a recommended method

Abstract

A recommendation method, comprising the following steps: S100: using a word embedding unit to convert text in natural language about users and items into numerical training data; S200: using an attention mechanism to add users and items to the training data S300: Use the convolutional neural network model based on the attention mechanism to extract the local features and core features of the training data, and finally get the hidden features that can express the global features; S400: Use the factorization machine to analyze the above hidden features Perform analysis to obtain the association between the user and the item, complete the user's rating prediction for the item based on the association, and finally complete the recommendation of the item to the user. Compared with existing methods, this method improves the precision and accuracy of recommendation, and improves the data utilization rate.

Description

technical field [0001] The disclosure belongs to the technical field of artificial neural network and personalized recommendation, and particularly relates to a recommendation method. Background technique [0002] With the rapid development of technologies such as cloud computing, big data, and the Internet of Things, a large number of application platforms such as shopping, education, and entertainment have emerged in the Internet and information industries, making the scale of multi-source heterogeneous data also grow rapidly. It is estimated that by 2020 The total amount of global data will reach 35.2ZB. These big data contain a wealth of value, which can guide people to change their behavioral decision-making mode from empiricism-based to data-driven. However, while enjoying the convenience brought by big data, it is difficult for people to extract valuable information from big data, which leads to the problem of "information overload". Therefore, how to mine effective...

AI Technical Summary

Problems solved by technology

The aggregation operation of the convolutional neural network will lose the position information of some words, and cannot consider the influence of historical words with high weight, which reduces the accuracy of feature extraction in natural language processing; although the recurrent neural network can consider dynamic information and The feature extraction of natural language has a good effect, but compared with CNN, its feature expression effect on static data is poor and the operation speed is too slow

[0004] To sum up, recommendation in a big data environment still faces three challenges: one is how to improve the accuracy of feature extraction from heterogeneous and multi-source data; the other is how to replace the traditional feature extraction method with manual extraction Turn to automatic extraction; the third is how to integrate deep learning and traditional recommendation methods into a highly collaborative hybrid model

Images