Research On Method Of Requirement Pattern Mining Based On Knowledge Graph

In the current complex Internet service environment,the double-blind phenomenon of service supply and demand side is widespread.Service demanders know little about external services and related domains,so it is difficult to accurately and fully express requirements.However,since the service providers have a limited user demand information,it is difficult to provide an accurate and appropriate service.In order to solve the above problems,we need to obtain and comprehensively analyze a large number of user requirements and extract demand patterns to help users match fuzzy requirements and provide provide potential user requirements for business.However,in the network,the user's requirements come from heterogeneous data sources,and the demand expression has diversity,non-standard,and unstructured features.Therefore,based on the requirements of fragmented users,obtaining valuable demand information from discrete data through knowledge extraction,knowledge fusion,and knowledge mining will be of great significance to the current service supply and demand market.This article uses the short text of the user requirements description in the crowdsourcing service site Free Lancer to study the above issues.First,each unstructured text is constructed into a structured ontology form through an entity relationship extraction method.Then we combine large-scale demand ontology into a heat integrated demand graph through the knowledge synthesis method.Among them,due to the heterogeneity of user expressions,we use entity-based and structure-based alignment methods to align entities in the string and semantic levels respectively;Then based on the synthesis requirement graph,we use the probability graph model to extract the link and cluster patterns defined by our research.In order to avoid the direct search in the huge requirement graph,we propose a pattern mining method based on domain perspective,which abstracts the large-scale requirement graph to the domain knowledge space.In this process,we identify the domain entity nodes through a domain term extraction algorithm,and adopt the represent learning method to complement the association between discrete domain knowledge.Finally,based on the above research results,we designed and developed a user requirement pattern matching tool to assist users in writing requirement texts through pat tern mining methods.Most of the user requirements on the Internet are fuzzy and noisy because there is no unified expression specification.Therefore,it is very difficult to study the above issues on this basis.At the stage of entity relation extraction,we made up for the deficiencies of the existing NLP tools through the rules definition;At the graph synthesis stage,we rely on the support of existing methods,integrated and improved various ontology alignment methods.For the demand pattern mining,we innovatively proposed a pattern extraction method based on the domain perspective to mine user demand patterns in a multi-granularity space.Finally,we proved the effectiveness of the methods we designed through comparing experiments,and proved the research value and significance of this topic through application examples.