Local citation recommendation system and method based on deep correlation matching

Abstract

The invention discloses a local citation recommendation system and method based on deep correlation matching. The method carries out the embedded representation of a word through a pre-trained multilayer language model, obtains the more contextualized representation of the word, and solves a problem that the word embedded representation is not abundant enough in a conventional method. According tothe invention, the problem of semantic fuzziness in a deep semantic matching method is solved. The interactive matrix learning is established for the citation context and the candidate paper content,so that the problem of great influence of a traditional model recommendation effect caused by great text length difference is solved. According to the method, the author network is innovatively constructed, the problem that the use characteristic is single in a traditional local citation recommendation method is solved, author information with the highest influence and correlation is fused into the model, and author characteristics and correlation characteristics are fully combined. According to the method, the same MLP network is used for learning each correlation feature, so that neural network parameters are effectively reduced, and the possibility of model over-fitting is reduced.

Description

【Technical field】 [0001] The invention belongs to the technical field of electronic information, and relates to a local citation recommendation system and method based on deep correlation matching. 【Background technique】 [0002] Citing relevant research results is an important link when researchers write academic literature. Researchers need to borrow their research ideas from the literature or describe it as the latest research progress. Researchers often need to cite a large number of references to support their views when writing academic literature, and the number of citations required varies greatly between different disciplines, especially for some relatively mature disciplines, and sometimes it is even necessary to deeply dig out all relevant literature. References, which will inevitably consume a lot of energy for researchers. How to quickly find suitable relevant literature among the academic resources of varying quality for researchers to make optimal choices is ...

AI Technical Summary

Problems solved by technology

This type of method generally adopts an iterative algorithm for model training, which often requires a long training time, so it is not suitable for citation recommendation in dynamically updated data sets; based on the translation model, the citation context and citation documents are regarded as two different "languages". ", and then use maximum likelihood estimation to calculate the probability of translation between them. This method has the problem of inconsistency between the words used in the citation and the target document, resulting in poor model performance.

The method based on deep semantic matching uses a deep neural network to automatically capture the similarity of words, phrases, and sentences, and infers the semantic relationship between the citation context and the target document at both ends of the text, so as to perform global matching. This method has become the current local citation method. Although the mainstream method of recommendation has made a lot of achievements, at present, due to the large difference in text length, there are problems such as semantic ambiguity and underutilization of literature information, which greatly affects the performance of partial citation recommendation.

Images