Building an Open Turkish EV Charging Intent Dataset

构建开放的土耳其语电动汽车充电意图数据集

Electric-vehicle assistants rarely have just one job. A short Turkish question may ask for a nearby station, a charging-price comparison, help planning a route, or an explanation of battery health. Before an application can retrieve current data or generate an answer, it needs to identify that intent reliably. We created the Turkish EV Charging Intent Dataset as a small, transparent starting point for that routing problem. 电动汽车助手很少只有单一的工作。一个简短的土耳其语问题可能是在询问附近的充电站、比较充电价格、寻求路线规划帮助，或是询问电池健康状况。在应用程序检索实时数据或生成答案之前，它需要可靠地识别出用户的意图。我们创建了“土耳其语电动汽车充电意图数据集”，作为解决该路由问题的一个小型、透明的起点。

Version 1.0.0 contains 192 Turkish queries distributed evenly across eight intent classes. It is open under CC BY 4.0, includes fixed train, validation, and test splits, and is maintained by TekPedal, an EV charging map and vehicle decision platform for Türkiye. You can explore the dataset interactively, inspect the source and validation workflow on GitHub, or cite the permanent Zenodo release with DOI 10.5281/zenodo.22062688. 1.0.0 版本包含 192 条土耳其语查询，均匀分布在八个意图类别中。该数据集在 CC BY 4.0 协议下开放，包含固定的训练集、验证集和测试集划分，并由土耳其电动汽车充电地图及车辆决策平台 TekPedal 维护。您可以交互式地探索该数据集，在 GitHub 上查看源代码和验证工作流，或通过 DOI 10.5281/zenodo.22062688 引用永久的 Zenodo 发布版本。

Why intent routing comes first

为什么意图路由至关重要

An assistant should not answer every EV question in the same way. Different requests need different tools and freshness guarantees: a station request needs a map or location index; a price request needs current tariff data; route planning needs distance, range, and charging-stop logic; a battery question needs careful educational content; a vehicle comparison needs structured specifications. An intent router makes that separation explicit. It can send each query to the correct retrieval source, product page, or application workflow. This also makes evaluation easier: teams can test routing independently before measuring the quality of downstream answers. 助手不应以相同的方式回答所有电动汽车相关问题。不同的请求需要不同的工具和时效性保证：站点请求需要地图或位置索引；价格请求需要当前的费率数据；路线规划需要距离、续航里程和充电停靠逻辑；电池问题需要严谨的科普内容；车辆比较则需要结构化的规格参数。意图路由器使这种区分变得明确。它可以将每个查询发送到正确的检索源、产品页面或应用程序工作流。这也简化了评估过程：团队可以在衡量下游答案质量之前，独立测试路由功能。

Dataset design

数据集设计

The taxonomy contains eight balanced classes, with 24 records in each class: FIND_STATION, COMPARE_PRICE, ROUTE_PLANNING, CHARGING_SPEED, VEHICLE_COMPARISON, HOME_CHARGING, BATTERY_HEALTH, OWNERSHIP_COST. Every record includes a stable ID, the Turkish query, the intent identifier, a human-readable Turkish label, a suggested TekPedal content route, the assigned split, the language, and a provenance marker. 该分类体系包含八个平衡的类别，每个类别有 24 条记录：查找站点 (FIND_STATION)、比较价格 (COMPARE_PRICE)、路线规划 (ROUTE_PLANNING)、充电速度 (CHARGING_SPEED)、车辆比较 (VEHICLE_COMPARISON)、家庭充电 (HOME_CHARGING)、电池健康 (BATTERY_HEALTH)、拥有成本 (OWNERSHIP_COST)。每条记录都包含一个稳定的 ID、土耳其语查询内容、意图标识符、人类可读的土耳其语标签、建议的 TekPedal 内容路径、分配的数据集划分、语言以及来源标记。

The split is deterministic inside every class: 16 training, four validation, and four test examples. Across the full dataset, that produces 128 training records, 32 validation records, and 32 test records. 每个类别内的划分是确定性的：16 个训练样本、4 个验证样本和 4 个测试样本。在整个数据集中，这总计产生了 128 条训练记录、32 条验证记录和 32 条测试记录。

