Topic Discovery And Evolution Research Based On Temporal Keyword Feature Analysis

With the rapid development and widespread application of Internet technology,data resources are showing an explosive growth trend.However,massive amount of information is mixed with much junk and invalid information,which often affects people’s judgment and wastes a lot of time.Moreover,the huge amount of data makes it difficult to obtain useful information.Therefore,mining research topics in massive amounts of information and obtaining high-quality information is particularly important.By discovering and sorting out the evolutionary context and relationships of topics,it is possible to predict cutting-edge hotspots of topics.The means of visual analysis can enhance the scientific and vivid results of evolution.The work of this paper is summarized as follows:First,this paper proposes multi-feature keyword extraction algorithms that incorporate temporal features.As the smallest constituent unit of a document,the quality of words has a significant impact on subsequent research such as topic discovery and evolution.Keyword extraction technology is an important method to improve its quality.This paper makes innovations based on previous research.Aiming at addressing the problem of ignoring temporal features of words in current keyword extraction algorithms,this paper proposes corresponding multi-feature keyword extraction algorithms by combining time as an independent feature with temporal network model and combining time as an influence factor with temporal decay function,respectively,to explore the influence of word temporal relationship on the quality of keyword extraction.These two methods effectively combine the information of temporal dimension and exploit the temporal features among words,then can improve the quality of keywords.Secondly,this paper further applies the keyword extraction algorithm with temporal features to the topic model,so as to improve the quality of the topic and realize the analysis of topic evolution in comprehensive and multi-angle ways.The LDA-based topic evolution model can discover hot topics from text information and is a classic topic discovery model.However,considering that the results obtained by directly applying the traditional topic model are often difficult to understand and exist a large number of meaningless,non-standardized or topic-irrelevant words,this paper uses the multi-feature keyword extraction algorithm integrating temporal information to process text information,which can further improve the effectiveness and interpretability of the topic model.In order to achieve a comprehensive analysis of the topic evolution,this paper first divides the texts into corresponding time windows according to time information,and then runs the topic model in each time window.The topic evolution analysis is divided into three parts,namely,topic relevance and novelty analysis,topic popularity evolution analysis and topic content evolution analysis.The evolution analysis can both achieve the purpose of predicting the overall hot topic popularity through the application of time series models in the field of deep learning,and observe the trend of topic splitting and merging in each time period.The experimental results show that the research methods proposed in this paper can dig out the temporal relationship of words and effectively enhance the effect of keyword extraction model,which can be applied to the topic model to enhance the effect of topic discovery and fully realize the evolutionary analysis research of topics.