Research On Personalized Recommendation System Combining Item Information And User Information

With the emergence and rapid development of Internet technology,it has had a huge impact on people's lives.Advertising is one of the important ways for people to obtain information.In real life,the advertisement push system is targeted to continuously play the same advertisement,which lacks pertinence and interactivity.Therefore,this thesis studies and designs a personalized recommendation system that combines with item information and user information.The system combines personalized recommendation algorithms with face recognition and expression recognition.After recognizing the user's face,it pushs advertisements based on the target user's historical behavior data and facial expressions,and then conducts human-computer interaction based on the target user's facial expression.Finally,the pushed advertisements can stimulate users' potential buying interest.The work content of this thesis is as follows:(1)In view of the sparseness and scalability problems of the user-item rating data matrix in the neighbor-based collaborative filtering recommendation algorithm,this thesis proposes a hybrid collaborative filtering recommendation algorithm based on inter-item dependency and users to effectively alleviate these two problems.The algorithm can be divided into the following two parts:Based on the item-based collaborative filtering recommendation algorithm,the concept of inter-item dependency is proposed and weighted with the modified cosine similarity formula,that is,the similarity calculation method based on inter-item dependency;On the basis of the user-based collaborative filtering recommendation algorithm,it is preferred to use the similarity calculation method based on the inter-item dependency to obtain the similarity coefficient between each item;Then use the formula of predicting the item score to obtain the target user's score for the unrated item.Thus,the user-item rating data matrix is filled.Then on this basis,according to the different feature attributes of the items,use the dichotomous K-Means algorithm to classify to form the sub-user-item rating data matrix;Finally,a user-based collaborative filtering recommendation algorithm is used to generate a list of item recommendations.(2)Using user information,research on interactive personalized recommendation algorithms based on facial feature point detection and expressions.Aiming at the problem that the embedded platform has limited resources and cannot quickly detect facial feature points,a rapid detection algorithm for facial feature points based on the golden ratio of faces is proposed.Experimental results show that this method can reduce the memory consumption of the model by 74% compared with the CNN model without losing a certain accuracy and ensuring a certain speed.After the facial feature point rapid detection algorithm based on the golden ratio of the face is used to achieve face alignment,it can effectively improve the accuracy of face recognition and expression recognition;Then combine with the user's historical behavior data to complete personalized recommendation;Finally,using the Raspberry Pi 3B+ as an embedded development platform,designs and completes a personalized recommendation system that combines with item information and user information.