Research On Deep Learning Recommendation Algorithm For Long-tail Items

The continuous breakthrough and update of Internet information technology drives the rapid development of many industries,and people have stepped into the information age.The development of science and technology provides great convenience for People’s Daily life,and the resulting information explosion also brings a new problem to People’s Daily life: information overload.The emergence of personalized recommendation system provides an effective way to alleviate information overload.Compared with traditional Internet search engines,the recommendation system does not require users to put forward search requirements independently,but predicts users’ personalized preferences by learning their behavior information,and presents the recommendation results to users.Although the recommendation system has made some progress,there are still some challenges,such as data sparsity,long tail,and cold start issues,which seriously hinder the development of the recommendation system and bring adverse experiences and dissatisfaction to users.Therefore,this article analyzes and studies the long tail problem in recommendation systems.The long tail problem is due to the uneven distribution of daily data.Some popular items have a large amount of interactive data,while most long tail items have very little interactive data.Traditional algorithms such as collaborative filtering methods often overestimate popular items and ignore a large number of long tail items,resulting in the emergence of the long tail problem.In recent years,many scholars have devoted themselves to research on recommendation of long tailed items.These studies have alleviated the long tailed problem to some extent,and have improved their diversity and coverage to varying degrees.But there are few guarantees about the number of long-tail items in the recommended list.At the same time,they have not established a good relationship between hot items and long tailed items,making the correlation between hot items and long tailed items weak,The recommendation rate of popular items is still high.From the user perspective,it is not possible to better mine the user’s preference for long tailed items.This paper presents a long-tail item recommendation algorithm based on deep learning,which can effectively link popular items with long tailed items,thereby solving the problem of high recommendation rates for popular items in the recommendation list,while ensuring accuracy,improving coverage,reducing the number of high-frequency recommendations,and achieving better recommendation results.The main research contents of the paper include:(1)For the long tail recommendation task,the accuracy problem of the traditional recommendation model in the long tail recommendation is not high,this paper proposes a long tail item recommendation algorithm model,which is based on a deep composite model algorithm – Fusion Transformer Graph Convolutional Layer Generation Countermeasures Network(Graph Convolutional Transformer-Generative Adversarial Networks,GCT-GAN),using graph convolutional networks as a discriminator for generating countermeasures networks,At the same time,the Transformer mechanism is added to enable the model to better perform feature extraction and feature fusion,and improve the accuracy of the long-tail recommendation.(2)This paper analyzes the reasons for the long tail phenomenon in the recommendation system,and proposes a long tail recommendation framework in view of the problems that few studies on the long tail recommendation can guarantee the recommendation rate of the long tail items in the recommendation list,the high repetition rate of the items in the recommendation list and the frequent recommendation of popular items.This paper defines long-tail items and high-frequency recommended items in the recommendation list,and sets the long-tail ratio in the recommendation list to establish the connection between long-tail items and popular items,so as to reduce the recommendation frequency of high-frequency recommended items in the recommendation list and improve the recommendation rate of long-tail items,effectively improving the coverage rate of the recommendation list.(3)The recommended experiment with a long tail ratio of 0.2 and 0.35 shows that when the long tail ratio is 0.35,the accuracy rate has been partially improved compared to the previous one.From the user’s perspective,it can help users mine their preferred long tail items,thereby mining their preferences for long tail items,improving user satisfaction and user stickiness.On MovieLens 1M real data set,Top-N recommendation experiment was carried out in this paper,and the effectiveness of the long-tail item recommendation algorithm in this paper was demonstrated by comparing experimental results.This algorithm can mine users’ preference for long-tail items,which has important application value for improving business performance and increasing user stickiness.