Query Classification Pipeline

Scenario

We're building a query intent classifier for a search engine. The classifier will be used in production to understand what users are looking for — enabling downstream task like result reranking.

We have a dataset of search queries. For each query, we also have the top search results (title, snippet, and URL) from our search engine. The goal is to build a classifier that assigns an intent category (sports, technology, shopping) to any incoming query.

You've been handed this codebase with a first attempt at solving this problem. Your task is to work with it, understand it, and improve it.

Input format

Each record in data/queries_with_results.json represents a single query with its search results:

{
  "query": "best running shoes 2025",
  "results": [
    {
      "title": "Top 10 Running Shoes for 2025",
      "snippet": "We tested dozens of running shoes to find the best options...",
      "url": "https://www.runnersworld.com/gear/top-running-shoes-2025"
    }
  ]
}

Categories

• sports - Queries related to sports, athletics, competitions, fitness
• technology - Queries related to software, hardware, programming, IT
• shopping - Queries related to buying, deals, price comparisons

How to run

There are two versions of the pipeline. Use whichever you prefer:

Python script:

uv run main.py

Jupyter notebook:

uv run jupyter lab pipeline.ipynb