Character string recognition method based on named entity model, electronic device, storage medium

Abstract

The invention provides a character string recognition method based on a named entity model, comprising: the input layer of the named entity model receives the character string input by the user, converts the character string into a word index array and outputs it to the word embedding layer, and the word embedding layer Each element in the word index array is converted into a multi-dimensional word vector and output to the bidirectional long short memory network layer. When the sigmoid activation function unit or tanh activation function unit of the bidirectional long-short memory network layer receives input data, it generates a sigmoid lookup request or a tanh lookup request, and calls a preset function interface, and uses the preset function interface for different lookup requests Using different table lookup methods to look up corresponding data in the same preset lookup table, and use the found data as the output result of the corresponding activation function unit. The bidirectional long-short memory network layer logically processes the output of the activation function unit and then outputs it to the fully connected layer, and the fully connected layer adds entity labels to the output result data. The scheme of the invention can effectively improve the data processing efficiency of the activation function.

Description

technical field [0001] The invention relates to the technical field of character recognition, in particular to a character string recognition method based on a named entity model, electronic equipment and a computer storage medium. Background technique [0002] In the field of automotive NLP (Natural Language Processing, Natural Language Processing), named entity recognition (Named Entity Recognition, NER) is a very basic task, which refers to identifying named referents from text and extracting relationships. The task is to pave the way. In a narrow sense, it is to identify three types of named entities: person names, place names, and organization names. In a broad sense, it can identify more named entities, such as singers and song names. Since the sigmoid and tanh functions of the two-way long-short memory network layer in the NER model are exponential functions, a floating-point unit or a dedicated hardware module is required for calculation. However, in automobiles, mo...

AI Technical Summary

Problems solved by technology

Since the sigmoid and tanh functions of the two-way long-short memory network layer in the NER model are exponential functions, a floating-point unit or a dedicated hardware module is required for calculation. However, in automobiles, most of the current on-board chips do not have floating-point units or dedicated hardware modules. Hardware acceleration module, and the performance of using general-purpose processor software to simulate exponential operations is very low, and it is impossible to achieve more efficient operation results

Images