--- library_name: transformers license: mit language: - en base_model: - google-bert/bert-base-uncased pipeline_tag: text-classification tags: - PyTorch - multi-class-classification --- This is Bert-base-uncased model fine-tuned for topic classification of therapist remarks in psychotherapeutic contexts. The task is a multi-class classification with the following labels: ```python id2label = {0: 'Time Up and Future Meetings', 1: 'Complex Emotions Toward Him', 2: 'Desires and Disappointments', 3: 'Personal Growth and Decision-Making', 4: 'Self-Acceptance and Relationships', 5: 'Understanding and Confronting Fear', 6: 'See and Understanding Conversations', 7: 'Clarifying Meaning and Intent', 8: 'Desire to Escape and Leave', 9: 'Uncertainty and Understanding Issues', 10: 'Open Conversation and Sharing', 11: 'Exploring Emotional Hurt and Bitterness', 12: 'Guilt and Self-Blame Dynamics', 13: 'Dynamics of Meaningful Relationships', 14: 'Struggles and Desires in Learning', 15: 'Gender Roles and Relationships', 16: 'Struggles with Personal Change', 17: 'Complex Mother-Sibling Relationships', 18: 'Voices and Perception of Sound', 19: 'Difficulties and Emotional Burdens', 20: 'Fear and Reflection on Aging', 21: 'Emotions of Crying and Tears', 22: 'Father-Child Relationships and Authority', 23: 'Possibilities and Potential Outcomes', 24: 'Inner Struggle and Helplessness', 25: 'Pursuing Meaningful Personal Goals', 26: 'Job Anxiety and Self-Reflection', 27: 'Marriage Anxiety and Dependence', 28: 'Expressions of Anger and Frustration', 29: 'Nurturing the inner child', 30: 'Therapy and Father Relationships', 31: 'Expressions of Happiness and Joy', 32: 'Revisiting the Past Together', 33: 'Drinking Habits and Concerns', 34: 'Managing and Increasing Energy Levels', 35: 'Safety and Self-Protection Strategies', 36: 'Understanding Depression and Its Roots'} ``` ## Usage ```python from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("AIPsy/bert-base-therapist-topic-classification-eng") model = AutoModelForSequenceClassification.from_pretrained("AIPsy/bert-base-therapist-topic-classification-eng") text = "You know, I mean, it seems like you could just go to work and feel so much better." encoding = tokenizer( text, truncation=True, padding="max_length", return_tensors="pt" ) output = model(encoding['input_ids'], encoding['attention_mask']).logits result = np.argmax(output.detach().numpy(), axis=-1) print(id2label[result[0]]) 'Job Anxiety and Self-Reflection' ``` ## Dataset The source material was the recordings of psychotherapeutic sessions posted on YouTube in the public domain. After conducting speaker diarization and transcription of the recordings 15324 items (sentences) were obtained. ## Recommendations Given the broad context of issues discussed in psychotherapeutic sessions, the authors believe that this model can be used to analyze human communication in general. ## Metrics Score metrics of trained model |Sample|F1 macro| |:-|:-:| |Test|0.74| |Validation|0.77| |Train|0.97| F1 score metrics for test sample across categories |ID|Topic|Precision|Recall|F1| |:-:|:-|:-:|:-:|:-:| | 0 | Time Up and Future Meetings | 0.81 | 0.80 | 0.80 | | 1 | Complex Emotions Toward Him | 0.86 | 0.88 | 0.87 | | 2 | Desires and Disappointments | 0.71 | 0.73 | 0.72 | | 3 | Personal Growth and Decision-Making | 0.64 | 0.77 | 0.70 | | 4 | Self-Acceptance and Relationships | 0.69 | 0.82 | 0.75 | | 5 | Understanding and Confronting Fear | 0.93 | 0.76 | 0.84 | | 6 | See and Understanding Conversations | 0.77 | 0.85 | 0.81 | | 7 | Clarifying Meaning and Intent | 0.81 | 0.76 | 0.79 | | 8 | Desire to Escape and Leave | 0.68 | 0.60 | 0.64 | | 9 | Uncertainty and Understanding Issues | 0.81 | 0.81 | 0.81 | | 10 | Open Conversation and Sharing | 0.79 | 0.77 | 0.78 | | 11 | Exploring Emotional Hurt and Bitterness | 0.74 | 0.70 | 0.72 | | 12 | Guilt and Self-Blame Dynamics | 0.80 | 0.69 | 0.74 | | 13 | Dynamics of Meaningful Relationships | 0.75 | 0.75 | 0.75 | | 14 | Struggles and Desires in Learning | 0.82 | 0.72 | 0.77 | | 15 | Gender Roles and Relationships | 0.83 | 0.83 | 0.83 | | 16 | Struggles with Personal Change | 0.76 | 0.60 | 0.67 | | 17 | Complex Mother-Sibling Relationships | 0.60 | 0.74 | 0.67 | | 18 | Voices and Perception of Sound | 0.74 | 0.78 | 0.76 | | 19 | Difficulties and Emotional Burdens | 0.75 | 0.67 | 0.71 | | 20 | Fear and Reflection on Aging | 0.62 | 0.64 | 0.63 | | 21 | Emotions of Crying and Tears | 0.65 | 0.76 | 0.70 | | 22 | Father-Child Relationships and Authority | 0.93 | 0.80 | 0.86 | | 23 | Possibilities and Potential Outcomes | 0.72 | 0.64 | 0.68 | | 24 | Inner Struggle and Helplessness | 0.68 | 0.71 | 0.69 | | 25 | Pursuing Meaningful Personal Goals | 0.58 | 0.65 | 0.61 | | 26 | Job Anxiety and Self-Reflection | 0.70 | 0.73 | 0.71 | | 27 | Marriage Anxiety and Dependence | 0.82 | 0.82 | 0.82 | | 28 | Expressions of Anger and Frustration | 0.73 | 0.76 | 0.75 | | 29 | Nurturing the inner child | 0.74 | 0.70 | 0.72 | | 30 | Therapy and Father Relationships | 0.80 | 0.82 | 0.81 | | 31 | Expressions of Happiness and Joy | 0.78 | 0.79 | 0.78 | | 32 | Revisiting the Past Together | 0.78 | 0.60 | 0.68 | | 33 | Drinking Habits and Concerns | 0.69 | 0.79 | 0.73 | | 34 | Managing and Increasing Energy Levels | 0.54 | 0.62 | 0.58 | | 35 | Safety and Self-Protection Strategies | 0.65 | 0.85 | 0.73 | | 36 | Understanding Depression and Its Roots | 0.61 | 0.67 | 0.64 | ## Citation - **Papers:** Vanin, A., Bolshev, V., & Panfilova, A. (2024). Applying LLM and Topic Modelling in Psychotherapeutic Contexts. ArXiv, abs/2412.17449. - **Developed by:** @myentity, @VadZhen, @Alek123 - **License:** MIT **BibTeX:** ``` @misc{vanin2024applyingllmtopicmodelling, title={Applying LLM and Topic Modelling in Psychotherapeutic Contexts}, author={Alexander Vanin and Vadim Bolshev and Anastasia Panfilova}, year={2024}, eprint={2412.17449}, archivePrefix={arXiv}, primaryClass={cs.LG}, url={https://arxiv.org/abs/2412.17449}, } ```