Building an Open Turkish EV Charging Intent Dataset

Description and use

An open Turkish EV Charging Intent Dataset (192 queries, 8 classes) was created to prototype text classifiers and evaluate intent routers. The synthetic, balanced data includes validation scripts and is licensed under CC BY 4.0.

Problem

Electric-vehicle assistants need to reliably identify user intent (e.g., finding stations, comparing prices, route planning) before retrieving data or generating answers, as different requests require different tools and freshness guarantees.

Outcome

A 192-record Turkish EV Charging Intent Dataset (v1.0.0) was created, distributed evenly across eight intent classes. It serves as a controlled regression suite for prototyping Turkish text classifiers and evaluating intent routers, with clear provenance markers and automated validation scripts.

Workflow

• Define eight balanced intent classes for Turkish EV queries (e.g., FIND_STATION, COMPARE_PRICE, ROUTE_PLANNING).
• Generate 192 synthetic-editorial examples (24 per class) using documented subject lists and templates to ensure privacy and reproducibility.
• Split data deterministically into train (128), validation (32), and test (32) sets.
• Implement automated validation checks for row counts, class balance, split sizes, unique IDs, and required fields.
• Load JSON Lines files using Python (pandas) for prototyping classifiers or evaluating embedding-based routers.
• Evaluate models using macro F1, per-class precision/recall/F1, and confusion matrices on the fixed test split.

Evidence

Version 1.0.0 contains 192 Turkish queries distributed evenly across eight intent classes.

Source

It is open under CC BY 4.0, includes fixed train, validation, and test splits.

Source

This dataset does not measure real search demand or the natural frequency of user intents. It is a controlled regression suite.

Source