Enhanced DARVO detector v2 - 84% accuracy, improved accountability detection

README.md

@@ -1,199 +1,205 @@
 ---
+language: en
+license: mit
 library_name: transformers
-tags: []
+tags:
+- text-classification
+- psychology
+- abuse-detection
+- darvo
+- manipulation-detection
+- mental-health
+- relationship-analysis
+- tether-pro
+datasets:
+- custom
+metrics:
+- mse
+- mae
+- accuracy
+- auc
+model-index:
+- name: tether-darvo-regressor-v1
+  results:
+  - task:
+      type: text-classification
+      name: DARVO Detection
+    metrics:
+    - type: mse
+      value: 0.043
+    - type: mae  
+      value: 0.171
+    - type: accuracy
+      value: 0.842
+    - type: auc
+      value: 0.881
 ---
 
-# Model Card for Model ID
+# Tether Pro DARVO Regressor v2
 
-<!-- Provide a quick summary of what the model is/does. -->
+## Model Description
 
+This model detects DARVO (Deny, Attack, Reverse Victim & Offender) manipulation tactics in text communication. DARVO is a psychological manipulation strategy where an abuser:
 
+1. **Denies** the abuse ever happened
+2. **Attacks** the victim for bringing it up  
+3. **Reverses** the roles to claim they are the victim
 
-## Model Details
+## Key Features
 
-### Model Description
+🎯 **Role-Aware Detection**: Distinguishes between genuine accountability and manipulation tactics
+🔬 **Research-Grade Accuracy**: 84% accuracy with 0.88 AUC
+⚡ **Real-Time Analysis**: Optimized for fast inference
+🛡️ **Professional Use**: Designed for therapists, legal professionals, and safety applications
 
-<!-- Provide a longer summary of what this model is. -->
+## Performance Metrics
 
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+| Metric | Score |
+|--------|-------|
+| **R²** | 0.665 |
+| **MAE** | 0.171 |
+| **MSE** | 0.043 |
+| **Accuracy** | 84.2% |
+| **AUC** | 88.1% |
 
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
+## Usage
 
-### Model Sources [optional]
+```python
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+import torch
 
-<!-- Provide the basic links for the model. -->
+# Load model and tokenizer
+tokenizer = AutoTokenizer.from_pretrained("SamanthaStorm/tether-darvo-regressor-v1")
+model = AutoModelForSequenceClassification.from_pretrained("SamanthaStorm/tether-darvo-regressor-v1")
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+# Example usage
+text = "You're the one being abusive to me right now"
 
-## Uses
+# Tokenize and predict
+inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
+with torch.no_grad():
+    outputs = model(**inputs)
+    darvo_score = outputs.logits.item()
 
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+print(f"DARVO Score: {darvo_score:.3f}")  # Higher scores = more DARVO tactics
+```
 
-### Direct Use
+## Score Interpretation
 
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+- **0.0 - 0.3**: Genuine accountability, healthy communication
+- **0.3 - 0.6**: Some defensive patterns, mild deflection  
+- **0.6 - 0.8**: Moderate DARVO tactics, concerning patterns
+- **0.8 - 1.0**: Strong DARVO tactics, victim reversal
 
-[More Information Needed]
+## Example Predictions
 
-### Downstream Use [optional]
+| Text | DARVO Score | Interpretation |
+|------|-------------|----------------|
+| "You're the one being abusive to me right now" | 0.870 | High DARVO - victim reversal |
+| "I don't remember saying that" | 0.224 | Low DARVO - simple denial |
+| "I take full responsibility for my actions" | 0.205 | Very low DARVO - accountability |
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+## Training Data
 
-[More Information Needed]
+Trained on 285 carefully curated examples including:
+- **High DARVO**: Explicit victim reversal tactics
+- **Medium DARVO**: Deflection and minimization patterns  
+- **Low DARVO**: Genuine accountability and healthy communication
+- **Contrast Examples**: Non-apologies vs real apologies
 
-### Out-of-Scope Use
+## Applications
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+### 🏥 Clinical Therapy
+- Help therapists identify manipulation patterns in client relationships
+- Assist in couples counseling to recognize unhealthy dynamics
+- Support trauma therapy by validating victim experiences
 
-[More Information Needed]
+### ⚖️ Legal Documentation  
+- Analyze communication patterns in domestic violence cases
+- Provide objective evidence of psychological manipulation
+- Support legal professionals in building abuse cases
 
-## Bias, Risks, and Limitations
+### 🏢 Workplace Safety
+- Identify harassment patterns in workplace communications
+- Support HR investigations with objective analysis
+- Create safer work environments through pattern recognition
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+## Ethical Considerations
 
-[More Information Needed]
+⚠️ **Important**: This model is designed to assist professionals and should not be used as the sole basis for serious decisions about relationships or safety.
 
