融合多维度属性的重要关键词识别方法研究

Research on Important Keyword Recognition Method Integrating Multi-Dimensional Attributes

[目的/意义]已有研究大多是通过频次研究关键词的热点程度和分布,较少有研究综合考虑多个维度的属性,对关键词的重要性进行考量。从词汇或词组集合中识别出重要关键词,有助于研究者把握学科领域的重点内容,为科研选题、确定研究内容等提供决策支持。[方法/过程]首先,通过理论分析,引入RFM模型,提出关键词重要性概念模型和指标体系;其次,构建特征数据集,参考GloVe词向量模型的思想,通过共现矩阵提取关键词的特征向量;再次,使用关键词重要性概念模型提取分类标签,对数据进行自动化标注;最后,通过人工智能相关算法进行模型训练和验证,证明提出的识别方法的可行性。[结果/结论]模型训练和评估,SVC算法的F1值达到0.79,BiLSTM模型的F1值达到0.87,具有较好的拟合效果,说明提出的重要关键词识别方法具有可行性。[创新/局限]研究的创新点在于提出了具有多维度属性的关键词重要性概念模型和指标体系,并在深度学习模型上得到较好的评估结果;局限之处在于需要进一步扩大数据量,选择更多学科领域的数据对关键词概念模型进行验证,这是进一步研究的重点。

[Purpose/significance]Most of the existing studies have studied the hotness degree and distribution of keywords by frequency,but few studies have considered the importance of keywords by integrating the attributes of multiple dimen-sions.Identifying important keywords from a collection of words or phrases helps researchers grasp the key content of the subject area and provides decision supports for research topic selection and determination of research content.[Method/process]Firstly,through theoretical analysis,RFM model is introduced to propose keyword importance concept model and index system.Secondly,the feature vectors of keywords are extracted from co-occurrence matrix with reference to the idea of GloVe word vector model.Thirdly,classification labels are extracted using keyword importance concept model and automated annotation is performed on the data.Finally,the model is trained and validated by artificial intelligence related algorithms to prove the feasibility of the rec-ognition method proposed in this paper.[Result/conclusion]Through model training and evaluation,the F1 value of SVC algo-rithm and BiLSTM model reach 0.79 and 0.87,respectively,which have good fitting effect,indicating the feasibility of the impor-tant keywords identification method proposed in this paper.[Innovations/limitations]The innovation of this paper lies in proposing a keyword importance concept model and index system with multi-dimensional attributes and obtaining better evaluation results on the deep learning model;the limitation lies in the need to further expand the data volume and select data from more discipline areas to validate the keyword concept model,which is the focus of further research.