Research On Service Matching Approach To Solve The Data-sparsity Problem

Service-oriented architecture technology has gradually become an important form of software development on the Internet.With the increase in the number of services,the matching of service supply and requirements has become increasingly important,which aims to help developers quickly and accurately find services that meet user needs.However,the types of services tend to be diversified,and the content of service descriptions is often nonstandard,which makes both structured description and natural language description services face serious semantic sparse problems,which brings great challenges to efficient service supply and requirements matching.Focusing on this problem,this thesis starts from analyzing the characteristics of two types of service descriptions,structured and natural language,and studies the corresponding efficient matching methods from a functional perspective.The main research work includes:(1)For the services described by structured language,a service supply and requirements matching method based on attention mechanism and neural topic model is proposed.The method first extracts the keywords related to the service topic from the labels and attributes of the structured description,enhances the sparse semantics through the external knowledge base,and generates the topic distribution through the neural topic model on the semantically enhanced text.Then,the encoded text vectors are weighted and aggregated using topic distribution and attention mechanism to improve the quality of text representation vectors.(2)For the services described in natural language,a service supply and requirements matching method based on functional goal enhancement is proposed.Based on the external knowledge enrichment mechanism,this method is further combined with the functional goal enhancement mechanism for matching.The function target is extracted from the service description by rules,and then the vectorized expression of the function goal is combined with the topic vector,and the weight of the attention mechanism is optimized,thereby improving the quality of the natural language text representation and improving the service matching effect.For services described by structured and natural language,experiments are carried out on public datasets such as SAWSDL-TC3,OWLS-TC4,and Programmable Web,respectively.The experimental results show that the method proposed in this thesis is superior to several current mainstream service matching methods in terms of accuracy,recall,NDCG and other indicators.