Reproducibility instead of hidden data

可复现性优于隐藏数据

All version 1 examples were generated editorially from documented subject lists and templates. They were not copied from search logs, customer conversations, support tickets, or third-party datasets. Each row is marked synthetic-editorial so the provenance remains visible during downstream use. That decision improves privacy and makes redistribution straightforward, but it also creates an important limitation: this dataset does not measure real search demand or the natural frequency of user intents. It is a controlled regression suite, not evidence about customer behavior. 所有 1.0 版本示例均由编辑人员根据记录的主题列表和模板生成。它们并非复制自搜索日志、客户对话、支持工单或第三方数据集。每一行都标记为“合成编辑”，以便在下游使用时保持来源可见。这一决定提高了隐私性并简化了再分发，但也带来了一个重要的局限性：该数据集无法衡量真实的搜索需求或用户意图的自然频率。它是一个受控的回归测试套件，而非关于客户行为的证据。

The repository includes automated checks for: the expected total row count; balanced class counts; fixed split sizes; stable and unique record IDs; required fields and allowed values; common direct-identifier patterns. This lets contributors review changes through the same validation process used for the published release. 该存储库包含针对以下内容的自动化检查：预期的总行数、平衡的类别计数、固定的划分大小、稳定且唯一的记录 ID、必填字段和允许的值、常见的直接标识符模式。这使得贡献者可以通过与已发布版本相同的验证流程来审查更改。

Quick start with Python

Python 快速入门

The JSON Lines files can be loaded directly from GitHub: JSON Lines 文件可以直接从 GitHub 加载：

import pandas as pd 
train = pd.read_json( 
    "https://raw.githubusercontent.com/gokimedia/" 
    "turkish-ev-charging-intents/main/data/train.jsonl", 
    lines=True, 
) 
print(train[["text", "intent"]].head()) 
print(train["intent"].value_counts())

For evaluation, we recommend keeping the supplied test split unchanged and reporting macro F1 together with per-class precision, recall, F1, and a confusion matrix. Any tuning performed after looking at test examples should be disclosed. 在评估方面，我们建议保持所提供的测试集划分不变，并报告宏观 F1 分数，以及各类的精确率 (Precision)、召回率 (Recall)、F1 分数和混淆矩阵。任何在查看测试样本后进行的调整都应予以披露。

What the dataset is useful for

数据集的用途

The release is intentionally compact. It works well for: prototyping a Turkish text classifier; evaluating embedding-based or zero-shot routers; building a deterministic regression suite for an EV assistant; demonstrating data-loading and classification workflows; testing which tool or content route should receive a question. It should not be treated as proof that a model is production-ready. The corpus has limited spelling-error, dialect, code-switching, multi-intent, and out-of-scope coverage. A real deployment also needs monitoring, rejection behavior, broader evaluation data, and current authoritative sources for prices, routes, taxes, charging stations, and vehicle specifications. 此次发布的数据集特意保持精简。它非常适合：土耳其语文本分类器的原型设计；评估基于嵌入 (embedding) 或零样本 (zero-shot) 的路由器；为电动汽车助手构建确定性的回归测试套件；演示数据加载和分类工作流；测试问题应分发给哪个工具或内容路径。它不应被视为模型已达到生产就绪状态的证明。该语料库在拼写错误、方言、语码转换、多意图和范围外查询方面的覆盖有限。实际部署还需要监控、拒绝机制、更广泛的评估数据，以及关于价格、路线、税费、充电站和车辆规格的实时权威来源。

Open distribution and citation

开放分发与引用

The dataset is available from several open-data services, but the project keeps two stable starting points: TekPedal Open Data for the human-readable overview and explorer; Zenodo DOI 10.5281/zenodo.22062688 for a versioned archival citation. The full documentation, datasheet, license, machine-readable files, and validation scripts are available in the GitHub repository. A rendered documentation edition is also available on Read the Docs. 该数据集可通过多个开放数据服务获取，但本项目保留了两个稳定的起点：用于人类可读概览和探索的 TekPedal Open Data；以及用于版本化存档引用的 Zenodo DOI 10.5281/zenodo.22062688。完整的文档、数据表、许可证、机器可读文件和验证脚本均可在 GitHub 存储库中找到。渲染后的文档版本也可在 Read the Docs 上查阅。

If you work on Turkish NLP, EV software, retrieval systems, or intent routing, contributions are welcome—especially carefully reviewed examples covering spelling variation, regional language, multi-intent requests, and safe out-of-scope handling. 如果您从事土耳其语自然语言处理 (NLP)、电动汽车软件、检索系统或意图路由相关工作，欢迎做出贡献——特别是那些经过仔细审查、涵盖拼写变体、区域语言、多意图请求以及安全处理范围外查询的示例。