Integration of contextual tags with named entity recognition models

• Main Authors: VU, DUY, BISHNOI, VISHNU, PHAM TUAN Q, GADDE SIDDHARTH PAVAN KUMAR, JOHNSON MARTIN E, HUANG CHRISTOPHER D V, DUONG THANH LONG
• Format: Patent
• Language: Chinese; English
• Published: 08.09.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

Techniques are provided for using contextual tags in a Named Entity Recognition (NER) model. In one particular aspect, a method is provided that includes receiving an utterance; generating an embedding of a word for the utterance; generating a regular expression for the utterance and a place name dictionary feature vector; generating a context tag distribution feature vector for the utterance; the embedding and regular expression, the place name dictionary feature vector and the context label distribution feature vector are connected in series or interpolated to generate a set of feature vectors; generating an encoded form of the utterance based on the set of feature vectors; generating a logarithmic probability based on the encoding form of the utterance; and identifying one or more constraints for the utterance. 提供了在命名实体识别(NER)模型中使用上下文标签的技术。在一个特定方面，提供了一种方法，该方法包括：接收话语；生成针对话语的词的嵌入；生成针对话语的正则表达式和地名词典特征向量；生成针对话语的上下文标签分布特征向量；将嵌入与正则表达式和地名词典特征向量以及上下文标签分布特征向量进行串连或插值以生成一组特征向量；基于该组特征向量生成话语的编码形式；基于话语的编码形式生成对数概率；以及识别针对话