UBA-DA: A Method for Automatic Data Annotation Based on User Behavior Analysis

Data annotation serves as a foundational component in developing algorithmic models for machine learning and artificial intelligence applications. Current methodologies face a dual challenge: Implementing data annotation during application while maintaining cost efficiency, which is crucial for facilitating model evolution in applications. This paper proposes UBA-DA, a method for automatic data annotation based on user behavior analysis. During the model deployment and application phase, we model users' typical human-computer interaction behaviors (including browsing, questioning, modifying, confirming, etc.) as a sample generation automaton. This sample generation automaton produces annotated samples during user interactions with the algorithmic model. These samples provide essential training data support for rapid domain adaptation and iterative refinement of the model in specialized business scenarios. Experiments show that compared with existing data annotation methods, the method proposed in this paper can effectively generate high-quality samples.