Language processing method and system based on statement discrimination and recognition and reinforcement learning action design

Abstract

The invention discloses a language processing method and system based on statement discrimination and recognition and reinforcement learning action design. The language processing method comprises the steps of core structure network construction, statement structured optimization classification processing and optimal text structure sampling judgment. The invention provides a statement reinforcement learning action design module for identifying important words and a design module for identifying an important structure to construct structured representation. After the attention mechanism is introduced, the attention degree of the phrases can be better covered, the attention mechanism can assist reinforcement learning in faster fitting, and the learning efficiency can be improved.

Description

technical field [0001] The present application relates to the technical field of language processing, in particular, to a language processing method and system based on sentence discrimination and recognition and reinforcement learning action design. Background technique [0002] The main research focuses on linguistics research oriented to natural language processing, mainly including the constitutional and morphological characteristics of Korean characters, syntactic structure characteristics and construction rules, and semantic expression forms. involve. [0003] In recent years, corpus linguistics of language texts has developed rapidly, and raw corpora and tagged corpora have been constructed based on relevant theoretical knowledge, and morphological knowledge, part-of-speech knowledge, syntactic knowledge, and semantic knowledge have been obtained through corpora. [0004] Domestic language and text phonetic orthography and related language and text language norms for...

AI Technical Summary

Problems solved by technology

[0005] The research on language text and natural language processing is currently stuck in text classification using bag-of-words modules or traditional neural network sequence modules. Simple superposition of algorithms, or optimization only from the perspective of word embedding, cannot improve the quality of language and text resources. Effective management, mining and utilization cannot meet the needs of language and text informatization and intelligent processing

[0006] In terms of text structuring, in the field of minority language information processing, there is currently no research on representational learning of language texts. The present invention provides a sentence-based identification and reinforcement learning action to solve language text data processing The designed language processing method and new processing method of the system, this method is suitable for most subsequent downstream tasks, and has further in-depth research and application value in the fields of text classification, sentiment analysis, emotion conversion, question answering system, recommendation system, etc.

[0007] In addition, domestic research on language, text, and natural language processing is still in the development stage. Compared with South Korea and even language texts, related research is still lagging behind. There is no unified planning and layout for the construction of language, text, and language resources, and there is no related series of standards. It makes it difficult to integrate the resources of various companies, which limits the further deepening of related research

Images