Shreevastava Cognitive Distortions: Multi-Label Therapy Text Classification

2530 human-annotated therapy forum posts and 646 synthetic entries form a multi-label dataset for classifying cognitive distortions. The data was used to train the DistortionMlp model for an on-device CBT journaling app and was uploaded to Hugging Face by masked-kunsiquat on 2026-04 17.

Use Cases

• Training multi-label text classifiers for cognitive distortion detection based on therapy forum posts.
• Evaluating the effectiveness of synthetic data generation for underrepresented distortion classes.
• Fine-tuning models for on-device mental health journaling applications based on the described use case.
• Researching linguistic patterns of cognitive distortions in therapeutic text based on the annotated corpus.

Strengths

• 2530 human-annotated entries provide a foundation of real-world data.
• Includes 646 synthetic entries to cover two specific distortion classes (DISQUALIFYING_POSITIVE and BLAME) not in the original corpus.
• Covers 10 distinct cognitive distortion classes for multi-label classification.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full combined dataset is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to the specific therapy forum source and annotation process.

Provenance

Source

Shreevastava et al. therapy forum posts (human-annotated) and Claude-generated synthetic data.

Collection Method

Human annotation and synthetic generation.

Time Range

null

Freshness

Last updated 2026-04-17 20:28:03; freshness should be verified.

Geography

null