Improved Recommendation Algorithm Based On BP Neural Network And Data Filling

Recommendation algorithm,as the core of recommendation system,determi nes the accuracy of recommendation to some extent,and the accuracy of recom mendation algorithm has a direct impact on the quality of recommendation resul ts.Therefore,it is of great significance to study and improve recommendation a lgorithm.The classic collaborative filtering recommendation algorithm is the first proposed and the most widely used recommendation algorithm.This algorithm uses the user's historical behavior data to mine and analyze,so as to discover t he user's interests and hobbies,group users according to different interests and hobbies,and recommend products with similar interests and hobbies to them.It is simple to implement and works well,but there are many shortcomings.This paper proposes an improved framework based on BP neural network and popul arity in view of the shortcomings in the scoring formula and user similarity cal culation formula of collaborative filtering algorithm(taking user-based collaborati ve filtering algorithm as an example),and at the same time(abbreviated as BP-P).Specific improvements are as follows:The third stage is to use the similar neighbor user data set to predict the r ating of the items not rated by the target user.The most commonly used predic tion formula is Resnick formula.Through analysis in this paper,it is found that the formula only involves the rating information of target users and similar nei ghbors,and only reflects the rating relationship between target users and similar users,but does not pay attention to the relationship between similar users.In order to solve the problems in the formula,BP neural network(Back Propagati on)was used to replace the original Resnick formula,and the trained BP neural network model was used to predict the scores of the unrated items of target u sers.The BP neural network instead of the original rating formula is designed t o mine the difference information in the user rating frequency to guide the reco mmendation,which makes up for the shortcomings of the original Resnick form ula and improves the accuracy of the recommendation results.The user-based collaborative filtering algorithm is generally divided into thr ee stages,of which the second stage is the acquisition of similar user sets.Pop ularity is to point to in a certain period of time,people often prefer to go well for hot project,the more users go well shows that the higher the popularity,t he program is popular,pop project will affect the accuracy of user similarity,th erefore proposed for the problem of similarity calculation formula in the second stage is to improve,will introduce a popular degree formula,By combining th em with linear weighting method,the similarity can be improved.In order to verify the effectiveness of the improved collaborative filtering r ecommendation algorithm on improving the quality of recommendation results,t his paper applies the above improved framework to the traditional Pearson Reco mmendation Algorithm(PFC),the recommendation algorithm based on cosine si milarity(CSC),and the collaborative filtering recommendation algorithm with m odified cosine similarity(ASCS).The experimental comparison between the impr oved algorithm and the previous algorithm shows that the three improved algorit hms are all more accurate than the original algorithm through observation of th e experimental results,which proves that the proposed improved BP-P framewor k can effectively improve the quality of the recommendation results.Figure[24]table[8]reference[45]...