Datasets:
license: cc-by-nc-4.0
task_categories:
- token-classification
task_ids:
- named-entity-recognition
language:
- en
- es
pretty_name: SearchQueryNER-100k
size_categories:
- 100K<n<1M
tags:
- token-classification
- named-entity-recognition
- search-queries
dataset_info:
features:
- name: id
dtype: int64
- name: tokens
sequence: string
- name: ner_tags
sequence:
class_label:
names:
'0': O
'1': B-WORK_P
'2': I-WORK_P
'3': B-COUNTRY
'4': I-COUNTRY
'5': B-PHONE
'6': I-PHONE
'7': B-CITY
'8': I-CITY
'9': B-REL
'10': I-REL
'11': B-WORK_S
'12': I-WORK_S
'13': B-PER
'14': I-PER
'15': B-ORG
'16': I-ORG
'17': B-ATTR
'18': I-ATTR
'19': B-STREET
'20': I-STREET
'21': B-CITY_PART
'22': I-CITY_PART
'23': B-REGION
'24': I-REGION
splits:
- name: train
num_bytes: 6351748
num_examples: 102931
- name: test
num_bytes: 1252122
num_examples: 20301
- name: validation
num_bytes: 1266344
num_examples: 20420
download_size: 2028140
dataset_size: 8870214
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
- split: validation
path: data/validation-*
train-eval-index:
- config: searchqueryner-100k
task: token-classification
task_id: entity_extraction
splits:
train_split: train
eval_split: test
col_mapping:
tokens: tokens
ner_tags: tags
metrics:
- type: seqeval
name: seqeval
Dataset Card for SearchQueryNER-100k
This dataset card describes SearchQueryNER-100k, a dataset of 100,000 annotated search queries designed for Named Entity Recognition (NER) tasks.
Dataset Details
Dataset Description
SearchQueryNER-100k is a high-quality dataset of search queries originally collected to gain insights into how users find the website putazon.com. The data has been carefully curated and annotated for NER tasks, making it ideal for training and evaluating models in search query understanding and entity extraction.
The dataset features:
13 entity types (e.g., B-WORK_P, B-CITY, B-ATTR, etc.).
A total of 143,652 search query rows filtered from 200,000 raw queries exported from Google Search Console.
Automatic annotation guided by hand-annotated examples and a fine-tuned Gemini-1.5-flash model.
Data split into training, validation, and test sets.
Curated by: Putazon Team
Language(s) (NLP): English (en), Spanish (es)
License: CC BY-NC 4.0
Uses
Direct Use
This dataset is intended for training and evaluating token classification models for Named Entity Recognition (NER), particularly in the context of search queries. Example tasks include:
- Understanding user intent.
- Extracting structured data (e.g., names, locations, organizations).
- Optimizing search engines and improving SEO strategies.
Out-of-Scope Use
- Commercial use is prohibited under the CC BY-NC 4.0 license.
- Using the dataset for tasks unrelated to NER or search query analysis may result in suboptimal performance.
Dataset Structure
The dataset includes the following fields:
| Field | Type | Description |
|---|---|---|
id |
int64 | Unique identifier for each search query. |
tokens |
sequence | List of tokenized words in the query. |
ner_tags |
sequence | Corresponding NER tags for each token. |
Example Data
{'id': 42, 'tokens': ['scor'], 'ner_tags': [11]}
{'id': 43, 'tokens': ['mujer', 'mexico'], 'ner_tags': [11, 3]}
{'id': 44, 'tokens': ['sussy', 'acompañante'], 'ner_tags': [13, 11]}
{'id': 45, 'tokens': ['lili', 'mujer'], 'ner_tags': [13, 11]}
{'id': 46, 'tokens': ['cscorts', 's', 'c', 'o', 'r', 't', 'mexico'], 'ner_tags': [1, 11, 12, 12, 12, 12, 3]}
Dataset Creation
Curation Rationale
The dataset was created to understand how users find putazon.com through search queries. By annotating these queries, we aimed to enhance tools for entity extraction and user intent recognition in search-based applications.
Source Data
Data Collection and Processing
- Data was exported from Google Search Console, containing ~200,000 distinct search queries along with impressions, clicks, and average position.
- Irrelevant or unsuitable queries were removed using internal SEO tools, leaving 143,652 rows.
- Initial 500 rows were hand-annotated to define label categories and guide annotation processes.
- GPT-4o pre-annotated 2,500 rows based on the hand-labeled data.
- Gemini-1.5-flash was fine-tuned on this data to automate labeling for the remainder of the dataset.
Who are the source data producers?
The search queries were sourced from anonymous users finding putazon.com through Google.
Personal and Sensitive Information
Efforts were made to remove any sensitive or personal data during the filtering and annotation processes. For example, all phone numbers were replaced with random digits from the 3rd digit onwards.
Bias, Risks, and Limitations
- Bias: The dataset may reflect biases inherent in user search behavior or the curation process.
- Risks: Misinterpretation of entity labels in contexts beyond search queries.
- Limitations: The dataset is focused on a specific domain and might not generalize to unrelated tasks.
Recommendations
- Use this dataset as a foundation to train smaller models for local environments (e.g., BERT).
- Cross-validate results with other datasets for improved generalization.
Citation
BibTeX:
@misc{searchqueryner-100k,
title = "{SearchQueryNER-100k}: A Dataset of 100,000 Annotated Search Queries for Named Entity Recognition",
author = "Putazon Team",
year = "2025",
url = "https://huggingface.co/putazon/searchqueryner-100k",
note = "Dataset for Named Entity Recognition in search queries, featuring 13 entity types. Licensed under CC BY-NC 4.0.",
howpublished = "Online resource",
website = "https://putazon.com"
}
More Information [optional]
For further details, please visit the dataset repository.
Dataset Card Authors [optional]
This card was created by the Putazon Team with assistance from automated tools.
Dataset Card Contact
For questions or feedback, please begin a conversation in this repository.