docs/PLAN.md

Dialog Emotion Timeline Demo Plan

Goal

Build a readable Gradio demo for Hugging Face Spaces that tracks emotional color across a dialogue over time.

The interface reads model-like outputs from CSV instead of calling a model directly. The default demo uses synthetic Russian messenger-style data, and users can upload another CSV with the same contract.

Decisions

• Chunk unit: one dialogue message per CSV row.
• Data mode: bundled synthetic CSV plus optional CSV upload.
• Refresh mode: static viewer, no live polling in v1.
• Probability semantics: six probabilities in [0, 1]; each row must sum to 1.
• Emotions: joy, warmth, sadness, anger, anxiety, neutral.
• UI language: Russian.
• Chunk details: shown inside each message block on hover/focus.

CSV Contract

Required columns:

• turn_index
• timestamp
• sender
• text
• joy
• warmth
• sadness
• anger
• anxiety
• neutral

Checklist

• Add dependencies and HF Spaces metadata.
• Generate deterministic synthetic dialogue CSV.
• Implement CSV loading and validation.
• Implement smoothing and Plotly figure generation.
• Build Gradio Blocks UI.
• Add focused tests.
• Run local verification.
• Commit changes in logical steps.

Commit Log

• Done: docs: add project plan
• Done: chore: add gradio dependencies and space config
• Done: feat: add csv contract and synthetic dialogue
• Done: feat: build emotion timeline ui
• Done: test: cover data loading and smoothing
• Done: docs: update implementation tracker
• Done: style: refine emotion timeline interface
• Current: graph supports line/stacked area modes, emotion focus, and a per-message slice chart.

Verification

• uv run pytest
• uv run python -m compileall app.py main.py dialog_emo_demo scripts tests
• uv run python app.py
• Browser smoke test at http://127.0.0.1:7860

Non-goals For v1

• Model inference inside the UI.
• Automatic CSV refresh or file watching.
• Plot-hover to message-block synchronization.
• Authentication, persistence, or multi-session state.