A Multi-Label Text Categorization Algorithm Incorporating Label-Guided Attention Mechanisms

This paper proposes a multi-label text classification algorithm based on causal relationships to address the current challenge of accurately capturing label correlations in multi-label text classification tasks. The algorithm comprises a basic prediction model, a label confusion model, and a label-guided attention mechanism. The basic prediction model extracts features from the text and generates a preliminary label distribution. The label confusion model generates simulated label distributions by calculating the causal correlations between text vector representations and labels at each level, thereby guiding the initial prediction results and improving prediction accuracy. To further reduce noise introduction, a label-guided attention mechanism is designed to enable the model to focus better on features related to the target labels. Experimental results demonstrate that the algorithm exhibits superior performance on established classification benchmark datasets, validating the effectiveness of the causal modeling approach in multi-label text classification tasks.