Text abstract generating method and system

Abstract

The invention relates to a text abstract generating method and system. The method comprises the steps of obtaining text information, and segmenting the text information into a plurality of words; inputting the segmented words into a word embedding model, thereby obtaining word vectors of the words; obtaining all characters which form each word, and sequentially inputting all the words which form the same word into a two-way recurrent neural network model, thereby obtaining joint vectors of the words; inputting the word vectors of the words and the joint vectors corresponding to the word vectors into a nonlinear model, thereby obtaining fusion features of the words; combining the fusion features of the words into a new input text, and inputting the new input text into a convolutional neuralnetwork, thereby obtaining high-level features which represent the text; and inputting the high-level features which represent the text into an encoding-decoding model, thereby obtaining an abstract.According to the method and the system, key information of the original text can be enhanced, and non-key information can be attenuated, so that the generated text abstract can better represent important information of the text.

Description

technical field [0001] The invention relates to the field of text data processing, in particular to a method and system for generating text summaries. Background technique [0002] With the explosive development of data, especially the rapid increase of text data, people have been unable to browse and understand all the texts of interest in time, but the omission of some important text data will cause a lot of organizational and application losses. Therefore, text summarization As the information that summarizes the important data of the text, it has become the focus of people's attention, and how to automatically generate summaries based on the text data has also become a hot research topic. [0003] At present, the existing methods for automatically generating text summaries mainly use the encoding-decoding model in machine learning. Specifically, the model first uses Recurrent Neural Networks (RNN) as an encoder to encode information from the original text. Compression a...

AI Technical Summary

Problems solved by technology

[0004] However, when the text summary is generated by the encoding-decoding model, the cyclic neural network directly compresses and extracts the original text information. However, due to the limited feature extraction of the original text by the cyclic neural network, many high-level features such as representing the key points of the original text The characteristics of the information cannot be identified and extracted, and the non-key information cannot be effectively identified and excluded, so that the generated text summary contains more impurity information, so that the final generated summary cannot well represent the important information of the text

Images