Similarity calculation-based large-model zero-sample text classification method and device

• Main Authors: LIU SHENGPING, LIANG JIA'EN, YANG MENGSHI
• Format: Patent
• Language: Chinese; English
• Published: 05.04.2024
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a large-model zero-sample text classification method and device based on similarity calculation. The large-model zero-sample text classification method based on similarity calculation comprises the following steps: constructing instruction data of an input model; representing or generating a to-be-classified text tag token by using an MLM model or a generation model; and calculating the cosine similarity between the to-be-classified text and the positive example text and the cosine similarity between the to-be-classified text and the negative example text tag token, if the to-be-classified text is more similar to the positive example text, the to-be-classified text belongs to positive classification, otherwise, the to-be-classified text does not belong to the positive classification, and the technical problem that in the prior art, zero sample text classification based on similarity calculation is low in accuracy is solved. 本发明公开了一种基于相似计算的大模型零样本文本分类方法及装置。其中，该基于相似计算的大模型零样本文本分类方法包括：构造输入模