Deep Learning Based Personalized Recommendation Algorithm Research

With the development of the Internet and electronic devices,recommendation systems have flooded people's lives.But with the advent of the era of big data,the traditional recommended learning method is not as good as before when processing large amounts of data.For example,the collaborative filtering model and its extended model of the widely used traditional recommendation system are shallow models,and most of them are recommended based on linear inner product.In the era of big data,the relationship between a large number of users and commodities is becoming more and more complicated,and the shallow linear model has not been well expressed.In some feature extraction,traditional recommendation systems require people to manually extract features,which is time-consuming and labor-intensive,and the results obtained are not necessarily good.In some weight distributions,the traditional recommendation system,after complex calculations,does not necessarily give satisfactory results.Therefore,a recommendation system based on deep learning came into being.It is better than traditional methods in the expression of some complex relationships,the extraction of features,and the intelligent allocation of weights.It is a trend to use deep learning for recommendation systems.This thesis thinks about the recommendation algorithm from the perspective of the association between the items,user preferences and weight distribution,and conducts related research on the recommendation algorithm based on deep learning.Based on the innovative algorithm proposed in this thesis,a linear regression algorithm is used to design a recommended prototype subsystem that combines multiple algorithms.The main results of this article are as follows:1.This thesis proposes an item-based convolutional collaborative recommendation algorithm.Consider the recommendation problem from the perspective of the association between items,focusing on mining potential information between items that each user has interacted with and items that have not.The historical information between each user's items is regarded as a graph,and the convolutional neural network is used to mine the nonlinear relationship characteristics of historical interactive / non-interactive items.The historical information between each user's items is regarded as a low-dimensional latent factor,and the target item is regarded as another low-dimensional latent factor,and their product(inner product)is used as the characteristics of the target item.Proved their superiority in personalized recommendation tasks on two real data sets.2.This thesis proposes a convolutional attention model based on long-and shortterm preferences.Consider the recommendation from the perspective of user preference and weight allocation.In terms of user preferences,the user preferences are subdivided,taking into account the user's long-term and short-term preferences.For the short-term preference of the user,the items that the user likes are sorted according to time,the sequence is regarded as a form of a linked list,and the sequence is expressed as an image.The convolution kernel of the convolutional neural network slides on this sequence to obtain the user's preference information in the short term.For the long-term preferences of users,the LFM latent factor model is used to extract user habits.In terms of weight distribution,the spliced long-term and short-term preferences will be used to assign weights intelligently using the attention mechanism,and finally recommend to users.Comparing with other models on two real data sets,it proves the effectiveness and superiority of the method.3.This thesis designs a prototype subsystem for movie recommendation that combines multiple algorithms.For old users,the system uses a linear regression algorithm to fuse the two deep learning-based recommendation algorithms proposed in this thesis,which complements the shortcomings of the innovative algorithm in this thesis due to different perspectives.For new users,the recommendation algorithm POP based on popularity is used to solve the cold start problem.