Home > AI Solutions > Artificial Intelligence > White Papers > Rethinking Hierarchical Text Classification: Insights from Multi-Agent Experiments with Small Language Models > Experiment 2
Overall, the T1 agent demonstrated substantial success with the new design, especially in extracting symptoms and assigning labels per symptom, which facilitated the identification of multiple symptoms that were mentioned in the text. While this model was a significant improvement on the previous ICC solution, it fell short in its ability to propose new labels consistently. In instances such as display-related issues, despite instructions to consider both [display (internal)] and [display (external)] labels when the type was unclear, the model showed inconsistent adherence to these rules. Further, the sensitivity of the symptom-extraction module sometimes led to the identification of minor symptoms that were not central to the classification task, which aims primarily to focus on root causes.