-### Recommendations
+- **Professional Use**: Best used by trained therapists, counselors, and legal professionals
+- **Context Matters**: Consider cultural, situational, and individual factors
+- **Not Diagnostic**: Does not diagnose psychological conditions
+- **Privacy**: Ensure consent when analyzing personal communications
 
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+## Technical Details
 
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+- **Base Model**: DistilBERT (distilbert-base-uncased)
+- **Architecture**: Custom regression head with 4-layer neural network
+- **Training**: 8 epochs with cosine learning rate scheduling
+- **Optimization**: Mixed precision training (FP16)
+- **Max Length**: 256 tokens for efficiency
+
+## Model Architecture
+
+```
+DistilBERT Base
+    ↓
+Linear(768 → 768) + GELU + Dropout
+    ↓  
+Linear(768 → 384) + GELU + Dropout
+    ↓
+Linear(384 → 192) + GELU + Dropout
+    ↓
+Linear(192 → 1) + Sigmoid
+    ↓
+DARVO Score (0.0 - 1.0)
+```
+
+## Version History
+
+### v2 (Current)
+- ✅ Enhanced training dataset (285 examples)
+- ✅ Improved architecture with deeper regression head
+- ✅ Better score calibration for accountability detection
+- ✅ Added contrast examples (fake vs real apologies)
+- ✅ 84% accuracy (up from 40%)
+
+### v1 (Previous)  
+- Basic DARVO detection capability
+- Limited training data
+- Lower accuracy performance
+
+## Citation
+
+If you use this model in research or professional practice, please cite:
+
+```bibtex
+@misc{tether-darvo-regressor-v1,
+  title={Tether Pro DARVO Regressor: Role-Aware Detection of Manipulation Tactics},
+  author={SamanthaStorm},
+  year={2024},
+  howpublished={\url{https://huggingface.co/SamanthaStorm/tether-darvo-regressor-v1}},
+}
+```
+
+## Contact & Support
+
+For questions about integration, licensing, or professional applications:
+- 📧 Enterprise: [email protected]
+- 🌐 Documentation: docs.tether.ai
+- 📅 Consultation: calendly.com/tether-pro
+
+## Related Models
+
+Part of the **Tether Pro AI Suite**:
+- 🛡️ **Boundary Health Detector**: `SamanthaStorm/healthy-boundary-predictor`
+- 🎯 **Abuse Pattern Detector**: `SamanthaStorm/tether-multilabel-v6`  
+- 🎭 **Sentiment Analyzer**: `SamanthaStorm/tether-sentiment-v3`
+- 🧩 **Fallacy Detector**: `SamanthaStorm/fallacy-detector` (coming soon)
+- 🎯 **Intent Classifier**: `SamanthaStorm/intent-detector` (coming soon)
 
-## How to Get Started with the Model
-
-Use the code below to get started with the model.
-
-[More Information Needed]
-
-## Training Details
-
-### Training Data
-
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-
-[More Information Needed]
-
-### Training Procedure
-
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-
-#### Preprocessing [optional]
-
-[More Information Needed]
-
-
-#### Training Hyperparameters
-
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-
-#### Speeds, Sizes, Times [optional]
-
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-
-[More Information Needed]
-
-## Evaluation
-
-<!-- This section describes the evaluation protocols and provides the results. -->
-
-### Testing Data, Factors & Metrics
-
-#### Testing Data
-
-<!-- This should link to a Dataset Card if possible. -->
-
-[More Information Needed]
-
-#### Factors
-
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-
-[More Information Needed]
-
-#### Metrics
-
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-
-[More Information Needed]
-
-### Results
-
-[More Information Needed]
-
-#### Summary
-
-
-
-## Model Examination [optional]
-
-<!-- Relevant interpretability work for the model goes here -->
-
-[More Information Needed]
-
-## Environmental Impact
-
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-
-## Technical Specifications [optional]
-
-### Model Architecture and Objective
-
-[More Information Needed]
-
-### Compute Infrastructure
-
-[More Information Needed]
-
-#### Hardware
-
-[More Information Needed]
-
-#### Software
-
-[More Information Needed]
-
-## Citation [optional]
-
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-
-**BibTeX:**
-
-[More Information Needed]
-
-**APA:**
-
-[More Information Needed]
-
-## Glossary [optional]
-
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-
-[More Information Needed]
-
-## More Information [optional]
-
-[More Information Needed]
-
-## Model Card Authors [optional]
-
-[More Information Needed]
-
-## Model Card Contact
+---
 
