Update README.md

README.md

@@ -34,95 +34,6 @@ This is Stage 2 of a hierarchical dissatisfaction detection pipeline. After Stag
 | **WRONG_ANSWER** | System response is incorrect | "No, I said 2 people not 3" |
 | **WANT_DIFFERENT** | User wants alternative options | "What else do you have?" |
 
-## Usage
-
-```python
-from transformers import AutoTokenizer, AutoModelForSequenceClassification
-import torch
-
-tokenizer = AutoTokenizer.from_pretrained("rootfs/modernbert-dissat-explainer")
-model = AutoModelForSequenceClassification.from_pretrained("rootfs/modernbert-dissat-explainer")
-
-id2label = {0: "NEED_CLARIFICATION", 1: "WRONG_ANSWER", 2: "WANT_DIFFERENT"}
-
-def explain_dissatisfaction(query, response, followup=None):
-    if followup:
-        text = f"[USER QUERY] {query}\n[SYSTEM RESPONSE] {response}\n[USER FOLLOWUP] {followup}"
-    else:
-        text = f"[USER QUERY] {query}\n[SYSTEM RESPONSE] {response}"
-    
-    inputs = tokenizer(text, return_tensors="pt", truncation=True, max_length=512)
-    
-    with torch.no_grad():
-        outputs = model(**inputs)
-        probs = torch.softmax(outputs.logits, dim=-1)[0]
-    
-    pred_id = probs.argmax().item()
-    return id2label[pred_id], probs[pred_id].item()
-
-# Examples
-print(explain_dissatisfaction(
-    "python tutorial",
-    "Select one to refine your search Options: beginners | advanced"
-))  # ('NEED_CLARIFICATION', 1.0)
-
-print(explain_dissatisfaction(
-    "Book for 2",
-    "Table for 3 confirmed",
-    "No, I said 2 people"
-))  # ('WRONG_ANSWER', 1.0)
-
-print(explain_dissatisfaction(
-    "Find restaurant",
-    "Italian Kitchen",
-    "What else do you have?"
-))  # ('WANT_DIFFERENT', 1.0)
-```
-
-## Training Data
-
-- **NEED_CLARIFICATION**: MIMICS search clarification dataset (522,955 examples)
-- **WRONG_ANSWER**: SGD NEGATE dialogue acts (13,379 examples)
-- **WANT_DIFFERENT**: SGD REQUEST_ALTS dialogue acts (19,869 examples)
-
-## Performance
-
-| Metric | Value |
-|--------|-------|
-| Accuracy | 99.88% |
-| F1 Macro | 98.60% |
-| F1 Weighted | 99.88% |
-
-### Per-Class F1
-
-| Class | F1 Score | % of Data |
-|-------|----------|-----------|
-| NEED_CLARIFICATION | 100.0% | 94.0% |
-| WRONG_ANSWER | 97.45% | 2.4% |
-| WANT_DIFFERENT | 98.36% | 3.6% |
-
-## Class Weights (for imbalance)
-
-```python
-class_weights = [0.355, 13.858, 9.331]  # NC, WA, WD
-```
-
-## Limitations
-
-- **NEED_CLARIFICATION** is tied to MIMICS format ("Select one to refine your search Options: ...")
-- General clarification requests ("Can you explain more?") classify as WANT_DIFFERENT
-- Best used in conjunction with Stage 1 binary classifier
-
-## Hierarchical Pipeline
-
-```
-User Follow-up → Stage 1 (SAT/DISSAT) → if DISSAT → Stage 2 (This Model)
-                                                    ↓
-                                        NEED_CLARIFICATION
-                                        WRONG_ANSWER
-                                        WANT_DIFFERENT
-```
-
 ## License
 
 Apache 2.0