Update README.md

README.md

@@ -1,88 +1,174 @@
+# AffectDynamics(Team AGI) — Longitudinal Affect Prediction Model
+
+AffectDynamics is a temporal affect modeling system developed for **SemEval-2026 Task 2: Predicting Variation in Emotional Valence and Arousal over Time from Ecological Essays**.
+
+The model predicts emotional **valence** and **arousal** from longitudinal text written by users across time. It combines transformer-based text encoding with temporal modeling and user-level conditioning to capture both **stable emotional baselines** and **dynamic emotional changes**.
+
+---
+
+# Model Details
+
+**Model name:** AffectDynamics-SemEval2026Task2  
+**Developer:** Harsh Rathva  
+**Institution:** Sardar Vallabhbhai National Institute of Technology (SVNIT), Surat  
+**Email:** u24ai036@aid.svnit.ac.in  
+
+## Architecture
+
+The system consists of four main components:
+
+### 1. Text Encoder
+- **RoBERTa-Large** transformer encoder
+- Produces contextual embeddings for each text input.
+
+Different pooling strategies are used depending on text type:
+- Essays → CLS / pooler representation
+- Feeling word lists → mean pooled token embeddings
+
+### 2. Temporal Encoder
+- **Unidirectional GRU**
+- Models longitudinal emotional dynamics across user timelines.
+- Ensures **causal temporal modeling** (no future information leakage).
+
+### 3. User Conditioning
+- **Gated user embedding**
+- Incorporates user-level statistics such as:
+  - number of samples
+  - timeline length
+  - emotional entropy
+
+This allows interpolation between **user-specific** and **global representations**.
+
+### 4. Prediction Heads
+The model supports three prediction tasks:
+
+| Task | Description |
+|-----|-------------|
+| **Subtask 1 (S1)** | Absolute valence and arousal prediction |
+| **Subtask 2A (S2A)** | Short-term emotional state change prediction |
+| **Subtask 2B (S2B)** | Long-term dispositional change prediction |
+
 ---
-license: mit
-language:
-  - en
-library_name: pytorch
-pipeline_tag: text-classification
-base_model:
-  - roberta-large
-tags:
-  - semeval
-  - semeval2026
-  - affective-computing
-  - emotion-regression
-  - valence-arousal
-  - temporal-modeling
-datasets:
-  - semeval2026-task2
-metrics:
-  - pearsonr
-  - r_within
-  - r_between
-model-index:
-  - name: AffectDynamics-SemEval2026Task2
-    results:
-      - task:
-          type: text-classification
-          name: SemEval-2026 Task 2 (Composite)
-        dataset:
-          type: semeval2026-task2
-          name: SemEval-2026 Task 2 Validation Split
-          split: validation
-        metrics:
-          - type: r_composite
-            name: Composite Correlation
-            value: 0.6990
+
+# Training Data
+
+The model was trained using the official **SemEval-2026 Task 2 dataset**.
+
+### Dataset statistics
+
+- **Total texts:** 5,285
+- **Training texts:** 2,764
+- **Users:** 182 total (137 in training)
+- **Time span:** 2021–2024
+
+Each entry contains:
+
+| Field | Description |
+|------|-------------|
+| user_id | Anonymous user identifier |
+| text | Ecological essay or feeling word list |
+| timestamp | Time of writing |
+| collection_phase | Study phase |
+| valence | Emotional valence (-2 to 2) |
+| arousal | Emotional arousal (0 to 2) |
+
+The texts were written by **U.S. service-industry workers** describing how they felt at the moment.
+
 ---
 
-### Model details
+# Training Details
 
-- **Model type**: Multi-task temporal regression (Subtask 1, 2A, 2B)
-- **Backbone**: `roberta-large`
-- **Temporal encoder**: 2-layer unidirectional GRU (hidden size 384)
-- **Personalization**: Gated user embedding (24-dim)
-- **Training objective**: Correlation-first, variance-aware losses aligned with task metrics
-- **Primary checkpoint**: `best-epoch=14-val_r_composite_avg=0.6990.ckpt`
+### Optimization
 
-### Intended use
+- Optimizer: **AdamW**
+- Scheduler: **OneCycleLR**
+- Batch size: 4
+- Training epochs: 10
 
