Research And Improvement Of Deep Relevance Matching Model Based On Information Retrieval

With the rapid development of network communication and e-commerce,the Internet has become an important way for people to obtain information in theinformation retrieval lives and work.For the huge resource base of the Internet,if there is no effective search tool,it is difficult for people to retrieve the information they need.In order to improve the speed of searching for effective information and reduce the difficulty for people to retrieve information,an information retrieval system is born.Text matching plays an important role in information retrieval systems.In the text matching process,there is a matching mistake problem.When the matching mistake means that the two paragraphs of text represent the same meaning by different words,the model cannot judge the matching meaning caused by the similar meaning.In response to this problem,most of the current research work is to expand the text by increasing the query words or the synonym of the document word,to increase the matching probability between the query word and the document,so as to alleviate the matching mistake problem,the method can be to a certain extent Solving the problem of matching errors,but it is computationally intensive and requinformation retrievales a huge resource pool;in deep learning,researchers use word embedding to calculate the similarity of synonyms,but the similarity between words and words still deviates,so it is not Can well alleviate the matching mistakes problem.For the problem of matching errors in text matching,this paper proposes the following two models:(1)A Deep Top-K Relevance Matching Model(DTMM)model,the contribution of which is to add document weights to the model to alleviate matching mistakes.problem.Since not all semaphores are good for text retrieval,the model will focus on learning K semaphores with similarity and weight of document words,making the information amount of the input model more reliable and effective,and then learning the query through the multi-layer connection layer.The score of the document;(2)a retrieval model based on the expansion of the knowledge map word,which finformation retrievalst extracts all the entities in the query and the document,and aligns the entities into the knowledge map.Since the synonymous entities conform to the linguistic "distance similarity" principle in the knowledge map,the word contexts with similar meanings are the same.This paper uses the SkipGram model to learn the word embedding of entity words and expands the text to enrich the semantic representation of the text.Finally,input the expanded text information into DTMM to further alleviate the matching mistake problem.In addition,the two models presented in this paper are validated on the MQ2007 dataset and the Robust04 dataset.The experimental results show that the DTMM model proposed in this paper and the retrieval model based on knowledge map word expansion can effectively alleviate the matching mistake problem.