Semantic Search Of Crowdsourcing Software Services Based On Knowledge Graph

In recent years,with the development of the Internet and the rise of the software crowdsourcing industry,more and more individuals and organizations crowdsource software services in crowdsourcing platforms in order to save costs and improve efficiency.One of the key research issues for crowdsourcing platforms is to understand users' search intent correctly and help them choose the correct services.However,the current service search of the crowdsourcing platforms has the following problems:(1)Most search methods are based on exact string matching,and do not achieve "soft matching" by understanding user semantics.Unless users accurately express their search intent,string-based exact matching will not be able to get the services that users want to search;(2)The platform lacks service text information due to the use of multiple pictures to display service content,which leads to the traditional statistical natural semantic search solutions for language processing more difficult to function.(3)Unlike general domain search such as Google and Baidu,software crowdsourcing platform search is a domain vertical search,which requires understanding of professional domain knowledge,so it is necessary to introduce domain knowledge to help search.In this context,this paper proposes a semantic search method based on knowledge graph for software crowdsourcing service platforms.This method first uses the rich information contained in the knowledge graph to enrich the information,such as description,attributes,synonyms,and subordinate words,etc.,to make a semantic expansion and make a preliminary search recall of related services.At the same time,the entity information of the knowledge graph is also used to help enhance the expression of the word embeddings,and to obtain the semantic relevance feature from the similarity of the word embeddings;construct the topic model,and extract the topic features from the topic probability distribution model.Then use these two types of features together and use the learning-to-rank algorithm to reorder the search results of the preliminary recall,so that more relevant services have a higher ranking.The contributions of this paper are summarized as follows:(1)Build a software crowdsourcing specific knowledge graph from the software service data and external data such as CN-DBpedia common domain Chinese knowledge map,Wikipedia,Baidu encyclopedia,Stu Q brain map and other external data.(2)Use the entity information in the knowledge graph,such as descriptions,synonyms,subordinate words,etc.,and make semantic expansion to user search statements and service description information to resolve user search statements that may not clearly express intent and lack of service text information problem.(3)Use the entity information in the knowledge graph to enhance the expression of the word embeddings,integrate the semantic information in the knowledge graph into the traditional word2 vec word embeddings,and build the re-ordering features of semantic relevance through the enhanced word embeddings.(4)Propose a hybrid feature extraction method using an auto-encoder for the topic model to extract the reordering features of the multi-level topics.This paper uses the data of Joint Force software crowdsourcing platform to carry out a series of experiments.Experimental results show that,compared with the existing DSSM algorithm,the proposed method improves the accuracy,recall,MRR and NDCG search result indicators by 42.633%,42.633%,27.465% and 34.977%,respectively.