Table semantic query rough arrangement method based on hybrid architecture

Abstract

The invention relates to the technical field of man-machine conversation and intelligent semantic search, in particular to a table semantic query rough arrangement method based on a hybrid architecture, which comprises the following steps of: encoding a context model, encoding a table name and a column name, constructing a unique identification symbol for the table name and the column name of a database, and establishing a mapping relationship; splicing identifiers, table names and column name meanings to form contexts, the contexts serve as priori knowledge of query statements, dictionary recognition being efficient, and meeting recognition of professional terms in the professional limited field; generating the context vector offline in advance, omitting the coding calculation process of a pre-training model, the online response time being shortened, meanwhile, meeting the generalization requirement, and common spoken words and similar words can be recognized.

Description

technical field [0001] The invention relates to the technical fields of man-machine dialogue and intelligent semantic search, and specifically relates to a method for roughly sorting table semantic queries based on a hybrid architecture. Background technique [0002] At present, the methods based on rough sorting in semantic analysis mainly include the inverted index method based on dictionaries or word segmentation, which is direct and fast, but lacks generalization ability for synonyms or synonymous phrases. However, directly using the training model (such as bert and other pre-training models) for real-time calculation, the online cost is relatively high, and cannot meet the high-concurrency online real-time calculation requirements of large-scale data. Therefore, this paper proposes a hybrid architecture coarse sorting method based on caching technology, which screens the candidate set from the full data set and then passes it to the fine sorting model to improve the res...

AI Technical Summary

Problems solved by technology

[0003] The present invention is used to solve the problem of rough sorting of entities of query statements in complex context environments in the financial field, and provides a method for rough sorting of table semantic queries based on a hybrid architecture, which is characterized in that a rough sorting model is constructed, and the method is specifically as follows:

Images