Intention recognition method and system based on keyword feature embedded language model

Abstract

The invention belongs to the technical field of natural language processing, and particularly relates to an intention recognition method based on a keyword feature embedded language model, which comprises the following steps of: performing word segmentation on extracted language information of an effective text by adopting a forward and backward maximum word segmentation algorithm to obtain different types of word segmentation results; for the obtained different types of word segmentation results, obtaining a keyword list related to candidate intentions corresponding to the different types of word segmentation results; removing universal high-frequency words and field-independent words in the keyword list related to the candidate intention corresponding to the word segmentation result of each category, obtaining a final keyword list corresponding to the word segmentation result of each category, and further obtaining different keyword feature vectors; embedding each obtained keyword feature vector into a pre-trained language model, and obtaining voice information of an effective text with keyword features; and encoding and classifying the voice information to obtain an intention recognition result of the language information of the effective text.

Description

technical field [0001] The invention belongs to the technical field of natural language processing and long text intent recognition, and in particular relates to an intent recognition method and system based on a keyword feature embedded language model. Background technique [0002] Intent recognition technology is an important technology to identify the speaker's intention hidden in long text, and it is also an important research content in the field of natural language processing. [0003] Traditional intent recognition technology uses keyword matching and statistical information such as word frequency, TFIDF (term frequency–inverse document frequency, word frequency inverse text frequency index) and traditional machine learning models such as support vector machines and mixed Gaussian models. These methods only make use of the underlying language statistical data, but the deep semantic information cannot be involved, and they are very dependent on the sample quality, and ...

AI Technical Summary

Problems solved by technology

[0003] Traditional intent recognition technology uses keyword matching and statistical information such as word frequency, TFIDF (term frequency–inverse document frequency, word frequency inverse text frequency index) and traditional machine learning models such as support vector machines and mixed Gaussian models. These methods only make use of the potential language statistics, but the deep semantic information cannot be involved. They are very dependent on the sample quality and the performance is not ideal.

[0004] Since the development of deep learning technology, advanced neural networks such as convolutional neural networks, recurrent neural networks, and transformers have been used to encode deeper semantic information, and have achieved remarkable results in various fields of natural language processing. However, these The methods are too dependent on the training data, only focus on specific tasks, and will be affected by the data set, learn other features of the data set and ignore the core semantic information of the task, once the data set changes, the performance will be greatly reduced

Images