Research And Design Of Scratch Project Recommendation System Based On Project Similarity

In the information age,programming education for teenagers attracts considerable attention.Because of its convenience and interesting,Scratch has become the preferred tool for learning zero-based programming.Scratch advocates a remix learning strategy and encourages students to discover,learn and remix,so as to achieve the purpose of improving programming ability.However,due to the uneven level of projects and the lack of appropriate measures in domestic and foreign platforms,it is difficult for learners to find projects that are similar to their programming level and interests,which brings great difficulties to learn from case study.Therefore,in-depth research on the structure of Scratch projects,discovering the similarities,and designing a project recommendation system,is of great significance to help learners increase interest and improve learning effects.Based on the in-depth study of Scratch code characteristics and similarity analysis,this thesis focuses on the similarity measurement and recommendation of Scratch 3.0 projects.In terms of the similarity measurement of Scratch projects,firstly,a code analysis scheme of logical reconstruction and sprite sorting is proposed,and a word embedding model for Scratch 3.0 is constructed;then based on Siamese network and using Bi-LSTM as a unilateral network for feature learning,the Bi-LSTM Based Siamese Network for Similarity Measure of Scratch Project(BSSM)is designed.In terms of project recommendation,in order to reduce the pressure of searching and comparison of full projects,firstly,based on project category and code similarity,proposed a double-level classification algorithm(DCPL).Combined with the user's personalized recommendation needs,a fast recommendation algorithm for Scratch projects is proposed.The experimental results show that the Scratch analysis and reconstruction technology designed in this paper is feasible and effective,the trained word embedding model can fully recognize Scratch standard blocks,The Scratch project similarity measurement model achieves an accuracy of 0.90.The proposed DCPL algorithm achieves a classification accuracy of more than 0.85 in each layer,the recommendation system can effectively reduce the server load while providing users with personalized project recommendations.