Research On The Collaborative Filtering Recommendation Algorithm Based On The Combined User Similarity

The rapid development of information technology and the advent of the era of big data have resulted in the growth of data bloating.The accumulation of massive data exceeds the reasonable range of information processing and utilizing by individuals and groups,which leads to information overload.In order to overcome this issue,the recommendation system is proposed,meanwhile,the recommendation system is also a research hotspot in the fields of data mining,information retrieval,and computational advertising.Collaborative filtering recommendation algorithm is one of the classic algorithms in the recommendation system field,whose main idea is to find similar users by the target users,and predict the evaluation of an item by the target users based on the evaluation of this item by the similar users.With the rapid increase of the data size in the recommended scenario,many issues such as high-dimensional sparse data,timeliness,and cold start come to seriously affect the recommendation accuracy of this method.This paper mainly focus on the research of the recommendation system from the definition and structure,analyses the mainstream recommendation technology,and compares the advantages and disadvantages of different recommendation technologies.In particular,aiming at the impact of high-dimensional sparse data on the user's similarity calculation caused by low accuracy and timeliness factors,a collaborative filtering recommendation algorithm based on the user's combined similarity is proposed.The specific research contents of this article are as follows:Firstly,the collaborative filtering recommendation algorithm is studied,and its working principle is described in detail.Furthermore,the existing problems in the similarity calculation and prediction score are also analyzed.Secondly,for the issue that the inaccurate similarity calculation caused by high-dimensional sparse data,a solution to the user's combined similarity is proposed.The user's combined similarity has two structurally measurements including user's preference distribution comparison and user's personal attribute information comparison,and the similarity calculation result between users is obtained by linear combination of these two measurements.The user's preferences distribution can be obtained by LDA model processing the integrated data,the similarity of different distributions is measured by the Bhattacharyya coefficient,and the scoring data in the process is onlyviewed as a screening basis,which slows down the effect of sparse data on calculation results.The comparison of attribute information is measured by the Hamming distance in the information theory after the personal attribute information is quantified,and the difference between individual user attribute information is obtained.Through the above process,the comparison of the user's preferences distribution is finally required.Thirdly,the effect of timeliness on the prediction scores is illustrated by the hot value attenuation of the specific item.Based on the hot value statistics of all items,an inverse proportional function is adopted to quantify the attenuation degree.After combining the hyperbolic tangent function mapping,a hot value impacted factor is obtained for the prediction process.The hot value impact factor reflects the change in the heat of the item and embodies timeliness' s influence on the predicted results.Finally,the combination of the user similarity and the improvement of the predictive computing constitute the collaborative filtering recommendation algorithm based on user's combined similarity.The experiment on the public data set testified the effectiveness and accuracy of the proposed algorithm.