| In the new era,the improvement o hard and soft power of scientific research has led to an unprecedented increase in the number of academic publications.These rich academic resources provide a strong basis for future research.However,the large volume of academic papers makes it difficult for researchers to efficiently identify appropriate research directions and track the rapid progress of research.In this context,it is highly meaningful and necessary to predict the development trend of research topics.In view of the above problems,the main contents are as follows:(1)In order to solve the problem of not accounting for the topological relationship between research topics when predicting the development of research topics based on paper title words,we propose the dual impact perception prediction model DIPP.This is a study that takes the title words of papers as the granularity of the research topic and relies on the characteristic of the title of the papers.The method first generates topical word graphs for each time steps and models the development of research topics on GRUs.Then,in terms of information interaction,we construct a module for selecting and interacting interested information of peer-conferences with attention mechanism;in terms of information fusion,we perform node propagation based on topical word graphs with GCN.Finally,we integrate the dual effects of topical word graphs and dependencies between peer conferences to predict the development trend of research topics.(2)In order to take full advantage of research sessions and tracks given by authoritative venues,we propose a unified framework RTTP.This is a study that takes fact-based annotations as the granularity of the research topic and relies on the characteristic of the citation count of the papers.First,a novel topic alignment strategy considering both global and local information is designed to overcome the semantic differences between cross-year and crossvenue.Next,an enhanced prediction network,which consists of an embedding module,an encoder,a decoder and a prediction module,is used to capture the latent research trend of known years for prediction.(3)In order to be able to integrate multiple attribute feature information of papers,we propose a framework MRTP based on rich information scenarios.This is a study that takes fos tags extracted from MAG as the granularity of the research topic and relies on the characteristic of title,abstract,keyword and citation count of the papers.The method first learns the representation of research topics from bottom to top.Then,we design a multiinformation depth prediction module that includes a metadata semantic features extractor and a research topic popularity predictor.Finally,we combine the eigenvectors of four characteristic attribute feature information and predict the development trend of research topics by a MLP.To sum up,this thesis discusses the research on the forecast of the development trend of research topics from three perspectives,and verifies with real-world datasets.The experimental results show that the problem has been well solved,which can guide the future direction of research groups,universities and technology companies. |
