Citation Search And Recommendation Analysis On Research Of Personalized Based On Deep Learning

The appropriate reference literature plays a vital role in writing academic documents,which can indicate the extent of the author's knowledge coverage in the field of his own research,and also help the author to demonstrate his views.According to the statistics of DBLP(Digital Bibliography & Library Project),about 300,000 academic documents are published every year in the computer-related fields.It can be seen that the number of academic literature has grown so rapidly,so it is of great significance to help scientific researchers dig out scientific research literature suitable for themselves from a large number of academic documents.Search and recommendation technology are two effective techniques to solve the above problems.This article is based on search and sort algorithm of Lucene,deep learning as the background,mining user's personalized interest model,explore the deficiencies of citation recommended algorithms,finalized the related algorithm research of personalized citation search and recommendation based on deep learning.The research work of the dissertation is as follows:(1)A personalized search algorithm based on deep learning was proposed.In the process of building a personalized search algorithm,we should first construct the user's personalized interest model,and reconstruct the deep learning Seq2 Seq model based on the LSTM and attention mechanism,and uses it to mine the user's research interests and vectorize it.Secondly,based on the Lucene search engine sorting algorithm,we fuse the interest model and then reconstruct the sorting algorithm.The search results are sorted according to the reconstructed algorithm to get the final search results.Experiments show that a personalized search algorithm for user interest mining method based on Seq2 Seq model,personalized search over traditional expressions of interest increased by 11% at MAP,increased 28% at P@10.The experiment shows that the proposed algorithm is feasible and has certain advantages.(2)A citation recommendation algorithm for multi-feature fusion was proposed.Only consider single features for traditional citation recommendation algorithms lead to overspecialization of recommendation results and lower quality of recommendation,a citation recommendation algorithm based on multi-feature factor fusion is proposed.The algorithm effectively integrates the global influence factor,local activity factor,query correlation factor and author correlation factor through a multi-feature fusion model.Among them,the query correlation factor is improved by restarting the random walk calculation by incorporating the ID2 vec improved state transition matrix.The experimental results on the AAN dataset show that the citation recommendation algorithm based on multi-feature fusion has a corresponding increase in recall rate compared to GloPageRank,TopicSim and BM25 models.Compared with the original algorithm,the improved restart random walk algorithm has improved 8.13% in recall rate and 29.7% in NDCG.Analyzed by experimental results,multi-feature fusion citation recommendation algorithm can effectively improve the citation recommendation quality.(3)The implementation of a personalized search recommendation system.Combined with the relevant algorithms developed in this paper,based on the current popular web framework and commonly used search engine framework,personalized citation search recommendation system is implemented.