Research On Personalized Recommendation Algorithm Based On Social Network Trust

Nowdays,the Internet is increasingly popular.The development of computer technology such as big data and social networks has made a breakthrough in the research of recommended algorithms.With the developed internet,users can easily obtain the information they want,but as the amount of information in the network increases,the ability of users to receive and process valid information decreases in a short period of time.Faced with information overload,the traditional artificial filtering search algorithm can not adapt to the current form.In response to this problem,personalized recommendation system came into and gradually formed a certain theoretical foundation under the guidance of many experts and scholars.The principle of the recommendation algorithm based on the traditional collaborative filtering(CF)is recommending according to the users' similarity,but the recommendation accuracy is low due to the difficulty of finding similar users.Therefore,introducing the social network trust degree recommendation into the traditional recommendation system is an effective way to solve the problem when the user rating data is relatively sparse.However,the current recommendation algorithm has a simpler method to deal with the trust degree and does not effectively mine trust users.In view of the low accuracy of the traditional collaborative filtering recommendation algorithm,this dissertation presents a personalized recommendation algorithm based on the trust of social networks.Aiming at the problem which the trust degree matrix of social network is sparseness,this dissertation proposes a dual-network time-domain evolution model based on the user's degree of deviation.According to the user's personal activities and relational networks,the model is used to enhance quality and efficiency of the recommendation by using the trust networks and similar networks on the social network.This model reduces the sparseness of trust network by creating a dual-trust network(be consists of a similarity network and a trust network for each user)on the basis of establishing users' bias,and then dynamically updates the double trust network by taking advantage of trust's dependence on time.According to the theoretical model,the dissertation makes relevant contrast experiments and it's results show that the proposed model has significant advantages in improving the quality of recommendation of sparseness problems,and has better comprehensive performance and practical value.DNTDEM is a proposal that lays a solid foundation for the next step.The dissertation makes a personalized recommendation for micro-video(a new form of user-generated content)on the basis of trust recommendation algorithm.The micro-video has characteristics such as low quality,ineffectiveness,providing the unique complementary information in different modes,all of which contribute to the recommendation of micro-video.The dissertation studies the issue of personalized micro-video recommendation through the exploration of social relations,themes and emotion analysis for users and micro-video.In the thematic analysis,the key word sequence of each micro-video is identified based on the LDA model,and the similar video list can be found by using the Jaccard similarity.In the emotion analysis,the comment content of micro-video is fully considered to accurately calculate the user's emotion similarity.On this basis,the algorithm can recommend personalized micro-vides to users with the using of matrix decomposition and the integration of three elements.The dissertation makes some relevant contrast experiments and it's results show that the micro-video personalized recommendation algorithm integrated in multi-factor in social networks can improve the accuracy and recall rate by more than 30%,and has better recommendation effect and application value compared with other basic algorithms.