Information extraction method based on bi-directional recurrent neural network

Abstract

The invention relates to the field of natural language processing, in particular to an information extraction method based on a bi-directional recurrent neural network. The information extraction method applies the technology of the bi-directional recurrent neural network, the basic elements, which include characters, words, punctuations and the like, of a text are subjected to the automatic learning of characteristics, series modeling is carried out through the RNN (Recurrent Neural Network), and the defect that the characteristics need to be manually set in a traditional way is overcome. In addition, the bi-directional communication RNN is used to overcome the problem of information asymmetry in a prediction process of a unidirectional RNN, so that the classification judgment result of a natural language series to be identified depends on both preamble information and postamble information, and therefore, information extraction and judgment accuracy is higher. The method is especially suitable for entity name extraction in big data analysis, and has an important application value in the big data analysis.

Description

technical field [0001] The invention relates to the field of natural language processing, in particular to an information extraction method based on a bidirectional recursive neural network. Background technique [0002] With the rapid development of the Internet, a large amount of public web page data has been generated, which has also spurred various emerging industries based on big data technology, such as Internet medical care, Internet education, corporate or personal credit investigation, etc. The rise and prosperity of these Internet industries is inseparable from the analysis of a large amount of data information; however, most of the data obtained directly from web pages are unstructured. In order to use these data, data cleaning has become the most time-consuming and energy-consuming task for major companies. place. In data cleaning, the extraction of specific information, especially the extraction of named entities, is a common occurrence. For example, when doing...

AI Technical Summary

Problems solved by technology

Feature templates include first-order words or multi-order phrases with a specified window size context, word prefixes, suffixes, part-of-speech tags and other state features; the construction of feature templates is very time-consuming and labor-intensive, but the recognition results are extremely dependent on feature templates; Manually set feature templates are often only based on the characteristics of some samples, which is poor in versatility; and usually only local context information can be used, and the use of each feature template is independent of each other. The prediction cannot rely on longer historical state information, nor can it Use longer future information feedback to correct possible historical errors, the forecasting process is complicated, and the forecasting results are difficult to achieve global optimality

Images