Research On Collaborative Filtering Recommendation Algorithms Based On Positive Correlation And Negative Correlation Nearest Neighbors

After entering the new century,with the increasing popularity of computers and smart phones and the rapid development of the Internet,the Internet has infiltrated into all aspects of our daily life.However,the information in the Internet shows an exponential growth,and this phenomenon will definitely lead to information overload.Therefore,it's getting harder and harder for people to acquire the information they need in the massive data of the Internet.The emergence of personalized recommendation system has provided a new coping tool for people in the era of Internet data explosion.Its outstanding performance has drawn continuous attention of researchers from all walks of life.Unlike traditional search engines,recommendation systems could actively provide users with accurate and personalized recommendation results,so it can enhance customer satisfaction.This dissertation discusses the relevant background knowledge of the recommendation systems,and introduces a variety of popular recommendation algorithms,as well as commonly used evaluation indicators.As the collaborative filtering recommendation algorithm is currently one of the most far-reaching and the most commonly used recommendation algorithm,has received a high degree of attention,and the two algorithms proposed in this dissertation are based on the research of collaborative filtering recommendation,so in the second chapter,the process of collaborative filtering algorithm is described in detail.The traditional collaborative filtering algorithm has some disadvantages,such as low diversity and low accuracy caused by sparse data.In order to solve these problems,this dissertation analyzes the steps of collaborative filtering algorithm,and finds that the traditional method only considers the positive correlation nearest neighbor while ignoring the negative correlation nearest neighbor.Therefore,this dissertation proposes a collaborative filtering algorithm based on positive correlation nearest neighbors and negative correlation nearest neighbors "PNCF".Firstly,this algorithm computes the similarities and coefficient of variation between users and sorts the neighbors according to the similarities which modified by the coefficient of variation.Secondly,select the positive nearest neighbors and the negative nearest neighbors based on the rules,then predict item ratings based on the positive nearest neighbors and the negative nearest neighbors respectively.Finally,the last prediction result is acquired by combining predicting ratings based on the positive nearest neighbors and the negative nearest neighbors with weight.Experiment result demonstrates that the proposed algorithm improves the accuracy and diversity of recommendation effectively.Although the PNCF algorithm proposed in this dissertation improves the accuracy and diversity of recommendation to some extent,the accuracy of the calculated similarity is low when the number of common items between users is too small,especially when the number of common items between users is zero,the similarity between users can not be calculated.Therefore,this dissertation further optimizes the similarity calculation of the PNCF algorithm,and proposes an improved collaborative filtering algorithm based on positive correlation nearest neighbors and negative correlation nearest neighbors "IPNCF".When the similarity of the algorithm is calculated,if the number of common items between users is less than the given threshold,two types of feature vectors are constructed for the users using the user-item ratings matrix and the item-attribute matrix.Then,the similarity between the corresponding feature vectors is calculated respectively.Finally,the similarity values are weighted to get the final similarity value.The experimental results demonstrate that the PNCF algorithm and IPNCF algorithm can effectively not only improves the accuracy of the recommendation,but also increases the diversity.