Associated knowledge generation method, auxiliary annotation system and application

Abstract

The invention relates to an associated knowledge generation method, an auxiliary annotation system and application, and the method comprises the steps: for a standard question, generating N generated questions through an AMBERT model, and carrying out the screening and annotation, thereby completing the generation of associated knowledge of the current standard question; repeating until the associated knowledge of all the standard questions is generated; the AMBERT model sets a MASK token in the BERT as a mixed MASK token, and generation question prediction is carried out to obtain N generation questions; the method is applied to a rear-end generation unit of the associated knowledge auxiliary labeling system, associated knowledge generated by the rear-end generation unit is output through an output interface of a front-end operation unit, and labeling content is input from an input interface; the system is applied to the tax question-answering system. Easily understood corpora needed by the language model of the question and answer system are constructed, the difficulty of manual annotation work is solved, and the recommendation performance of the language model and the manual annotation efficiency are improved by screening out the data needing to be annotated of the model.

Description

technical field [0001] The invention relates to the technical field of electrical digital data processing, in particular to a method for generating associated knowledge, an auxiliary labeling system and applications. Background technique [0002] With the continuous development of science and technology, the demand for intelligent information consulting services in all walks of life is increasing, and it is becoming more common to realize intelligent consulting services through deep learning and related technologies of natural language processing. Under this premise, the core point of the service is that the intelligent consulting service can accurately recommend the answers to the user's consulting questions, and this puts forward higher requirements for the performance of the question answering system in the consulting service. [0003] In fact, language model recommendation services alone cannot achieve excellent performance in intelligent consulting services, and constru...

AI Technical Summary

Problems solved by technology

[0006] However, although BERT works very well, it is not clear from the model mechanism why its effect is good, which limits the further assumptions of researchers and the improvement of the architecture driven by this.

Unlike CNNs or other traditional model architectures, Transformers are not cognitively motivated, and the size of these models limits the ability to experiment with pre-training and conduct ablation studies. In the past year, there has been a large amount of research trying to understand BERT. The reason for its excellent performance has not been realized. While BERT has greatly improved the performance of various natural language understanding tasks, its bidirectional nature makes it difficult to apply to Natural Language Generation (Natural Language Generation) tasks, which limits the use of BERT in The NLG field continues to exert its performance superiority

Images