Newly-developing hot topic detection system based on attention mechanism

Abstract

The invention relates to a newly-developing hot topic detection system based on an attention mechanism. The newly-developing hot topic detection system comprises a data preprocessing module, a hierarchical sequence model, a word sequence coding layer, a word level attention layer, a sentence level coding layer, a sentence level attention layer and a topic prediction module. By use of the newly-developing hot topic detection system based on the attention mechanism, on the basis of a bidirectional recurrent neural network, two layers of attention mechanisms are added to reinforce the vector representation of a topic, a hierarchical recurrent neural network model based on the attention mechanism is put forward, each dimension of data in social media can be used as characteristics to train high-quality topic vector characteristics, so that the newly-developing hot topic is detected, and the detection ability of the newly-developing hot topic is improved.

Description

technical field [0001] The invention relates to the field of natural language processing, in particular to an emerging hot topic detection system based on an attention mechanism. Background technique [0002] At present, there are some emerging hot topic detection methods that are biased towards topic content features (static features). The growth rate, etc., are used as real features, and then use a judgment function (such as a classification algorithm) to determine whether it is an emerging hot topic. [0003] At present, there are still some biases that use the propagation characteristics of topics to detect emerging hot topics. The basic idea is to use related specific data structures (such as: trees, graphs, particle swarms, neural networks, etc.) to calculate or train topics. Features, here the features tend to be disseminated, that is, there is a connection between the data, not static. Then use the classification algorithm to find out whether the topic is an emergi...

AI Technical Summary

Problems solved by technology

[0004] However, although these method models have achieved corresponding results to a certain extent, they have also promoted the development of topic detection tasks; but there are also shortcomings. Methods based on static characteristics of topic content are certainly accurate in predicting emerging hot topics. rate, but it lacks the contextual semantic analysis of the topic text, so the tracking effect of the topic is poor

Those based on propagation features (dynamic features) also take into account the contextual semantic information of the text in the topic, and there is a certain delay in the prediction time of emerging hot topics, so its accuracy is not enough, but it has better performance in topic tracking

Images