Research On Intelligent Resource Recommendation Method Based On Deep Learning

Nowadays,people live in an era of information explosion.For information consumers,they need to take a lot of time and energy to filter the information they are interested in,and for information producers,they need to consider how to intelligently recommend the information of interest.To solve the problem,the recommendation technology came into being.According to the different recommendation algorithm,recommendation system can be divided into Content-based Recommendation,Collaborative Filtering Recommendation,Hybrid Recommendation and so on.Among them,the Collaborative Filtering Recommendation is the most widely used recommendation algorithm.However,due to the limitations of traditional recommendation technology,the recommendation system often has some problems such as data sparsity,low accuracy and so on.The rise of deep learning has provided new ideas for the field of recommendation system.Based on the traditional recommendation technology,deep learning will be introduced to better solve the problems existing in the recommendation system and further improve the accuracy.In the subject,based on the related research,experiment and analysis of recommendation system and deep learning,an intelligent resource recommendation method based on deep learning,Convolutional Long Short-Term Matrix Factorization model(CnnLstmMF)is proposed Method steps are as follows:(1)Traditional recommendation systems often use simple user behavior data such as rating data.Compared with the rating data,the review data do better reflect the user's real emotions and the specific reasons for the preference of an item,which has higher application value.Therefore,the subject not only uses the user's rating data on the item,but also introduced the auxiliary behavior data,that is,the user's review data on the item.(2)Due to the limitations of the traditional textual representations of Bag of Words model and the difficulty of processing the review texts,the subject introduces the Word Embedding text representation method to train user's review texts on the item,to generate review text word vector,and construct review text matrix which is used as input of deep neural network to deeply dig the deep meaning of review text.(3)Due to the success of Convolution Neural Network and Long Short-Term Memory Neural Network in the field of Natural Language Processing,the subject will combine the advantages of both to train the review text matrix,extract the item features reflected in the review text,generate the item latent vectors,and reconstruct the item latent feature matrix of Probabilistic Matrix Factorization model.(4)Finally,the subject uses a new Hybrid Recommendation method combining Probabilistic Matrix Factorization and Deep Learning to predict the user's rating of the item and intelligently recommend the user's favorite item.Through a large amount of data evaluation,the proposed model is obviously superior to other existing models,effectively solving the problem of data sparsity,further exploring the potential interests of users,improving the accuracy of the algorithm and improving the recommended performance.