-[More Information Needed]
+*Built with ❤️ for safer communication analysis*

config.json

@@ -1,33 +1,30 @@
 {
+  "activation": "gelu",
   "architectures": [
-    "RobertaForSequenceClassification"
+    "DistilBertForSequenceClassification"
   ],
-  "attention_probs_dropout_prob": 0.1,
-  "bos_token_id": 0,
-  "classifier_dropout": null,
-  "eos_token_id": 2,
-  "hidden_act": "gelu",
-  "hidden_dropout_prob": 0.1,
-  "hidden_size": 768,
+  "attention_dropout": 0.1,
+  "dim": 768,
+  "dropout": 0.1,
+  "hidden_dim": 3072,
   "id2label": {
     "0": "LABEL_0"
   },
   "initializer_range": 0.02,
-  "intermediate_size": 3072,
   "label2id": {
     "LABEL_0": 0
   },
-  "layer_norm_eps": 1e-05,
-  "max_position_embeddings": 514,
-  "model_type": "roberta",
-  "num_attention_heads": 12,
-  "num_hidden_layers": 12,
-  "pad_token_id": 1,
-  "position_embedding_type": "absolute",
+  "max_position_embeddings": 512,
+  "model_type": "distilbert",
+  "n_heads": 12,
+  "n_layers": 6,
+  "pad_token_id": 0,
   "problem_type": "regression",
+  "qa_dropout": 0.1,
+  "seq_classif_dropout": 0.2,
+  "sinusoidal_pos_embds": false,
+  "tie_weights_": true,
   "torch_dtype": "float32",
-  "transformers_version": "4.51.3",
-  "type_vocab_size": 1,
-  "use_cache": true,
-  "vocab_size": 50265
+  "transformers_version": "4.53.0",
+  "vocab_size": 30522
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b199dbf389cbe30289fc8cc9430d4a1ca05b4a809734cb731ccd2dde6d9dca00
-size 498609748
+oid sha256:6a4e929f41e35442855d7ded918ee23f9d68a7e0fb3c7675df0728d5421c0e86
+size 267829484

model_info.json

@@ -0,0 +1,14 @@
+{
+  "model_type": "regression",
+  "task": "darvo-detection",
+  "version": "2.0",
+  "performance": {
+    "mse": 0.043,
+    "mae": 0.171,
+    "accuracy": 0.842,
+    "auc": 0.881,
+    "r_squared": 0.665
+  },
+  "training_examples": 285,
+  "architecture": "distilbert-base-uncased + custom regression head"
+}

special_tokens_map.json

@@ -1,51 +1,7 @@
 {
-  "bos_token": {
-    "content": "<s>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "cls_token": {
-    "content": "<s>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "eos_token": {
-    "content": "</s>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "mask_token": {
-    "content": "<mask>",
-    "lstrip": true,
-    "normalized": false,
-    "rstrip": false,
-    "single_word": false
-  },
-  "pad_token": {
-    "content": "<pad>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "sep_token": {
-    "content": "</s>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  },
-  "unk_token": {
-    "content": "<unk>",
-    "lstrip": false,
-    "normalized": true,
-    "rstrip": false,
-    "single_word": false
-  }
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
 }

tokenizer_config.json

@@ -1,65 +1,56 @@
 {
-  "add_prefix_space": false,
   "added_tokens_decoder": {
     "0": {
-      "content": "<s>",
+      "content": "[PAD]",
       "lstrip": false,
-      "normalized": true,
+      "normalized": false,
       "rstrip": false,
       "single_word": false,
       "special": true
     },
-    "1": {
-      "content": "<pad>",
+    "100": {
+      "content": "[UNK]",
       "lstrip": false,
-      "normalized": true,
+      "normalized": false,
       "rstrip": false,
       "single_word": false,
       "special": true
     },
-    "2": {
-      "content": "</s>",
+    "101": {
+      "content": "[CLS]",
       "lstrip": false,
-      "normalized": true,
+      "normalized": false,
       "rstrip": false,
       "single_word": false,
       "special": true
     },
-    "3": {
-      "content": "<unk>",
+    "102": {
+      "content": "[SEP]",
       "lstrip": false,
-      "normalized": true,
+      "normalized": false,
       "rstrip": false,
       "single_word": false,
       "special": true
     },
-    "50264": {
-      "content": "<mask>",
-      "lstrip": true,
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
       "normalized": false,
       "rstrip": false,
       "single_word": false,
       "special": true
     }
   },
-  "bos_token": "<s>",
   "clean_up_tokenization_spaces": false,
-  "cls_token": "<s>",
-  "eos_token": "</s>",
-  "errors": "replace",
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
   "extra_special_tokens": {},
-  "mask_token": "<mask>",
-  "max_length": 256,
+  "mask_token": "[MASK]",
   "model_max_length": 512,
-  "pad_to_multiple_of": null,
-  "pad_token": "<pad>",
-  "pad_token_type_id": 0,
-  "padding_side": "right",
-  "sep_token": "</s>",
-  "stride": 0,
-  "tokenizer_class": "RobertaTokenizer",
-  "trim_offsets": true,
-  "truncation_side": "right",
-  "truncation_strategy": "longest_first",
-  "unk_token": "<unk>"
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "DistilBertTokenizer",
+  "unk_token": "[UNK]"
 }