Query intention recognition method based on bert + bolstm + crf and fused with sentence pattern analysis

Abstract

The invention discloses a query intention recognition method based on bert + bolstm + crf and fused with sentence pattern analysis. The method comprises the steps that a data set is marked and constructed, domain intentions are defined, intention classification and intention word grooves are clarified, a domain term set corresponding to the word grooves is constructed, a question set is collected, and question intention classification and word groove information are marked. According to the marked question and the corresponding intention information, a sentence pattern for matching verification in an intention recognition model reasoning stage is generated; and a query intention recognition model is constructed based on bert+bilstm+crf. According to the method, slot dictionary matching, sentence pattern analysis, slot entity recognition based on bert+bilstm+crf and intention classification based on bert are fused, the problem that the model is wrongly recognized when the model training data and the model prediction data are inconsistent in distribution is well solved, and the accuracy of recognizing the user input port linguistic content intention by the model is improved.

Description

Technical field [0001] The present invention relates to natural language processing, artificial intelligence technology, and more particularly to a query intent recognition method based on Bert + BilSTM + CRF and fusion sentence pattern analysis. Background technique [0002] The existing so-called intention is actually the natural language understanding (NLU), is one of the important tasks in natural language. Natural language understanding is to hope that the machine is like people, with the normal human language understanding, through the intention of accurate identification, clarify the user's demand, thereby providing to the user accurate service. [0003] With the development of deep learning, the neural network model based on deep learning algorithms such as CNN, RNN, LSTM, and attention mechanism is widely used in text classification, intended to identify the field, and continuously obtain SOTA effects. The intent identification model of deep learning is high, but the ide...

AI Technical Summary

Problems solved by technology

The intent recognition model based on deep learning has a high intent recognition rate, but the recognition results are not interpretable, and the model recognition rate is heavily dependent on the quality of the training data

Images