This repo references https://github.com/acocos/business_embeddings, updates and fixed bugs from the original repo
This repo contains code used to generate business embeddings for the Yelp Academic Dataset as detailed in this blog.
| file/directory | description |
|---|---|
src/pipeline.sh |
This script demonstrates how to extract business/context pairs from the Yelp data and use them to train word embeddings using word2vecf. You can run this pipeline (after downloading the Yelp data and installing word2vecf, see below). |
src/extract_contexts.py |
Generates business/context pairs from the Yelp data |
src/infer.py |
Script from the original word2vecf code, useful for loading and manipulating the resulting vectors |
src/examine_places.py |
Script used to produce results given in blog post |
data/ |
Download the Yelp data and extract it to this directory |
data/processed |
If you run pipeline.sh, the resulting vectors will end up here. Or you can download them and put them there on your own. |
- The code in this repo depends on the
word2vecfadaptation of the popularword2vecsoftware, allowing the use of arbitrary contexts to train vectors. It was developed by researchers at Bar-Ilan University and is available here. You'll need to download and install before running the pipeline to train the vectors on your own. - If you want to train your own vectors, you'll also need to download the Yelp data and extract it to the
./datadirectory. - Once you have done those two things, you will be able to run
src/pipeline.shto generate your own business embeddings.