DeepSearch

DeepSearch is an experimental project designed for embedding-based text similarity search in web pages. It utilizes the Xenova/multilingual-e5-large model for feature extraction and computes cosine similarity to identify relevant text segments.

Demo

Overview

DeepSearch allows users to search for semantically similar text within web pages using embeddings. By leveraging WebGPU for fast processing, the tool dynamically highlights matching text segments and enables easy navigation through search results. It is primarily designed as a Chrome extension for enhanced browsing experience.

Installation

# Clone this repository
git clone https://github.com/your-repo/DeepSearch.git
cd DeepSearch

# Install dependencies
npm install

# Build the project
npm run build

How to install?

1. Open Chrome and navigate to chrome://extensions/.
2. Enable Developer mode (toggle in the top right corner).
3. Click Load unpacked and select the build/ directory inside the DeepSearch project.
4. The extension should now be installed.
5. Open a web page and press Ctrl + Shift + F to activate DeepSearch.

How to use?

1. Open a web page.
2. Press Ctrl + Shift + F to activate the search.
3. Enter a query to find semantically similar text.
4. Navigate through highlighted results.

Known Issues & Limitations

• Similarity scores tend to cluster within a narrow range.
• Performance may vary depending on WebGPU support.
• Large web pages with excessive text may lead to high memory usage.