Intelligent text automatic generation system based on deep learning and implementation method thereof

Abstract

The invention provides an intelligent text automatic generation system based on deep learning and an implementation method thereof, and the implementation method mainly comprises the following operation steps: step A, data preprocessing: preprocessing a data file submitted by a user, and converting the data file into a data format which can be recognized and processed by the intelligent text automatic generation system; step B, field-content selection coding: coding the preprocessed data, and calculating staged output of a sequence by using a pre-designed neural network model; and step C, descriptive decoding: performing new calculation and prediction on the staged output by using the neural network model again to obtain output probabilities of words in all vocabulary lists, and selectinga text with the maximum probability as the output.

Description

technical field [0001] The invention relates to an intelligent text automatic generation system based on deep learning and an implementation method thereof, belonging to the fields of natural language processing and data mining. Background technique [0002] Intelligent automatic text generation is a technology that uses natural language processing technology in the field of artificial intelligence to realize automatic text analysis, content summary and text generation. Accurate and comprehensive acquisition of key information has important practical significance in both industry and commerce. However, the mainstream methods currently used in the field of text generation are all based on the neural network language model. Due to the uninterpretable and black-box characteristics of the neural network language model, the generated text begins to have reference errors, such as: inconsistencies between sentences and source data Issues such as lack of fidelity. [0003] How to ...

AI Technical Summary

Problems solved by technology

Most of the methods currently used are based on the sequence-to-sequence neural network architecture, and the dependencies between the generated text and the source data are aligned through the encoder-decoder structure and the attention mechanism. However, the attention mechanism still has some defects. When generating long text, the alignment effect of the attention vector will gradually weaken, and the complex structural information in the data cannot be extracted well

Moreover, the neural network language model models the text generation problem as a statistical probability problem. When outputting the text, the word with the highest posterior probability is found as the target word, but for the rare words that have not appeared in the training data set, they are also called unregistered words. doesn't generate well

[0004] For the problem of rare words, most of the current processing methods use the copy network to copy the words that the model cannot generate from the source sequence to the output text, but the additionally trained neural network often cannot share parameters with the model, thus increasing the training model. Difficulty and resource consumption

Images