Text Classification
sentence-transformers
Safetensors
English
questionnaire-harmonization
prototype-learning
impulse-control
Instructions to use julia-pfarr/HarmoniCA_impulse-control with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use julia-pfarr/HarmoniCA_impulse-control with sentence-transformers:
from sentence_transformers import SentenceTransformer model = SentenceTransformer("julia-pfarr/HarmoniCA_impulse-control") sentences = [ "The weather is lovely today.", "It's so sunny outside!", "He drove to the stadium." ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [3, 3] - Notebooks
- Google Colab
- Kaggle
metadata
language: en
license: cc-by-4.0
base_model: hkunlp/instructor-large
tags:
- questionnaire-harmonization
- sentence-transformers
- prototype-learning
- impulse-control
pipeline_tag: text-classification
HarmoniCA — Impulse Control Model
Part of the HarmoniCA collection for harmonising psychiatric questionnaire items across studies. This model assigns items from impulse control questionnaires to one of six theoretically motivated symptom dimensions.
Dimensions
| ID | Label | Description |
|---|---|---|
| 1 | Pathological Gambling | preoccupation with gambling, difficulty controlling gambling urges, time and money spent on gambling |
| 2 | Hypersexuality | preoccupation with sex, difficulty controlling sexual urges, time and money spent on sexual activities |
| 3 | Compulsive Buying | preoccupation with buying, difficulty controlling buying urges, excessive spending |
| 4 | Compulsive Eating | preoccupation with eating, difficulty controlling eating urges, overeating |
| 5 | Punding & Hobbyism | stereotyped or repetitive behaviors, excessive engagement with hobbies, purposeless repetitive activities |
| 6 | Dopamine Dysregulation Syndrome | compulsive use of dopaminergic medication beyond therapeutic need |
Items rated as not belonging to the impulse control construct are assigned dimension -1 (Does not fit).
Model Architecture
- Type: PrototypeModel (prototype-based learning with a fine-tuned sentence transformer)
- Base encoder: hkunlp/instructor-large
- Inference: Fine-tuned encoder with learned class prototypes
The PrototypeModel learns a single prototype embedding per dimension. At inference, items are assigned to the nearest prototype in the fine-tuned embedding space.
Training Data
- Training items: 58 expert-labeled questionnaire items
- Questionnaires: QUIP-C, QUIP-RS
- Labels: Expert consensus from two rounds of surveys (psychiatrists/neurologists)
Evaluation
Evaluated on 124 held-out items from unseen questionnaires against expert labels.
| Cohen's κ | Accuracy |
|---|---|
| 0.884 | 90.3% |
Limitations
- Trained on QUIP-RS and QUIP-C; generalization to questionnaires that frame items very differently may be limited.
- Dopamine Dysregulation Syndrome (dimension 6) is Parkinson's-specific and may not be applicable in non-PD contexts.
- Dimension structure reflects theoretical consensus among a specific expert panel and may not match all theoretical frameworks.
Citation
in preparation