-- Research use for longitudinal affect forecasting on SemEval-style data.
-- Produces continuous predictions for:
-  - Subtask 1: `pred_valence`, `pred_arousal`
-  - Subtask 2A: `pred_state_change_valence`, `pred_state_change_arousal`
-  - Subtask 2B: `pred_dispo_change_valence`, `pred_dispo_change_arousal`
+### Learning rates
 
-### Out-of-scope use
+| Component | Learning Rate |
+|----------|---------------|
+| RoBERTa encoder | 2e-6 |
+| GRU | 3e-4 |
+| Task heads | 2e-5 |
 
-- Clinical diagnosis or mental health decision support.
-- High-stakes individual-level decision making.
-- Use on domains, languages, or demographics not represented in SemEval Task 2 data without re-validation.
+### Loss Functions
 
-### Training and evaluation data
+| Task | Loss |
+|----|----|
+| Subtask 1 | Ordinal regression with label smoothing |
+| Subtask 2A | Smooth L1 loss for delta prediction |
+| Subtask 2B | Mean squared error |
 
-- Source task: SemEval-2026 Task 2 (shared-task format).
-- Training corpus in this repo includes:
-  - `data/train_subtask1.csv`
-  - `data/train_subtask2a.csv` (or computed from Subtask 1 timeline)
-  - `data/train_subtask2b_user_disposition_change.csv`
-- Validation strategy: temporal per-user split to prevent future leakage.
+---
 
-### Metrics
+# Evaluation Results
 
-- **Subtask 1**: `r_within`, `r_between`, `r_composite` (per SemEval evaluator)
-- **Subtask 2A/2B**: Pearson correlation (`r`) on forecasting targets
-- **Checkpoint selection signal**: `val_r_composite_avg`
+Official evaluation results from SemEval-2026 Task 2:
 
+| Task | Metric | Valence | Arousal |
+|----|----|----|----|
+| **Subtask 1** | Composite correlation | **0.600** | **0.452** |
+| **Subtask 2A** | Pearson correlation | -0.167 | -0.147 |
+| **Subtask 2B** | Pearson correlation | 0.086 | -0.081 |
 
-### Limitations and bias
+The model demonstrates strong performance on **absolute affect prediction**, but exhibits limitations in **change detection tasks**, highlighting a trade-off between temporal stability and sensitivity to emotional transitions.
 
-- Performance depends on temporal history quality and per-user data sparsity.
-- Arousal typically has lower correlation than valence due to lower target variance.
-- Predictions are correlation-optimized for benchmark metrics and may require calibration for deployment settings.
+---
+
+# Intended Use
+
+This model is intended for **research purposes** including:
+
+- longitudinal affect modeling
+- emotion prediction from text
+- temporal NLP modeling
+- ecological momentary assessment analysis
+
+---
+
+# Limitations
+
+Several limitations should be considered:
+
+1. **Stability bias**
+   - Temporal modeling tends to smooth predictions, reducing sensitivity to abrupt emotional changes.
+
+2. **Dataset domain**
+   - Data comes from a specific population (U.S. service-industry workers), which may limit generalization.
+
+3. **Small number of users**
+   - Training data includes only 137 users.
+
+4. **Change prediction difficulty**
+   - Predicting emotional deltas is significantly harder than predicting absolute states.
+
+---
+
+# Ethical Considerations
+
+Emotion prediction models must be used responsibly.
+
+Potential concerns include:
+
+- **Privacy risks** when modeling personal emotional data
+- **Misuse for emotional manipulation**
+- **Bias from dataset demographics**
+
+This model should **not be used for clinical or psychological diagnosis**.
+
+---
+
+# Reproducibility
+
+Code and training pipeline are available at:
+
+**GitHub Repository**  
+https://github.com/ezylopx5/AffectDynamics-SemEval2026Task2
+
+---
 
-## Task Overview
+# Citation
 
-**Three interconnected subtasks:**
-- **Subtask 1**: Longitudinal Affect Assessment - Predict valence/arousal for each text in a user's timeline
-- **Subtask 2A**: State Change Detection - Predict short-term emotional shifts between consecutive texts  
-- **Subtask 2B**: Dispositional Change - Predict long-term changes in baseline emotional state
+If you use this model, please cite the system description paper: