readme.md

Universal Sentence Encoder Clustering with DBSCAN

This code provides an implementation of clustering text data using the Universal Sentence Encoder (USE) and the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm.

You can provide a list of queries (sentences or words) and it will cluster them for your SEO needs (or other use cases) in groups using this NodeJS repo.

Requirements

This code requires the following packages to be installed:

• @tensorflow/tfjs-node
• @tensorflow-models/universal-sentence-encoder
• density-clustering

If you run npm install or pnpm install the repo will install all the necessary dependencies.

Usage

To run the clustering on your own data, modify the kws.js file to contain your desired keywords, then run the script using Node.js:

node ./src/index.js

Output

Modes

There are three different output modes:

• logs: Run pnpm run logs and the output will be printed in the console
• csv: Run pnpm run csv and the output will be safe as a csv in out.csv
• text: Run pnpm run txt and the output will be safe as a txt in out.txt

Logs example:

9app:
  - 9app
  - 9apps

adobe flash player descargar:
  - adobe flash player descargar
  - descargar adobe flash player

among us download pc:
  - among us download pc
  - among us pc download

anydesk descargar:
  - anydesk descargar
  - descargar anydesk

anydesk download:
  - anydesk download
  - download anydesk

anydesk download free:
  - anydesk download free
  - anydesk free download

To have the output in a file you can use this shell command

node ./src/index.js > out.txt

The output will display the updated data with cluster assignments and a list of clusters with their corresponding keywords. The output will also display any noise points that were not assigned to any cluster.

Note: This is the MVP version of the clusterizer

License

This code is licensed under the MIT license. See the LICENSE file for more details