Study Of Academic Recommendation Based On Collaborative Filtering

The rapid development of Internet makes the explosive growth of information onthe Internet. Because of the huge amount of information in the academic database,when users want in search of the article, it is often quite difficult to fast and accurateto access to the desired information. This greatly increases the user’s cognitive burdenon the acquisition of knowledge. But the existing search engines are more and moredifficult to meet the needs of users. Personalized services arise at the historic moment.Although it had been a lot of personalized service, but for personalized academicresearch were rare. This paper mainly studies the academic recommendation based oncollaborative filtering and meet the personalized needs of users’ information search.This paper presents the method of academic recommendation based oncollaborative filtering technology. Firstly, the existing recommendation technologiesare summarized, and analyze the advantages and disadvantages. In academicrecommendation, how to accurately predict user’s interest is a key issue. So this paperdescribes every step in building the user interest model in detail and analyzes theadvantages and disadvantages of each step to choose the method this paper is reallyneed. Through the method of this paper, we can build the user interest model based onthe concepts and relationships between concepts, on the basis of the interest model,the paper is not only relies on the concept but fuse the concept similarity and therelationship similarity together to find the collaborative users.Secondly, this method is implemented based on refined user profile in whichconcept weights is adjusted. In the adjustment, the continuity feature of user’sbrowsing content is taken into account, which is helpful in discovering collaborativeusers. Based on the new user profile, the concepts that their weights are bigger thanthe threshold are selected as key concepts. Then through recalculate the weight ofcandidate users to redefine the candidate users. Collaborative users are discoveredbased only on key concepts which can improve the efficiency of prediction.Finally, organize all the candidate users’ browse content based on semantics, andextract the related concepts according to the weight, information quantity ispresented as the evaluation attribute, then select the concepts with high informationquantity as the semantic representation of user’s prediction interest. In addition, semantic relations between concepts are considered when computing informationquantity, which can ensure the accuracy of prediction. Experimental resultsdemonstrate the validity and effectiveness of this method.