Scientific Research Document Retrieval And Recommendation System Based On Doc2Vec

With the progress of informatization scientific research work in domestic universities,massive scientific research documents have been accumulated.These documents contain rich scientific research information,which has not been effectively used at present.With the increasing size of scientific research documents,it becomes more and more difficult for scientific researchers to obtain the required scientific research information quickly and accurately.In response to the above problems,this thesis builds a scientific research document retrieval and recommendation system to help scientific researchers obtain scientific research document information conveniently and efficiently.The core work of constructing the system is to transform scientific research documents into a form understood by computers.To this end,this thesis extensively investigated the text representation technology based on machine learning at home and abroad,and combined with the text features of scientific research documents,selected the classic Doc2Vec as the basic model.In order to make the training document vectors represent the document features as much as possible,this thesis makes an in-depth study on the model,and proposes a document embedding model called Weighted Doc2Vec(WDV)that incorporates word weight information as an improvement to the original model.In order to verify the WDV's document embedding ability,this thesis compares the text classification effect of Doc2Vec and WDV on the public movie review data set IMDB.Experimental results show that the classification accuracy of WDV fused with reasonable word weight information is higher than Doc2Vec.This thesis analyzes the user's needs in depth,designs the system from the perspective of function and technology,and finally implements a research document retrieval and recommendation system based on WDV.The system's document retrieval module helps users retrieve the required scientific research documents and display hot search named entity words;the document recommendation module uses user personal information,search records and scientific research document information to recommend related scientific research documents.This thesis studies and implements scientific research document retrieval and recommendation based on WDV,which provides a new method for scientific research information acquisition and has high research and application value.