Interactive Academic Citation Recommendations For User Intent

Academic literature is an important carrier for scientists to display their scientific achievements.With the advent of information age,convenient access to information has further promoted the flourishing development of academic fields and the explosion of various academic resources.In the course of scientific research task,it is an important work to use academic resources efficiently.At present,the acquisition of academic literature mainly depends on the literature retrieval services provided by Google Learning,China Knowledge Network,Microsoft Learning and other academic resources.By entering the query text and setting the corresponding search conditions,the system uses the query text to match the titles,keywords and topics of academic literature in the resource library.This type of retrieval service only supports short text queries,which often do not accurately describe the user’s retrieval needs due to the limited information contained in the short text.Inaccuracy and impropriety of query words will also cause the search results to be skewed,and users still need to spend a lot of time reading the text of the literature for secondary screening after obtaining the results.In order to optimize the accuracy of document matching in retrieval system and improve the efficiency of scientific research,extensive attention has been paid to the study of academic citation recommendations based on the text contents and citation network of academic literature.From the point of view of precise matching of academic citations,The research is carried out from two aspects: hierarchical coding of academic papers and text matching feature mining.The main research contributions are as follows:(1)Mining the fine-grained features of the paper and constructing a hierarchical citation matching framework.As a summary of the research achievements of scientific scholars,academic papers not only put forward and verify academic opinions,but also introduce the background,theory and experiment of the research.In the course of academic retrieval,the rich semantic information contained in the thesis cannot be accurately described by using only title and keyword information.In this paper,the title and abstract text are used as input to mine the granularity semantic features of academic papers.Aiming at the problem that the text matching method is not suitable for long text,a hierarchical model framework for academic long text is designed.By using attention mechanism to mine different granularity text information representation,the attention weight of the long text inside text is weighted to obtain a more reasonable academic text encoding representation.By comparing experiments,the proposed model can improve the Acc and F1 indexes of the prediction task,and prove that the proposed model of academic citation recommendation based on hierarchical paper representation achieves better results in academic citation matching task.(2)Design cross-document interactive attention mechanisms to mine potential reference-matching features.Scholars choose academic citations not only because of the semantic similarity between two academic papers,but also because of the local content matching.These fine-grained information matches have important influence on the recommendation effect of academic citations.Based on the citation matching model framework of hierarchical paper representation,this paper designs the interactive attention mechanism between papers,and mines the finely grained reference matching features between papers through cross-document attention.Through the ablation experiment,it is proved that the interactive attention mechanism between papers can improve the performance of citation matching task.Based on the above,this paper constructs a citation matching model based on hierarchical attention mechanism with the help of deep neural network and attention mechanism,introduces the mining of citation matching features between academic citations and verifies the validity of the model.