Multi-task active learning method and system for sentiment classification and regression simultaneously

Abstract

The invention discloses a multi-task active learning method and system for emotion classification and regression simultaneously, belonging to the field of emotion computing. The present invention combines the active learning classification method on the EC task, and the value measurement of the unlabeled sample by the active learning regression method on the ER task, and obtains the total value measurement of the unlabeled sample on multiple tasks by active learning, and simultaneously mines the category type Emotional and dimensional emotional information, so that you only need to select as few samples as possible for labeling, and you can simultaneously train an EC model with good performance and an ER model on single or multiple dimensions. Experimental verification, in the same number of queries Under the circumstances, the method proposed in the present invention has better performance than the EC model trained by the single-task active learning method and the multi-dimensional ER model, and greatly reduces the labeling cost.

Description

technical field [0001] The invention belongs to the field of emotion computing, and more specifically relates to a multi-task active learning method and system for emotion classification and regression simultaneously. Background technique [0002] Affective computing enables machines to recognize, understand, express and adapt to human emotions, and is the core and foundation of human-computer interaction. Emotional recognition is an important step in emotional computing, and the emotional state of a person can be obtained by analyzing and processing collected physiological signals or other non-physiological signals. There are two ways to express emotion: 1) Categorical emotion (discrete), which can express emotion simply and intuitively as several independent emotion categories, such as the six basic emotions (happy, sad, Surprise, fear, anger, disgust); 2) Dimensional emotion (continuous), it is believed that emotion has basic dimensions, and each dimension is a measure o...

AI Technical Summary

Problems solved by technology

However, the current research only focuses on the active learning on the EC model or the ER model, which needs to select and label samples separately, and the training cost is relatively high

Images