Neural collaborative filtering model recommendation method based on lambda Mart

Abstract

The invention provides a neural collaborative filtering model recommendation method based on lambda Mart, and the method comprises the following steps: S1, inputting user information which comprises user basic information and movie comment information, and the movie comment information comprises scored movie information and unscored movie information; S2, enabling the embedding layer to map the user information into a user feature vector, and mapping the movie comment information into a movie feature vector; S3, inputting the user feature vector and the movie feature vector into a neural collaborative filtering model, extracting high-order feature information, and extracting sorting information at the same time; and S4, processing the high-order feature information and the sorting information to obtain a recommendation result, and outputting the recommendation result. According to the method, an LMNCF model is put forward, a neural collaborative filtering model is improved, implicit high-order feature information is extracted through nonlinear feature processing of a multi-layer perceptron, sorting information is extracted through a lambda Mart personalized sorting algorithm, and recommendation is more accurate.

Description

technical field [0001] The invention relates to the technical field of personalized recommendation, in particular to a neural collaborative filtering model recommendation method based on lambdaMart. Background technique [0002] With the development of science and technology, there are more and more online data, which makes users unable to obtain really useful information when faced with huge information overload, and the efficiency of using information is greatly reduced. Personalized recommendation is an effective way to solve information overload , the recommendation through this method can better grasp the user's viewing habits, and provide users with high-quality movie recommendations. [0003] At present, the methods used to realize movie recommendation include: neural network collaborative filtering recommendation algorithm based on inception structure, hybrid recommendation model based on deep learning, sparse data recommendation algorithm based on noise detection an...

AI Technical Summary

Problems solved by technology

The disadvantage of this technology is that the offline training of the recommendation system using the NCF-i model will get better recommendation results, but due to the complexity of the calculation using the neural network, when the scale of users and products is huge, it will consume a lot of resources. and time, so the model cannot be applied to real-time online recommendation

[0005] The hybrid recommendation model based on deep learning, for the traditional matrix decomposition algorithm, only uses the scoring information as the recommendation basis. When the scoring data is sparse, the implicit feedback cannot be accurately obtained, which affects the accuracy of the recommendation. Make full use of the auxiliary information for implicit features. The extraction of user information has become one of the research hotspots. A deep learning-based recommendation model HRS-DC is proposed, using deep neural networks and convolutional neural networks to extract the hidden feature vectors of users and items from auxiliary information, and then the feature vectors New scoring matrix derived from improved neural collaborative filtering

The disadvantage of this technology is that the number of network layers of NCF, the dimension of hidden factors and other parameters have not yet found the best value, but only found a relatively optimized value, and the accuracy of prediction needs to be improved. In terms of sparsity and cold start There is still room for improvement

The disadvantage of this technology is that the training process of restricted RBM is relatively complicated, and hyperparameters need to be fine-tuned, so it cannot be applied to dynamic big data recommendation applications, such as real-time data flow scenarios such as news recommendation and stock recommendation.

The disadvantage of this technology is that when using i-tem2vec to learn item vector representation, a small number of items are not in the training set, that is, cold start items. For items without labels, further processing is required to be applicable. There are a large number of cold items in real life. And the input sequence of RNN is adjacent on the time axis, without considering the influence of the time interval between the two items

Images