Accurate Classification for Government Data: A Tree-of-Thoughts-Driven Few-Shot Learning Approach

Mengxiang Zhu, Yunchuan Guo, Duo Zhang, Ziyan Zhou, Haoyang Yu, Lingcui Zhang

Abstract: Government data is essential for social and economic development, underpinning policy-making, public services, and data-driven innovation. However, the heterogeneity of government datasets significantly increases the risk of misclassification. Currently, obtaining a robust and accurate classification model for government data is challenging due to label limitation and structural complexity, resulting in poor classification performance. To address this challenge, this paper utilizes a tree-of-thoughts-driven few-shot learning approach to propose an accurate classification for government data without requiring extensive feature engineering. In our approach, we employ a tree-of-thoughts-based structure to dynamically adjust the classification process. To enhance classification accuracy, we develop a label expansion method that further extends and clarifies the original classification terms. Additionally, we design a dynamic tree generator based on semantic clustering and association to achieve accurate classification. Experiments demonstrate that our approach achieves a precision of 85.15% in government data classification, surpassing existing methods in comparative studies.

External IDs:dblp:conf/icic/ZhuGZZYZ25