Research On Braille Book Recommendation System Based On Fusion Preference And User Attention

With the development of the Internet,the information explosion caused by massive data needs to be solved uergently.People are faced with the problem of selection jin the data stream of millions.The recommendation system came into being.For visually impaired users,it's difficult to obtain information through the sense of sight because of visual defects,but only rely on other senses such as touch,hearing,etc.,ordinary reading methods can not meet their needs for knowledge acquisition.The book recommendation system algorithm commonly used on the market is applicable to able-bodied people but not the men visually imparement.Therefore,book recommendation system algorithms that suit their behavior habits are required.In order to provide accurate and effective book sets for users,this thesis proposed a algorithm based on fusion preference.The thesis explored users' behavior-based interests and content-based intersts to predict the preferences of visually impaired users,and conbined these preferences to recall books.In the calculation of interest preference,this thesis improved content similarity method,constructed the user fusion preference by using text feature,and realized the expansion of the recall book set on the sparse data.This algorithm solved the problem that insufficient recommended books caused by data sparseness,and effectively improved the accuracy of recommended books for visually impaired users.For intelligently ranking the recall book set and help users find right book quickly,this thesis proposed a context-aware algorithm based on users' attention.The thesis introduced an improved calculation method for user attention to solve the problem of inaccurate prediction on users' current attention when their action history is cluttered.This thesis used the attention mechanism to preserve the users' reading tendencies during the reading process,analyse users' personal features and book content features,and improve the accuracy of the recommendation by merging the feature space.This algorithm adapted to the sequential browsing mode of visually impaired users,and effectively assists them to quickly find books of interest.Finally,the improved algorithm proposed and comparative experiments were employed on the dataset collecting from the China Braille Library and fit the habit of visually impaired users.The effectiveness of the improvement is proved in each experimental comparison results.