Recommendation method based on self-attention mechanism

Abstract

The invention relates to the field of article recommendation, discloses a recommendation method based on a self-attention mechanism, and aims to improve the training efficiency of a recommendation model and the effect of personalized recommendation. The method comprises the following steps: firstly, collecting and preprocessing historical interaction information of a user to form a training sample set; designing a recommendation model, taking the training sample set as the input of the recommendation model, and adopting a square loss function as an optimization target to train the recommendation model; finally, calculating through the trained recommendation model to obtain the interaction probability of the user and to-be-recommended items, sorting the interaction probability according to the size, and generating a recommendation candidate set of the user.

Description

Technical field [0001] The present invention relates to the field of recommended items, particularly relates to a method based on self-recommended attention mechanisms. Background technique [0002] In the era of information explosion, the rapid growth of the number of goods, difficult for users to find what they are interested in a short time, how to use the user's history quickly and accurately generate personalized recommendations become a research hotspot. Recommendation algorithm to help users find content of interest, the current mainstream methods are: collaborative filtering recommendation algorithm based content recommendation algorithms, based on depth learning model recommendation algorithm. Among them, the recommendation algorithm depth learning model based on its ability to fit the generalization ability has become the most important method. [0003] In terms of recommendation algorithm based on deep learning model, Jin et al proposed a preference vector nerve percep...

AI Technical Summary

Problems solved by technology

[0004] The above research is mainly based on the recommendation based on the user’s historical information. The neural perception recommendation method proposed by Jin et al. does not consider the preference weight between the user and each item. All interactions use the same weight value, and the missing information may affect the model’s performance. learning result

At the same time, this method needs to additionally calculate the overall quality score of the store through user feedback, and the frequency of changes in the score and evaluation of the store will increase the training cost of the model

DeepCoNN uses two feature extraction modules. Each user needs to obtain the features of items and products from user reviews and product reviews through this module. The calculation cost is relatively large, and a large amount of text information learning may have a negative impact on the recommendation effect.

[0005] In fact, the position of the selected item in the user record can reflect the interest of the user at the corresponding time node. The above studies did not use the position information in the sequence, and ignored the degree of correlation between items in the historical information.

Images