Hybrid Recommendation System Based On Attention Model

With the rapid development of information technology,especially big data technology,great changes have taken place in the way users obtain information.Owing to the explosive growth of information,the problem of "information overload" has become increasingly prominent,and the recommendation system has attracted extensive attention,as an effective way to solve the problem of information overload.Collaborative filtering algorithm is the earliest recommendation algorithm.Its simple and easy to implement make it the mainstream recommendation algorithm,and widely used currently.But this algorithm has two main defects:(1)for a specific commodity,whether the target user likes or not is only strongly related to the same category of items,but weakly related to other category of items.The preference data of different categories of items and the strong correlation between the categories of recommended items can influence the accuracy of recommended results.(2)In the selection of recommendation results,mix the similarity of users and the preference degree of similar users by a simple sorting algorithm,which weakens the influence of user similarity and different user preference on the recommendation system,and reduces the accuracy of the recommendation results.With the rise of machine learning,especially deep learning algorithms,Construct new hybrid recommendation system which integrates the traditional recommendation system with machine learning and deep learning is now a research hotspot.Introducing machine learning,especially deep learning techniques,into the recommendation system will effectively improve the above defects and bottlenecks.For this reason,a hybrid recommendation system based on attention model is proposed in this paper.Firstly,the attention model in deep neural network is used to weight the item attributes of specific recommended products to providing user recognition degree of pre-recommended products.Then,the traditional loss ranking model is replaced by Ada-Boosting model,which makes the related evaluation indicators such as accuracy and recall rate greatly improved.At the same time,this paper introduces the concept of average recognition degree of similar users for the first time,which is used to evaluate the average recognition degree of recommended items among similar users.By evaluating ARD,the recommendation system performance can be more accurately evaluated in the user experience dimension.Finally,the correctness and feasibility of the proposed algorithm are verified through system simulation,and the performance comparison and analysis between the improved algorithm and the current mainstream collaborative filtering recommendation algorithm are given.