Datasetpublic

FoodSecuriry/turkish-ev-charging-intents

Turkish EV Charging Intent Benchmark A compact, balanced Turkish-language benchmark for routing electric-vehicle and charging questions. It contains 192 editorially generated queries across eight intent classes, with fixed train, validation, and test splits. The dataset was created and is maintained by TekPedal, an independent EV charging map and vehicle decision platform for Türkiye. Intended uses Evaluate Turkish intent classifiers and embedding-based routers.… See the full description on the dataset page: https://huggingface.co/datasets/FoodSecuriry/turkish-ev-charging-intents.

sourceHugging Facecc-by-4.0updated 1mo agoView on Hugging Face

0likes23downloads

Dataset Card

Turkish EV Charging Intent Benchmark

A compact, balanced Turkish-language benchmark for routing electric-vehicle and charging questions. It contains 192 editorially generated queries across eight intent classes, with fixed train, validation, and test splits.

The dataset was created and is maintained by TekPedal, an independent EV charging map and vehicle decision platform for Türkiye.

Intended uses

—Evaluate Turkish intent classifiers and embedding-based routers.
—Test retrieval routing before connecting an assistant to maps, price guides, route planning, or battery content.
—Use as a small, deterministic regression suite for Turkish EV assistants.

This dataset is not a record of real user searches and must not be used to infer demand, search volume, demographics, or individual behavior.

Dataset structure

Field	Type	Description
`id`	string	Stable example identifier
`text`	string	Turkish query
`intent`	class label	One of eight routing intents
`intent_label_tr`	string	Human-readable Turkish label
`target_path`	string	Recommended product-content route, relative to TekPedal
`split`	string	`train`, `validation`, or `test`
`language`	string	Always `tr` in v1
`provenance`	string	Always `synthetic-editorial` in v1

The eight balanced classes are FIND_STATION, COMPARE_PRICE, ROUTE_PLANNING, CHARGING_SPEED, VEHICLE_COMPARISON, HOME_CHARGING, BATTERY_HEALTH, and OWNERSHIP_COST.

Counts and quality controls

—192 total rows; 24 rows per class.
—128 train, 32 validation, and 32 test rows.
—No names, email addresses, device identifiers, session logs, or other personal data.
—No copied search-query logs; all samples are editorially generated.
—Stable IDs and deterministic generation for repeatable regression testing.

Run node scripts/generate-huggingface-dataset.mjs from the TekPedal source repository to regenerate the files. The companion interactive Space exposes the taxonomy and a transparent rule-based baseline.

Limitations

The corpus is deliberately small and template-based. It does not represent the full diversity of Turkish dialects, misspellings, code-switching, or multi-intent requests. A strong score on this dataset is not evidence that a system is safe or production-ready. Route, price, tax, and battery answers require current authoritative sources.

License and attribution

The data is available under CC BY 4.0. Attribute as “TekPedal Turkish EV Charging Intent Benchmark, v1.0.0”. Brand names and trademarks remain the property of their respective owners.

bibtex

@dataset{tekpedal_turkish_ev_intents_2026,
  title     = {Turkish EV Charging Intent Benchmark},
  author    = {TekPedal},
  year      = {2026},
  version   = {1.0.0},
  url       = {https://huggingface.co/datasets/FoodSecuriry/turkish-ev-charging-intents},
  license   = {CC-BY-4.0}
}

Project: tekpedal.com · EV charging map