Research On Collaborative Filtering Algorithm Based On User's Preference For Attribute And Common Rating

With the advent of the Web 3.0 era,the issue of "information trek" is imminent.The recommendation system is the third method of dealing with information overload problems that was born after the catalog and search engines.It analyzes users' history behavior to mine users' interests to meet users' individual needs.The collaborative filtering recommendation algorithm is an important algorithm in the recommendation field.It uses the collective wisdom of the cluster in which the target user is located to recommend to the target user.Because the algorithm principle is simple and easy to implement,it is widely used in business,but the algorithm still faces some problems: 1)In the case of data sparse,the recommendation accuracy is not high due to low-precision similarity calculation;2)relying solely on users' rating data to calculates the users' similarity,and ignores the influence of the item attribute characteristics on the user's interest preference.3)The lack of consideration of the user's interest migration caused by time effect,the algorithm can not adapt to the change of user interest over time.These problems hinder the performance improvement of the recommendation algorithm.Therefore,this paper studies the similarity measure in the collaborative filtering algorithm based on the above problems.(1)This paper designs a similarity measure model that considers the user 's preference for attribute,and introduces attributes of an item into the similarity calculation of the recommendation algorithm as an important factor affecting the user's interest.Since the user's rating of an item is the result of the user's comprehensive evaluation of the various preferences for attribute of an item.This paper maps the user's rating to the rating of attributes,and uses each rating to calculate the user's attribute interest.The sparse user-item rating matrix is transformed into a relatively dense user-attribute preference matrix,and the cosine similarity is used to calculate the user's preference for attribute similarity.This model solves the problem that the traditional similarity model performs poorly in the case of sparse data to some extent.By introducing attribute features,the search range of neighbor users is expanded.The experimental results show that the similarity model considering user's preference for attribute improves the coverage and accuracy of the recommendation algorithm.(2)This paper adds a time dimension based on the user 's preference for attribute model to track the user's preference for attribute over time.By introducing a time decay function,the forgetting function,the user's memory retention ratio of the attribute is calculated with the function,and the ratio is used as the weight of the user attribute rating,and the attribute rating of the earlier time is penalized,and the attribute rating of the newer date is weighted,and we can get the user's corrected attribute rating.Then we can get user's interest degree for the attribute by accumulating the corrected rating value and establish the user-attribute preference matrix.The user's preference for attribute similarity is calculated by using the cosine similarity model.The simulation results show that the user's preference for attribute model considering time effect further improves the accuracy of recommendation than the original proposed model.(3)Finally,in this paper,the Jaccard similarity coefficient based on the user's common rating is used as the relaxation factor,and the user 's preference for attribute similarity model considering the time effect is blended with the traditional common rating similarity model linearly.The relaxation factor can adjust the weight of two models in the fusion model dynamically according to the ratio of common rating.When there are more common rating items between users,the weight of the common rating similarity model increases,and vice versa.The experiment proves that the fusion model has a significant improvement in recommendation accuracy and coverage,and has improved the recommendation quality.