Research On The Research Hotspots Recognition Methods Based On Citation Network Structure And Text Semantic

Research hotspots refer to the research problems or directions that are widely concerned in a research field.Research hotspots can reflect the focus of a field,and guide the development direction of the field.Identifying research hotspots is of great significance for scientific research planning,and seizing the high ground of scientific research.Currently,the main method for identifying research hotspots is based on scientific literature.Due to the large and exponentially growing number of documents,using computer technology for hotspot identification is a necessary means.The hotspots identification method based on the fusion of citation network structure and text semantics has been the main research field in recent years and has been highly valued by the academic community.In this context,the following aspects of work were carried out:First,an improvement was made to deep learning-based topic modeling technology.A grouping structure was proposed for topic modeling,which combined the latest advances in topic modeling in the field of deep learning.An auto-encoding variational inference model for grouped topic was proposed,whose neural network structure and loss function were designed through theoretical analysis.The experimental results showed that the model had improved clustering accuracy and semantic consistency compared to existing benchmark models.Second,a meta-clustering fusion strategy was introduced into constructing the fusion method for hotspot identification.Meta-clustering was used as a fusion strategy,breaking through the traditional fusion approach of similarity weighted fusion,and simplifying the construction process of fusion methods.Third,a semantic extraction method for hotspot identification results was proposed.Based on the AVITM method,a cross-entropy function was added to the loss function,transforming unsupervised AVITM into a supervised topic model.The results of the fusion method were used as category labels,and this model was used to extract the topic semantics of the hotspots to improve the readability of the hotspot identification results.Through experiments,it was demonstrated that this semantic extraction method could accurately extract the topic semantics of each hotspot from the hotspot identification results obtained by the fusion method,which was convenient for people to interpret and analyze research hotspots.Fourth,an empirical study was carried out to verify the effectiveness of the fusion method proposed in this study.Literature data in the field of library and information science from 2015 to 2017 was collected.Through comparison with single-category methods and hotspot identification results of other scholars,it was confirmed that the fusion method based on citation network structure and text semantics could improve coverage and timeliness,and could identify high-quality and appropriately granular hotspot topics.Further comparison with ESI research hotspots and the results of expert evaluations also evaluations also verified the effectiveness of the fusion method.The evaluation process considered both qualitative and quantitative evaluation,which played a certain reference role for similar research in the future.In summary,this study focused on hotspot identification methods,improved topic modeling technology for fusion methods,designed fusion strategies,designed semantic extraction methods for identification results,and verified the effectiveness of the fusion method through empirical research.Compared to existing methods,some improvements were made in hotspot identification effectiveness.The research process had theoretical significance for research on hotspot identification methods and fusion methods,and the research results had a certain application value.