civicdex / README.md
JadeSamLee's picture
Update README.md
b46543e verified
|
Raw
History Blame Contribute Delete
3.79 kB
metadata
dataset: CivicDex
language: multilingual
task: intent-classification, text-routing
license: mit

CivicDex: Multilingual Civic Request Dataset

CivicDex is a structured multilingual dataset designed for understanding and routing public-service requests written in Tamil, Tanglish (romanized Tamil), English, and code-mixed language.

It is built to support AI systems that handle real-world civic service interactions such as complaints, information requests, application support, and grievance escalation in low-resource language settings.


Motivation

Public-service systems often receive citizen requests in unstructured and multilingual formats. In regions like Tamil Nadu, users frequently mix Tamil and English or write Tamil using English characters (Tanglish), making automated processing difficult.

CivicDex addresses this gap by providing a clean, labeled dataset that enables machine learning models to:

  • Understand citizen intent
  • Classify civic service requests
  • Route requests to correct departments
  • Estimate urgency and severity
  • Support multilingual NLP research

Dataset Overview

Each record in CivicDex contains a civic-service request with structured annotations:

  • Raw user input (Tamil / Tanglish / English / code-mixed)
  • Normalized text version
  • English interpretation (gloss)
  • Language type label
  • Intent classification
  • Service category
  • Department mapping
  • Urgency level
  • Severity tag
  • Location metadata (generalized)
  • Dataset split (train/validation/test)

Supported Languages

CivicDex supports real-world multilingual input patterns:

  • Tamil (native script)
  • Tanglish (romanized Tamil)
  • Code-mixed Tamil-English
  • English civic-service requests

This reflects real citizen communication behavior in digital civic platforms.


Key Tasks Supported

CivicDex enables multiple NLP and AI tasks:

1. Intent Classification

Identify the purpose of the request:

  • complaint
  • information_request
  • status_query
  • application_help
  • document_help
  • follow_up
  • escalation

2. Category Classification

Identify civic domain:

  • water_supply
  • roads
  • sanitation
  • electricity
  • drainage
  • public_health
  • certificates
  • welfare
  • transport

3. Department Routing

Map requests to responsible departments such as:

  • Water Supply Department
  • Sanitation Department
  • Roads Department
  • Electricity Department
  • Public Health Department

4. Urgency Prediction

Classify priority level:

  • low
  • medium
  • high

Dataset Design

CivicDex is designed as a seed dataset, meaning:

  • Small but high-quality and structured
  • Focused on realism and usability
  • Designed for extensibility
  • Suitable for baseline ML benchmarking

It can be expanded into a larger civic AI dataset over time.


Use Cases

CivicDex can be used for:

  • Civic grievance automation systems
  • Multilingual chatbot development
  • Government service request routing
  • Low-resource NLP research
  • Intent classification benchmarking
  • AI-powered public service dashboards

Limitations

  • This is a seed dataset (not large-scale)
  • Contains synthetic + realistic samples
  • Geographic scope is limited
  • May not represent all dialect variations
  • Intended for research and prototyping only

Ethics & Privacy

  • No personally identifiable information (PII)
  • No real citizen data is included
  • Locations are generalized or anonymized
  • Designed for safe AI research and civic innovation

Summary

CivicDex provides a structured foundation for building multilingual civic AI systems that can understand, classify, and route public-service requests in low-resource language environments.

It bridges the gap between real-world civic communication and machine learning-ready datasets.