Enterprise abbreviation generation and model training method and device

Abstract

The invention provides an enterprise abbreviation generation method and device and an enterprise abbreviation model training method and device. Wherein the structure of the enterprise abbreviation generation model sequentially comprises an embedded layer, an encoder and a decoder, and the training method comprises the following steps: generating an associated word library of an enterprise full name based on an enterprise full name sample set and an enterprise abbreviation sample set, and generating a word library based on the enterprise full name library and the associated word library; carrying out word segmentation on the enterprise full name sample set according to characters to obtain a word vector of a first dimension, and inputting the word vector into the embedding layer to carry out dimension reduction processing to obtain a word vector of a second dimension; inputting the word vector of the second dimension into an encoder for encoding to generate an intermediate semantic vector; inputting the intermediate semantic vector into the decoder based on the word stock for decoding to generate an enterprise abbreviation candidate set; wherein in the decoding processing process, the prediction character generated in each step is adjusted to the word stock according to the similarity. According to the embodiment of the invention, the enterprise abbreviation generation accuracycan be improved.

Description

technical field [0001] This application relates to the field of deep learning technology, in particular to a method and device for generating corporate abbreviations and training their models. Background technique [0002] Enterprise abbreviations are used more frequently than full names in natural language. In the general named entity recognition (NamedEntity Recognition) task, organization name recognition has become the most difficult type of entity because of the widespread use of business abbreviations. In addition, in practical applications, such as natural language retrieval, question answering, knowledge map construction and other fields, identifying the abbreviation is not the ultimate goal. It is also necessary to standardize the abbreviation of the enterprise before the follow-up work can be carried out. [0003] In the prior art, rules and traditional conditional random field (CRF) algorithms are mainly used to generate enterprise abbreviations. The CRF model i...

AI Technical Summary

Problems solved by technology

[0004] The applicant found in the research that the assumptions in the prior art easily lead to the loss of abbreviation information, resulting in a loss of recall rate in the application process

For example, the corresponding abbreviation of "Chongqing Three Gorges Paint Co., Ltd." is "Yusanxia", and the corresponding abbreviation of "Beijing Shenzhou Automobile Leasing Co., Ltd." is "Shenzhou Rental Car".

Moreover, there are many steps in the prior art, and each step has a loss of accuracy, resulting in a decrease in overall accuracy and affecting practical applications.

Images