FantasticSearch

Automatic Search

1. Write a simple description of the product you want to find in the "Search query" field (large input box).
2. Check the "AI assisted filters" checkbox.
3. Start the search by clicking the "Search" button.
4. Below the search bar, results and recommended filters will appear. If you want to refine your search using custom filters, you can repeat the search with manual settings.

Filtering Rules

• Positive filter: A word that the product description must include.
• Negative filter: A word that the product description must not contain.
• If a product description contains at least one negative filter or does not contain all positive filters at the same time, it will be excluded from search results.
• Category filter: Filter by "Category" field.

Manual Search

1. Modify the product description if necessary.
2. Uncheck the "AI assisted filters" checkbox.
3. Remove any filter by clicking on its badge.
4. Remove all positive filters using the "Reset positive" button.
5. Remove all negative filters using the "Reset negative" button.
6. Add multiple filters by typing them separated by spaces in the "Input filter word" field (small input box), then click "Add positive" or "Add negative".
7. Start the search by clicking "Search" (make sure the "AI assisted filters" checkbox is unchecked).

Find Similar

Search Results

1. Click on the product ID (e.g., B0CQVRZTXD) to go to its actual Amazon product page.
2. Information about prices, availability and ratings may not be current.

Search Tips

1. If search results match the category but not the exact product, try modifying the filters and using manual search.
2. If you got related products in output (e.g., searching for a phone but getting phone cases), negative filters (e.g., "case", "cases", "cover", "covers") can help.
3. Be cautious with filters to avoid excluding the right product.
4. Negative filters can include synonyms of words that must be excluded.
5. If you know the exact manufacturer, name, model, or unique number, add them to positive filters. If a product description likely includes a specific number or letter-number combination, adding it to filters helps refine results.
6. Avoid adding synonyms in positive filters (e.g., "football", "soccer") because any single synonym might exclude matching results.
7. If search results do not match your intent, revise the main query and try automatic search again.

Examples

1. Good "Exact" simple query with good manual filters
2. Good "Product type" simple auto query with good optional manual filters
3. Good non-English "Brand" simple auto query with good optional manual filters
4. Good "Use Case" simple query
5. Good "Symptom" simple auto query with good optional manual filters
6. Complex query with mutually exclusive synonyms
7. When additional manual negative filters would be useful
8. When positive AI filters must be removed manually
9. Range filters
10. Variation filters (like shoes size)
11. Ambiguous query must be rewritten

About

This project was created to test the capabilities of product search for e-commerce using incomplete and unstructured data, AI, and various database options. The project utilizes the 2023 dataset (https://huggingface.co/datasets/McAuley-Lab/Amazon-C4), so some data may be outdated. The search covers 1,058,417 products listed on "Amazon.com" in 2023, with data preserved in the specified dataset. Currently, some of these products are no longer available for sale.