Research On Personalized Knowledge Recommendation Method For Engineering Design Task

With the advent of the era of knowledge economy,knowledge has become an important asset to enterprises,and enterprises in the engineering field are also paying more attention to creating value for itself by “knowledge reuse”.However,in the face of the explosive growth of knowledge,the inefficient traditional knowledge reuse methods cannot meet the knowledge needs of engineering users,which severely affects the value transformation of enterprise knowledge resources.In this regard,using the advanced personalized recommendation algorithm for engineering knowledge recommendation can effectively improve the efficiency of knowledge reuse,assist engineers to quickly solve engineering design problems and complete engineering design tasks,thereby improving enterprise production efficiency,and achieving innovative development.However,due to the neglect of domain features,such as engineering semantics and engineering context,the direct application of traditional personalized recommendation algorithms for knowledge recommendation may cause problems of “popular recommendation” and “homogeneous recommendation”,as well as “sparse matrix” and “cold start”,which can lead to the effect reduction of knowledge recommendation results in a specific engineering context.To rectify these problems,the research was carried out from two different perspectives of “diversity” and “accuracy”on how to meet the knowledge needs of users in different engineering context.In addition,an application case study of proposed theories and methods was carried out in the background of ship outfitting design to verify their effectiveness and advancement.The main contents of these research are outlined as follows:(1)An engineering knowledge recommendation method considering the diversity of topics was proposed to solve the “popular recommendation” and “homogeneous recommendation” problems.By combining the defined content similarity,application context similarity and task similarity of knowledge documents,a knowledge relevance network was built for knowledge topic discovery.Based on the divided knowledge topic communities,the user-topic professionalism was defined to improve the user-based collaborative filtering method,and the user-knowledge interest model and user-topic interest model were built to implement the context-aware diversity-oriented recommend strategy by considering the defined knowledge context-based effectiveness simultaneously.(2)Aiming at the “sparse matrix” and “cold start” problems,an engineering knowledge recommendation method based on the context-aware generative adversarial network model was proposed.Based on the user background information and behavior log,the user background context vector and question context vector were constructed to get the user representation vector contextual-detailed.By constructing the semantic spreading activation model based on knowledge relevance network,the semantic extending vector was constructed for semantic extension of user historical rating vector to fully explore user interests.To obtain the high recommendation efficacy,the generative adversarial network model(GAN)was introduced,and the collaborative training method,using the user representation vectors with context information as training dataset,was proposed to construct the context-aware GAN(Cxt GAN)model.The Cxt GAN can be used to solve the “sparse matrix” and “cold start”problems and ensure the effectiveness of the knowledge recommended for the target user with a specific engineering design question.(3)An empirical research on the application of the proposed knowledge recommendation methods in a domestic shipyard was conducted.For the knowledge recommendation in the research,the specific engineering design task of "engine room layout design" and the engineering design problem of "engine room pipeline layout design" included in the engineering design task were taken as examples,and by collecting the user personal information,behavior log data and corresponding knowledge document corpus data from the knowledge management system of the shipyard,the two proposed methods were implemented to recommend the appropriate knowledge for the engineering task and engineering problem.The corresponding diversity and accuracy evaluation index were used to verify their effectiveness and advancement.