Semantic fuzzy matching method

Abstract

The embodiment of the invention provides a semantic fuzzy matching method. The method comprises the following steps of: extracting characteristics of the text identified by voice to obtain the characteristic data; carrying out named entity identification on the characteristic data by a conditional random field (CRF) to find the key semantic categories of sentences; and accurately matching the key semantic categories, performing fuzzy matching when the accurate match is failed, calculating the similarity of the key semantic categories and the key words in the dictionary, selecting the key words with largest similarity to replace the key semantic categories, and marking the categories. By the method of the embodiment, the CRF is used for marking the sequence, the key semantic categories in the inquire statement are initially marked and located; the fuzzy matching range is shortened; the similarity is calculated according to the domain dictionary; the dictionary entries with the largest similarity are used for replacing the wrong key semantic categories in the user query; the calculation amount is reduced; and the identifying speed is improved.

Description

technical field [0001] The present application relates to the field of speech recognition, in particular, to a semantic fuzzy matching method. Background technique [0002] In the human-computer interaction system, the user makes an inquiry request through spoken language, and the system provides information services. A typical human-computer interaction system includes four components: automatic speech recognition, spoken language understanding, dialogue management and speech synthesis. The spoken language comprehension part is to convert the query sentence after speech recognition into the corresponding semantic representation. However, spoken language comprehension often encounters such problems, that is, the user's query sentences have pronunciation variations, recognition errors, and incomplete key semantic concepts brought about by speech recognition. How to obtain correct sentences while obtaining some key information To understand the results, this requires fuzzy m...

AI Technical Summary

Problems solved by technology

The traditional fuzzy matching algorithm is mainly to find the starting position of the substring matching the pattern string in the given text string, and most of them use the edit distance as the similarity function. In this method, each Chinese character in the user query sentence must participate in the calculation, if the sentence is longer, the calculation speed will be greatly reduced

Images