Research And Application Of Keyphrase Generation Technology Based On Pointer-Generator Networks

In recent years,the information management of the scientific research platform has been carried out rapidly which has led to a rapid increase in the amount of scientific research project documents.How to exploit and utilize the rich information contained in scientific research documents and give full play to the value of data is an urgent problem for the scientific research platform.Keyphrase information can highly summarize the main idea of the article and organize the content of the article.At the same time,the keyphrase information can be used in multiple fields of text mining,such as information retrieval,hotspot analysis,and abstracts generation.Therefore,the mining and utilization of document keyphrase information is one of the effective ways to leverage the value of documents.However,the accuracy of the current keyphrase extraction technology for extracting keyphrases is not high enough,which is the main factor that restricts the effective mining and utilization of keyphrase information.To improve the accuracy of keyphrase extraction is of great significance to data mining of the scientific research documents in the scientific research platform.Based on the research of word co-occurrence information and pointer-generator networks,this paper proposes a new keyphrase generation model called Co-occurrence Pointer Model(CoPM).The CoPM model uses the sequence-to-sequence neural network structure to learn not only the semantic information of words in the article but also the linguistic features of the word co-occurrence information in the article,so that it can analyze the article information in more dimensions and generate keyphrases of the article more accurately.This paper evaluates the model on the dataset of scientific research documents.The result of experiment shows that compared with the classic keyphrase extraction models,the CoPM model can produce more accurate results in terms of present keyphrase prediction and absent keyphrase prediction.In addition,based on the results of keyphrase generation for scientific research documents by the CoPM model,this paper implements keyphrase generation service for project and keyphrase-based project search service in the scientific research platform.Also this paper proposes a keyphrase-based hotspot analysis algorithm and applies it to the keyphrase-based hotspot analysis service.These applications improve the service of project analysis and decision system in the scientific research platform,and leverage the value of the research project documents effectively.