-view semantic knowledge, contextual information, and long-distance dependence are not prominent. It is challenging to dig deep-level text features, thus limiting academic papers recommended accuracy. To this end, this paper proposes a paper automatic recommendation model based on multi-view fusion TextRCNN, which combines three-view features of paper including title, keywords, and abstract. This method uses the convolutional neural network(CNN), bidirectional long and short-term memory network(BiLSTM), and attention mechanism to build a model to realize the automatic classification and recommendation of papers in different disciplines. The experimental results show that the model has improved precision, recall, and F1-score, which is 3.40%, 3.57%, and 3.49% higher than the machine learning method on average, and it is also better than single-mode and existing deep learning methods. This method effectively utilizes multi-view knowledge and contextual semantic information to improve the accuracy of paper recommendations, thereby saving the time and energy spent by scientific researchers in retrieving required papers. Meanwhile, the multi-view fusion TextRCNN model can improve scientific researchers’ efficiency and recommend academic papers that meet their research needs, which have good theoretical value and application expansion.

Key words: paper recommendation, multi-view fusion, TextRCNN, attention mechanism, deep learning