| 祝婷.融合网络表示学习与文本信息的学术文献推荐方法[J].情报工程,2022,8(3):081-092 |
| 融合网络表示学习与文本信息的学术文献推荐方法 |
| An Academic Paper Recommendation Method by Network Representation Learning and Text Information |
| |
| DOI：10.3772/j.issn.2095-915X.2022.03.006 |
| 中文关键词: 网络表示学习；Node2vec；标签；BERT；推荐；学术文献 |
| 英文关键词: Network representation learning; Node2vec; tag; BERT; recommendation; academic paper |
| 基金项目:陕西省科学技术情报学会项目“网络表示学习在图书馆文献资源推荐系统中的应用研究”（2022KTF-06）。 |
| 作者 | 单位 | | 祝婷 | 西安工业大学图书馆 西安 710021 |
|
| 摘要点击次数: 896 |
| 全文下载次数: 503 |
| 中文摘要: |
| [ 目的/ 意义] 为了从引文网络、文献内容、标签等多角度挖掘文献间的深层次关系，进而提高学术文献推荐的效果，提出一种融合网络表示学习与文本信息的学术文献推荐方法。[ 方法/ 过程] 首先，使用网络表示学习、BERT、标签针对文献库分别生成基于引文网络的特征向量表示、基于长文本内容的特征向量表示以及基于短文本标签的特征向量表示；其次，针对网络表示学习及BERT 生成的向量进行一次特征融合，采用余弦相似性算法分别计算特征融合及标签对应的文献相似度矩阵，并对其进行二次相似度矩阵融合，获取文献综合相似度矩阵；最后，按照相似度大小对待推荐的文献进行排序，实现Top-N 推荐。[ 结果/ 结论] 在CiteUlike 数据集上进行实验验证，相比于对比方法在准确率、召回率和F 值上平均提升了31.05%、28.51% 和29.70%，结果表明本文方法较于单一推荐方法可以有效提高学术文献推荐的质量。 |
| 英文摘要: |
| [Objective/Significance] In order to dig out the deep-level relationship between academic papers from multiple perspectives such as citation network, content, and tags, and improve the effect of academic paper recommendation, an academic paper recommendation method by network representation learning and text information is proposed. [Methods/Process] Firstly,use network representation learning, bert algorithm, and tags to generate feature vectors based on citation network, long text content, and short text tags for the paper dataset; Secondly, according to the vectors generated by the network representation learning and bert, the first feature fusion vector is calculated, and based on the first feature fusion vector and the vector calculated by paper labels, the secondary similarity fusion matrix is carried out by the cosine similarity algorithm; Finally, sort the similarities of papers and realize Top-N recommendation. [Results/Conclusions] The proposed method is validated in the CiteUlike dataset, the precision, recall and F-measure of this method increased by 31.05%, 28.51% and 29.70% on average compared with the comparison method, and the results show that this method can effectively improve the quality of academic paper recommendation compared with single recommendation method. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|
