Design And Implementation Of Personalized Recommendation System Based On Academic Data

The intersection of computer science and other subjects is the key to the development of information technology,and at the same time,scientific research achievements reflect the degree of integration of the intersection of subjects.For scholars,obtaining valuable information from the existing massive academic data and outputting personal research achievements is the only way for their scientific research,and it is also the process of gradually improving the interdisciplinary results.Therefore,design and realization a computer-oriented interdisciplinary service system has very important theoretical value and practical significance.After fully referring to the functions and characteristics of the existing academic service systems,this paper designs an academic recommendation system around issues such as information crossover,academic exchange,personalized recommendation,and cooperation among scholars,then uses deep learning technology to propose the journal recommendation model and academic collaborator recommendation model,they improves the accuracy of personalized recommendation,improves the information retrieval method,and speeds up retrieval.The main research contents are as follows:(1)A journal recommendation model for scientific research achievements is proposed.Aiming at the problems of information asymmetry and information sparseness in journal recommendation,the paper profile and journal profile are respectively described,and a "journal-paper" data model is constructed.By using paper attributes to characterize journal topics,a journal recommendation model LDA-BERT(L-BERT for short)is proposed.First,use word segmentation technology and LDA topic model to classify the topic words which can characterize the paper,and determine the number of topic words combined with the preset perplexity value.Secondly,the BERT model is used to realize the vectorization of subject words,and the vector value representing the journal is determined by combining the idea of mean value.Finally,the Euclidean distance is used to obtain the similarity value of the paper and the journal to be recommended,and the final recommendation result is determined by combining the high similarity value.(2)An academic collaborator recommendation model for scholars is proposed.Aiming at the diversity of scholars' portraits and the difference of academic influence,a "journal-interests" data model is constructed,and the scholar's academic impact factor is used to characterize the characteristics of scholars,and a multi-academic impact factor recommendation model is proposed.First,the importance of different attributes is determined by using the difference of scholars' attributes,and then the scholar's influence formula is obtained.Secondly,use the research interests of scholars to form a journal-based academic circle.Finally,use the influence formula to obtain scholars to be recommended in the academic circle.(3)A multi-dimensional journal retrieval method is implemented.Aiming at how to improve retrieval efficiency,a "journal-topics" data model was constructed by using word segmentation technology and LDA model,then the data that after model training was stored in the database and indexed,so as to meet the needs for rapid retrieval of journals based on keywords.(4)Design and implement an academic recommendation system that integrates personalized recommendation and information retrieval.By analyzing the functional structure of the system,combined with the Django framework,an academic recommendation system for computer-related research is developed,which realizes the recommendation and retrieval of journals.