DiSCo Streamlit Demo

This repository contains a Streamlit demo for
Domain-informed Summarization through Contrast (DiSCo).

The demo follows the IUI 2026 demonstration paper:

Demonstrating Domain-informed Summarization through Contrast (DiSCo)
Proceedings of the ACM Conference on Intelligent User Interfaces (IUI 2026)

The application compares:

• a standard (frequency-based) summary of accommodation reviews, and
• a DiSCo summary, which highlights what is overrepresented or underrepresented relative to other accommodations in the same domain.

---

📁 Project structure

.
├── app.py                  # Streamlit application
├── requirements.txt
├── data/
│   ├── summaries.csv
│   ├── <domain>_signatures.csv
│   ├── <domain>_freqs.csv
│   └── <domain>_dvr.csv
Each <domain> corresponds to an accommodation domain (e.g., hotels, apartments).

📊 Expected data files

The application expects all input data to be placed under the data/ directory.

For each accommodation domain <domain>, the following CSV files are required.

---

<domain>_signatures.csv

Domain-level divergence signatures per accommodation.

• Rows: accommodations
• Columns:
  • The first column must be accommodation_id
  • All remaining columns correspond to topic–sentiment elements
• Values: numeric divergence scores

---

<domain>_freqs.csv

Topic mention frequencies per accommodation.

Required columns:

• accommodation_id
• element
• frequency

---

<domain>_dvr.csv

Global domain reference weights.

Required columns:

• element
• global_weight

---

summaries.csv

Textual summaries displayed in the UI.

Required columns:

• accommodation_id
• baseline_summary
• expectations_aware_summary

---

Notes

• All accommodation_id values are treated as strings.
• Only accommodations present in both the domain files and summaries.csv will be selectable in the application.
• File names must follow the exact naming conventions shown above.

🚀 Running the app locally

1. Clone the repository

git clone https://github.com/ScanLab-ossi/DisCo--Demo.git
cd DisCo--Demo

2. Create a virtual environment (recommended)

python -m venv .venv
source .venv/bin/activate   # macOS / Linux
# or
.venv\Scripts\activate      # Windows

3. Install dependencies

pip install -r requirements.txt

4. Launch Streamlit

streamlit run app.py

The app will open automatically in your browser.

🧠 How to interpret the demo

• Standard summary Reflects what guests frequently mention within the selected accommodation.

• DiSCo summary Highlights topics that are:

  • Overrepresented: mentioned more than expected compared to the domain
  • Underrepresented: mentioned less than expected compared to the domain

Topics are ranked by absolute divergence magnitude, so the strongest deviations from the domain appear first.

📄 Citation If you use this demo or code, please cite:

@inproceedings{disco2026,
  title     = {Demonstrating Domain-informed Summarization through Contrast (DiSCo)},
  booktitle = {Proceedings of the ACM Conference on Intelligent User Interfaces (IUI)},
  year      = {2026}
}