The Research And Implementation Of Personalized Recommendation Based On User Profile

With the prevailing of resources sharing website,the shared resources are exploding on the network,it is more and more difficult for users to find the information which they interested in.Resources accuracy of current recommender systems,mainly depends on the correlation matching of the query and resource description,although different users can input the same query conditions,but they may have the different demand for resources because of the different interest,therefore,is of great significance to the personalized recommendation.At present, the researches of personalized recommendation are mainly focused on how to get the user's interest model and in the form of a suitable, as well as how to apply interest model to recommend system in order to improve the accuracy of resources and user satisfaction. Traditional recommendation methods usually do not have the comprehensive treatment for query result, not only does not take into account the user's individualized demand, is also not do the detail study on the characteristic of resources, so the retrieval is poor and user satisfaction is low. At the same time, the birds of a feather flock together, although everyone has his own interests, but making the users with same interests into group, is of great significance to improve the performance of personalized recommendation. To solve above problems, this topic combined with group recommendation, proposes a personalized recommendation method based on user profile.At first, in view of the current user profile and resource profile are based on the TF-IDF(inverse document frequency)method to construct, the TF-IDF is a traditional method to construct profiles, but this method only consider the absolutely term frequency, the total number of users and the number of users who use the label when building user profile, lead to the tag weight of active user will be far higher than that of ordinary users. therefore, this paper puts forward the improvement of TF-IDF method,the fusion processing is conducted with the number of tags which the user had used and the total number of resources which the user had marked while constructing user profile, the tag weights of active users has relatively reduced, also, the fusion processing is conducted with the number of tags which had labeled the resource and the number of users who had annotated the resource, so that the profile could more accurately reflect the characteristics of users and resources.Second, this paper had further studied the traditional cosine similarity calculation method, and had put forward improvement strategy to improve the accuracy of the retrieved result as a whole, that is combining with the number of tags which has matched.Then, on the basis of the traditional recommendation system, calculated the correlation of the preliminary result and the computed user profile, had applied the user interest model into search effectively. The effectiveness of the proposed method is verified by some related experiments.Finally, used the method proposed in this paper to experiment with MovieLens dataset repeatedly, and had collected users into groups using clustering technology according to the similarity of user profiles, had computed the user group interest to updating user's interest model, combined personalized recommendation and group recommendation perfectly, and make the recommendation result more convincing.