Cognitive Behavioral Therapy (CBT) serves as a powerful tool in addressing and resolving irrational thoughts, especially when related to mental illness. Therefore, accurate identification of cognitive pathways is essential for its successful application in providing psychological support. Today, people’s negative emotional expressions on social media about certain topics (such as “depression”) often contain cognitive distortions and follow incorrect cognitive paths. However, there is a distinct lack of data to analyze cognitive pathways, as well as tools to help psychotherapists identify them effectively. In this study, we defined the cognitive path extraction task as a hierarchical text classification task based on cognitive behavioral therapy, which contains four parent categories and nineteen subcategories. We collected a total of 5,282 sentences from social media and then annotated the data to create a cognitive path extraction dataset. We proposed an interpretable tree structure model and explored three different hierarchical classification strategies. It was verified that our model outperformed five other encoder-based models and far outperformed four decoder-based models while maintaining interpretability. On another public comparison dataset, it also achieved better performance than the other seven models. We have released the code publicly: https://github.com/liuhanfei0426/Hierarchical-Text-Classification-wise2025 .