brand_embedding

The repo maps the similarity or closeness between different commercial brands. It builds a brand embedding which can be queried to find the most similar or the least distant instances representing similar brands.

Download Glove Embeddings

Download Glove word embeddings from glove.6B.zip to brand_embedding/data/glove_embeddings.

The file structure should look like: brand_embedding/data/glove_embeddings/glove.6B.200d.txt

Methodology

• release 1: It is a simple technique which takes the mean of the Glove word embedding for each word present on the landing page of the brand in Wikipedia. The idea is that similar brands talk about overlapping concepts. Working with word embeddings frees us from worrying about the exact match in jargon or vocab.

How to build the brand embeddings?

Use the build function in brand_embedding/build_emb.py scrip to query Wikipedia iteratively and generate the brand embeddings for the set of brands passed as an argument via set_brands. The default set of brands are as listed in brand_embedding/data/brand_list which is loaded as part of the module __init__. The listings in brand_embedding/data/brand_list must match the brand name as on Wikipedia otherwise ambiguous results may arise.

How to query the brand embeddings?

Use the query or the query_list function in brand_embedding/query_emb.py to get most similar brands to a particular brand or get the same for a list of brands. The top_n parameter can be set to only return the closes n brands.

Examples

• Building the embeddings:

cd brand_embedding

python brand_embedding/build_emb.py build

This should iterate through the brand list giving out logs as pages are processed from wikipedia

04/21/2018 21:35:04:INFO: building knowledge base
04/21/2018 21:35:14:INFO: Manchester United: https://en.wikipedia.org/wiki/Manchester_United_F.C.
04/21/2018 21:35:15:INFO: Honda: https://en.wikipedia.org/wiki/Honda
04/21/2018 21:35:16:INFO: Los Angeles Lakers: https://en.wikipedia.org/wiki/Los_Angeles_Lakers
04/21/2018 21:35:18:INFO: Gap Inc.: https://en.wikipedia.org/wiki/Gap_Inc.
04/21/2018 21:35:19:INFO: Ford Motor Company: https://en.wikipedia.org/wiki/Ford_Motor_Company

and towards the end:

04/21/2018 21:36:05:INFO: saving knowledge base to: `/home/bluefire/workspace/projects/brand_embedding/brand_embedding/data/brand_emb.json`
04/21/2018 21:36:05:INFO: knowledge base compiled

The embeddings would be saved to the fpath_save path in the build function.

• Querying the embeddings for similarity:

cd brand_embedding

python brand_embedding/query_emb.py query --target_brand_name=Toyota --top_n=5

["Honda", 0.5642089940473477]
["Ford Motor Company", 0.5905087678182305]
["General Motors", 0.6618795641050003]
["Suzuki", 0.7336843346403088]
["Chevrolet", 0.8425528515308258]

cd brand_embedding

python brand_embedding/query_emb.py query_list --list_target_brand_name='["Liverpool F.C.", "Google"]' --top_n=5

Google:         [["Facebook", 0.563797545845138], ["Amazon Inc", 0.634932045374182], ["Apple Inc", 0.6743377279657304], ["Gap Inc.", 1.0265872931260083], ["Nestle", 1.0296603447159702]]
Liverpool F.C.: [["Chelsea F.C.", 0.3479228805034268], ["Manchester United", 0.37073191660239563], ["Arsenal F.C.", 0.41666102106971], ["Cincinnati Bengals", 1.1081518545348459], ["Los Angeles Rams", 1.171184292207359]]