Research On Type Transformation Sequence Recommendation Combining With Semantic Web Technology

With the rapid growth of the software scale, more and moredevelopers prefer to adopt some mature frameworks during thedevelopment process to improve the efficiency. However, there is anextremely common scenario where developers know the type of objectbut totally confused in how to get it in their program context. Ifdevelopers cope with this problem through reading related documentationand source codes, it will take lots of time and efforts. Consequently, thereis a need to improve development efficiency through coderecommendation tools. Therefore, how to provide an effective method torecommend code snippet in the above scenario becomes the focus ofmany researchers’work.In the existing research works, the proposed recommendationmethods can be classified into two categories: the method based onhistory code and the method based on information of program context.The former one compares the program context with history code to findout similar scenes. The source code in these scene will be provided todevelopers as example codes. The method based on information ofprogram context analyzes the type transformation relationship throughprogram code mining, and computes the type transformation sequence bycombining with the current program context. Both two methodsmaintain disadvantages: methods based on history code relies on thehistory code, but the quality of history code is hard to guarantee; methodsbased on information of program context have limitations when thesequence starts with a global member, which may lead to a spaceexplosion.Semantic Web technology allows computers recognize therelationship in data instead of text content so that computers can provideservice in an intelligent way. Through defined classes and attributes in ontology as well as inferable rules, computer can make certain judgmentsand inference, thus assisting human beings on the tasks of informationcollection, organization and retrieval. In the type transformation sequencerecommendation problem, computers need to understand the relationshipof codes and to provide effective code retrieval capability. The SemanticWeb technology provides a new approach for this problem. With the fastdevelopment of Semantic Web these years, it has been outstanding inmany domains include program language. There are some works usingSemantic Web in software evolution and code clone detection, which laida solid foundation for the research of this thesis.This thesis presents a complete recommendation method for typetransformation sequence through improving the method based oninformation of program context. It efficiently solves the space explosionproblem when recommending the sequence starts with a global member,through decreasing the number of the global member entry points usingreachability analysis. Based on the proposed method, an Eclipse plug-inis implemented for evaluation. Comparing with the existingrecommendation tools, the result of experiment shows that the methodproposed in this thesis can not only provide a well hit rate whenrecommending the sequence starts with global member, but also give abetter accuracy when recommending other sequences as expected.The main research work and achievements of this thesis include thefollowing:(1) Analyzes the existing works of type transformation sequencerecommendation. According to the advantages and disadvantages of theseworks, this thesis proposes a complete method for recommendation.(2) Creates a proper ontology model for the semantic description oftype transformation. Solve the problem of space explosion whenrecommending a sequence starts with a global member.(3) Implements an Eclipse plug-in based the method proposed by thisthesis.(4) Designs a comparison experiment on several open source projects.Compare with the existing tools, the result validates the efficiency of theproposed method.