Collaborative Filtering Recommendation Method And Application Oriented To Big Data In Educational Scientific Research

The rapid development of information science has brought an enormous amount of information,making it harder and harder for users to get valuable information,namely,users being confronted with a severe problem of information overload and information anxiety.Recommender systems,aiming at providing individualized recommendation services for users by extracting their hidden preferences based on historical data,prove to be effective means to reduce information overload and information anxiety,so they have attracted wide attention from both the academic and industrial fields.The number of high-quality resources in the education field,especially video resources in teaching and academic paper resources in scientific research,has shown explosive growth.How to accurately recommend high-quality video resources and academic paper resources for learners is the focus and difficulty of current research.Currently,mainstream recommendation systems make recommendations by digging users’interests and preferences from historical interactive data of users and items.However,these recommendation systems are faced with various challenges,such as data-sparse,user or item cold start,expandability,robustness,explicability,and so on.To meet these challenges,numerous researchers have given their own solutions.However,we cannot address these problems solely by relying on some limited information like user and item scoring matrix.In recent years,the everchanging information technology including big data,artificial intelligence has made more and more Multi-source heterogeneous information recorded and stored,for example,user rating information,social network information,item attributes,user attributes,location,time,etc.Most of this multi-source heterogeneous information contains users’ interests and preferences,which provides means to alleviate the problems facing recommendation systems.Meanwhile,there is a difference in internal importance among information of the same source.How should different source information be integrated?These are significant researches to improve the performance of recommendation systems and to tackle the problems facing recommendation systems.Big data technology brings opportunities for conducting educational scientific researches efficiently,however,there are still challenges to precisely recommend high-quality educational resources such as video resources academic papers.This article is written on the basis of the collaborative filtering recommendation model based on matrix factorization by making use of user side information,item side information,and user-item interactive information.Researches on how to pay attention to the key information in auxiliary information,how to use more auxiliary information of users and projects,how to achieve the integration of auxiliary information features,how to design paper recommendations,and particularly the common problems and difficult problems faced by the recommendation model based on deep learning in the field of high-quality educational resources were carried out.Focusing on the above issues,the author conducted researches on the recommendation model and application from the following aspects.(1)Deep hybrid recommendation algorithm incorporating an attention mechanism.A large number of researchers make use of user or item-side information to reduce the problems of datasparse and cold start in video recommendation,which prove to be effective to some degree.However,these researches fail to pay attention to the key information in auxiliary information.To solve this problem,the author proposed a deep blended recommendation model integrated with a dual attention mechanism.This model extracts hidden factors of the item side with a convolutional neural network that is integrated with the self-attention mechanism.Meanwhile,it extracts hidden factors of the user side with a stack denoising auto-encoder that is integrated with the self-attention mechanism.In this way,it can deeply mine important information from both the item and user sides,which is then combined with probability matrix factorization to realize video scoring prediction.Repeated experimental results in two public datasets show that the proposed method outperforms the most advanced baseline model.(2)An auto-encoder collaborative recommendation model based on user attributes and video categories.Video recommendation is an essential part of the recommendation service.Presently,numerous researches utilize user or video side information to ease the problems of data-sparse and cold start in video recommendation.Progress has been made compared with traditional models.However,user and video side information is very complicated.Moreover,the importance of different information varies,even the same piece of information has a different significance to different users or items.Based on this fact,the author put forward an autoencoder that makes use of user attribute information and video type information simultaneously.This autoencoder integrated with attention mechanism focuses on more than one key information of the side information.It combines user-item rating matrix and probability matrix factorization to achieve scoring prediction.A number of experiments in two public datasets demonstrate that the proposed method can effectively improve video recommendation quality compared with the state-of-the-art model.(3)An academic paper recommendation model based on the integration of deep features.As the number of published academic papers keeps growing,how to search for relevant research papers from the countless papers becomes more and more difficult.Most researchers search by keywords or follow the cited papers or authors and other publication information to search for related articles.It is time-consuming but unsatisfactory in result sometimes.Therefore,the author proposed an academic paper recommendation model based on the integration of deep features.It uses title and abstract text information and citation information.It uses a dual auto-encoder that is integrated with a multi-head attention mechanism to learn the potential factors,which are then weighted and combined with matrix factorization to achieve individualized paper recommendations.Experimental results in two public datasets show that the model put forward gets better results compared with the state-of-the-art baseline models.(4)A paper recommendation in educational research big data platform.Educational research big data platform collects and sorts out information including scholars,institutes,policies,projects,journals,conferences,papers,patents,books,and prizes in the educational field,aiming to provide a big data service platform for researchers in the educational field.The research in this dissertation provides technological support for this platform.The framework and detailed implementation strategies have been put forward.The author divided paper recommendation into a hot paper recommendation,citation recommendation,lexical recommendation,content recommendation,and blended recommendation.As for a blended recommendation,through innovative points such as the breakthrough attention mechanism,multi-source heterogeneous information,and feature weighted fusion in the previous three chapters,a combination of text information including user paper interactive data,user attribute information,paper category information,title and key words,network information of citation between papers and paper quality are made full use of through the deeply blended recommendation system based on matrix factorization.Meanwhile,attention has been paid to important information in different auxiliary information and auxiliary information on the paper side is weighted to realize auxiliary recommendations.In summary,the research in this dissertation has significant study and application value,especially in this AI era.Making full use of big data and deep learning technology,the author studied collaborative filtering recommendation method oriented to big data in educational scientific research,explored the application of deep learning in high-quality educational resources recommendation,serving as a solid foundation for application in other fields.