s-emanuilov/query-expansion

Query Expansion Dataset

This dataset is designed to train search query expansion models that can generate multiple semantic expansions for a given query.

Purpose

The goal of this dataset is to serve as input for training small language models (0.5B to 3B parameters) to act as query expander models in various search systems, including but not limited to Retrieval-Augmented Generation (RAG) systems.

Query expansion is a technique used to enhance search results by generating additional relevant queries. While advanced search systems often use large language models for query expansion, this can introduce latency. The purpose of this dataset is to enable the development of smaller, efficient query expander models that can perform this task without the added latency.

This query expansion dataset can be used in a variety of search system architectures, such as the one illustrated below:

The dataset serves as a key component in training query expansion models, which generate additional relevant queries to enhance the retrieval process and improve the overall performance of search systems.

This dataset is the first step. In the near future, I plan to release the trained query expander models as well.

Dataset creation

This dataset was created using a diverse set of state-of-the-art large language models. These LLMs were prompted with queries covering a wide range of topics and lengths, representing different user intents.

The choice to use multiple LLMs was made to reduce the bias that might be introduced by using a single model. The broad spectrum of topics covered and the variety of query intents (informational, navigational, transactional, commercial) ensures the dataset is comprehensive and diverse. After generation, the data underwent manual curation to ensure high quality.

Structure

Format: JSONL
Query types: Informational, Navigational, Transactional, Commercial
Each entry:

• Original query
• 3-7 semantic expansions

Example

{"query": "Best gaming laptops 2024", "expansions": ["Top rated gaming laptops 2024", "Highest performance gaming computers", "Premium gaming notebook recommendations", "Best value gaming laptops this year"]}

Usage

from datasets import load_dataset
dataset = load_dataset("s-emanuilov/query-expansion")

Limitations and alternative approaches

While this dataset provides a valuable resource for training query expansion models, it's important to note that alternative approaches, such as thesaurus-based methods or using large language model APIs, may be more suitable depending on the specific use case and requirements.

License

This dataset is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).

Citation

@misc {simeon_emanuilov_2024,
    author       = { {Simeon Emanuilov} },
    title        = { query-expansion (Revision e58bb74) },
    year         = 2024,
    url          = { https://huggingface.co/datasets/s-emanuilov/query-expansion },
    doi          = { 10.57967/hf/3881 },
    publisher    = { Hugging Face }
}