工业工程

工业工程 ›› 2021, Vol. 24 ›› Issue (5): 132-140,151.doi: 10.3969/j.issn.1007-7375.2021.05.017

• 实践与应用 • 上一篇    下一篇

考虑主题多样性的工程领域知识推荐方法

王临科1, 蒋祖华1, 李心雨2   

  1. 1. 上海交通大学 机械与动力工程学院，上海 200240;
    2. 新加坡南洋理工大学 机械与宇航工程学院，新加坡 639798
  • 收稿日期:2020-03-25 发布日期:2021-11-02
  • 通讯作者: 蒋祖华(1966—)，男，浙江省人，教授，博士，主要研究方向为知识管理、设备预防性维护、人因工程。E-mail：zuhuajiang@sjtu.edu.cn E-mail:zuhuajiang@sjtu.edu.cn
  • 作者简介:王临科(1995—)，男，河南省人，硕士研究生，主要研究方向为推荐算法、知识管理
  • 基金资助:
    国家自然科学基金面上资助项目(71671113)

An Engineering Knowledge Recommendation Method Considering the Topic Diversity

WANG Linke1, JIANG Zuhua1, LI Xinyu2   

  1. 1. School of Mechanical Engineering, Shanghai JiaoTong University, Shanghai 200240, China;
    2. School of Mechanical and Aerospace Engineering, Nanyang Technological University, 639798, Singapore
  • Received:2020-03-25 Published:2021-11-02

RichHTML

75

PDF

5464

摘要： 工程问题的解决涉及到多个学科和专业的知识，主题单一的推荐列表已无法满足当前工程领域用户对知识多样性的需求。对此提出一种基于知识主题网络的多样性工程领域知识推荐方法(topic diversity collaborative filtering, TDCF)。该方法构建有向图形式的知识主题网络，定义并计算用户对知识主题的专业度；提出考虑用户专业度的评分矩阵预填充方法以缓解评分矩阵稀疏问题；利用知识主题网络和用户专业度改进传统协同过滤推荐算法中的用户相似度计算方法，从而有针对性地提升推荐结果多样性，进而提高推荐准确度和用户满意度。结合国内某船厂实际数据，设计对比实验。结果表明，TDCF推荐算法的F1分数和多样性指标最高分别达到0.52和0.44，均优于对比算法。因此，TDCF算法具有可行性与优越性，能够为用户提供更好的知识推荐服务。

关键词: 工程领域知识, 主题多样性, 知识主题网络, LDA主题模型, 知识推荐方法

Abstract: Engineering problem-solving involves multidisciplinary knowledge, and the users' knowledge needs cannot be satisfied by offering single filed knowledge. Therefore, an engineering knowledge recommendation method, topic diversity collaborative filtering (TDCF), was proposed to optimize the topic diversity of the recommendation list. In TDCF, a knowledge topic network was first constructed in the form of a directed graph, and the user's profession on the knowledge topic was calculated. Then a rating matrix pre-filling algorithm considering the user's profession was developed to alleviate the sparseness of the rating matrix. Finally, the knowledge topic network and user's profession were used to improve the user similarity calculation method in the traditional collaborative filtering recommendation algorithm, and thereby the diversity of recommendation results can be improved in a targeted manner and the recommendation accuracy and user satisfaction can be promoted. Experiment based on the real data collected from a domestic shipyard showed that TDCF's recommended results achieve higher F1-Score and diversity, which are 0.52 and 0.44 respectively, thus outperforming other benchmarking algorithms. Therefore, TDCF is feasible and superior, and can provide better knowledge recommendation services.

Key words: engineering knowledge, topic diversity, knowledge topic network, latent dirichlet allocation topic model, knowledge recommendation method

中图分类号: