Deep Learning Recommendation Algorithm Of Fusion Ratings

With the rapid development of distributed computing,artificial intelligence,cloud computing and other technologies,due to the application of various network services,the scale of data and the volume of information have shown an exponential growth,the recommendation system and its related technologies are becoming important tools to deal with "information overload".The recommendation system needs to record the explicit or implicit behavior in the user's interaction behavior,discover the user's preference features,and then generates a list of items that match the user's preferences.Since the traditional recommendation system mainly extracts the characteristics of users or items through manual or shallow models,it is difficult to obtain deep expressions of users or items.In addition,when the traditional recommendation system analyzes the user's preference characteristics,it ignores that the user's preference will change dynamically with time.In order to deal with above problems,the research content of this paper is as follows:First of all,when using the recurrent neural network to predict the items that the user will interact with at the next moment,this article uses the Word2 Vec model to characterize the items in the input sequence,and proposes a W-LSTM recommendation model.Due to the contextual relationship between the user's historical behaviors,the Word2 Vec framework can be used to model user behaviors that are related to each other,and then it generate low-dimensional,dense item word vectors,and obtain hidden semantic relationships between user behaviors to make the recurrent neural network has better performance in predicting the item that the user clicks at the next moment.Secondly,in the movie recommendation task,in order to make the recurrent neural network pay attention to the user's preferred genre features and some important behaviors,this paper builds a GE-RA-LSTM model based on the rating attention mechanism and fusing genre features.In addition,in order to pay attention to the genre information preferred by the user,the model concatenates the genre features of the movie with the movie word vector in the input part.In order to pay attention to the user's partof the important behavior,the model combines the RF-IIF and the attention mechanism to make the model focus on high-score movies and unpopular movies in the user's historical sequence to mine the user's unique preferences.The GE-RA-LSTM model not only realizes the mining of users' dynamic preferences,but also recommends more interesting movies for users,while increasing the recommendation of unpopular films.Finally,this paper validates the above model in the Movie Lens-10 M dataset.The experimental results show that the W-LSTM model using Word2 Vec for the movie representation of the user's historical sequence has a certain improvement in accuracy and coverage and other indicators than the model that randomly initializes the movie word vector,and greatly reduce the training time of the model.Introducing a certain number of genres in the input part of the model can further improve the accuracy of the recommendation.The GE-RA-LSTM model using the rating attention mechanism has a greater improvement in the coverage index than the recommendation model using the attention mechanism.Compared with the traditional recommendation algorithm,the W-LSTM and GE-RA-LSTM models proposed in this paper has a better recommendation effect.In summary,Word2 Vec can better characterize items by extracting contextual connections of user behavior.If the model considers the influence of user preference genres,the accuracy of the recommendation results can be improved.In addition,using the rating attention mechanism in the model can make the model focus on some important behaviors of the user,mine the user's unique preferences,and improve the recommendation performance.