Neural network model based on attention mechanism

Abstract

The invention discloses a neural network model based on an attention mechanism. The neural network model comprises position coding and data coding, a multi-head self-attention mechanism (Mti-Head-Self Attention), residual connection and layer normalization (Addamp; according to the method, five modules (Norm), a feed-forward neural network (feed forward) and a convolutional neural network (CNN) are used, a residual connection and layer normalization module is used twice, other modules are respectively used once, language and image data information is fused by using a deep learning technology, cross-media data is used for public opinion analysis, and two types of data can share model parameters, so that cost resources are saved, and the method is suitable for popularization and application. The meaning is thoroughly understood, and public opinion analysis can be effectively carried out.

Description

Technical field [0001] The invention relates to the field of natural language processing and computer vision, language processing, image processing and deep learning technologies, in particular to a neural network model based on attention mechanism. technical background [0002] At present, mobile users need to generate or receive more information in a shorter time to participate in online social interaction, and users can't tolerate the anxiety caused by reading large texts or watching lengthy videos. Therefore, the cross-media public opinion data on various media platforms show important characteristics such as small volume, miscellaneous content, large scale, fast production speed, etc. For example, the length of short video is usually less than five minutes, and the length of Weibo is mostly within 100 words, but the research work of public opinion analysis is generally single text data or image data. [0003] This project intends to study a cross-media public opinion analysi...

AI Technical Summary

Problems solved by technology

[0002] Currently, mobile users need to generate or receive more information in a shorter period of time to participate in online social interactions, and users cannot tolerate the anxiety caused by reading large texts or watching lengthy videos

Therefore, cross-media public opinion data on various media platforms present important characteristics such as small size, complex content, large scale, and fast generation speed. For example, the duration of short videos is usually less than five minutes, and the length of Weibo is mostly concentrated in 100 characters. However, the research work of public opinion analysis is generally a single text data or image data

Images