Research And Application Of Note Recommendation Model Based On Two-channel Deep Topic Feature Extraction

Under the background of the highly developed Internet and the explosive growth of online content,the phenomenon of "information overload" has become one of the main problems of daily access to information.In the online teaching system in the field of education,teachers and students use online notes to manage teaching reference resources,but the system often lacks reference materials other than textbooks.These materials are closely related to the needs of users and have individual characteristics.In the teaching environment,traditional search engines may not be available.Therefore,recommending personalized reference materials to users according to actual application scenarios in the note-taking system is a topic worthy of study.Through research,it is found that a personalized recommendation engine based on deep learning design is a feasible solution.Firstly,the application status and characteristics of recommendation systems in different scenarios are analyzed through literature retrieval and review;on the basis of parsing and reproducing the current mainstream recommendation models,this paper proposes an article recommendation model based on dual-channel deep topic feature extraction(TCDTM,Two-channel Deep Topic Model);using this model to design and implement a classroom cloud note-taking software with online resource recommendation function.The main work and contributions of this paper are as follows:(1)A two-channel deep topic feature extraction article model TCDTM is proposed.Compared with other recommendation models based on feature engineering semantic similarity measure,this model adds topic information,combines topic information with pre-trained context representation,and enriches the expression of semantic features.Through comparative experiments and ablation experiments,the effectiveness of the model is verified.For example,the performance of the NDCG@3 and NDCG@5standards on the Aminer dataset is improved by 6.1% and 7.2%,respectively,compared with the optimal comparison method.(2)Based on the improvement of the TCDTM model,a recommendation framework TCDTM-os is proposed.Aiming at the cold start in the process of personalized article recommendation and the problem of sparse user feedback,single-sample learning is integrated into the recommendation framework,and personalized recommendation is realized by combining the personalized score in the support set and the article similarity score.Experiments show that adding single-sample learning can effectively alleviate the cold-start problem of the recommender system.For example,on the Aminer dataset,the NDCG@3 and NDCG@5 standards are improved by 3.4% and 3.2%,respectively,compared with the original method.(3)A classroom cloud note-taking software was developed using TCDTM-os.Firstly,the current situation of the note-taking system is analyzed and sorted,and the main functions and technical selection of the software are determined.Then,the recommendation framework TCDTM-os is used as the recommendation engine,and finally the front-end interface and back-end interface of the software are realized.Experiments show that the classroom cloud note software can recommend personalized and content-related online reference resources for teachers and students.Research and practice show that there is room for improvement in terms of models and recommended content types.The proposed model and software have practical significance and practical value.