Open-sourcing PyTorch-BigGraph for faster embeddings of extremely large graphs(ai.facebook.com)
ai.facebook.com
Open-sourcing PyTorch-BigGraph for faster embeddings of extremely large graphs
https://ai.facebook.com/blog/open-sourcing-pytorch-biggraph-for-faster-embeddings-of-extremely-large-graphs
7 comments
Don't see anything but my main problem is finding a good open source graph engine (NEO4J level maturity or close) that I can install locally and develop my application on and that allows me to spin up multiple graphs on the same machine easily.
Your use case might be different than what is being offered here. This isn't meant to allow SPARQL to query nodes and edges.
This provides embeddings that follow graph relationships to improve correlations between entities when looking through text. For example, if you use word2vec you will see that terms in the same context appear in similarly trained vectors...but those vectors are lacking formal relationships and just go off where they are likely to appear in text. A good use case for this might be Entity Linking / Disambiguation, when performing Named Entity Recognition on unstructured text...such as knowing that William Shatner and Bill Shatner are the same person, and 'She drank a Manhattan' is referring to the cocktail and not the city.
This provides embeddings that follow graph relationships to improve correlations between entities when looking through text. For example, if you use word2vec you will see that terms in the same context appear in similarly trained vectors...but those vectors are lacking formal relationships and just go off where they are likely to appear in text. A good use case for this might be Entity Linking / Disambiguation, when performing Named Entity Recognition on unstructured text...such as knowing that William Shatner and Bill Shatner are the same person, and 'She drank a Manhattan' is referring to the cocktail and not the city.
You could have a look at ArangoDB (https://www.arangodb.com), a multi-model database - it may cover your use cases. Hopefully linking is ok in comments, will remove if its breaks community rules - no affiliation etc..
binarymax in another comment is correct about what PBG does. Still, I am interested in combining deep learning and Neo4J (and possible RDF data stores in the future) and I am experimenting with some code for a book project.
BTW, I am running their code that produced the article results (on small freebase data set) right now and also studying their code.
BTW, part 2: I know that Facebook and Google get a lot of criticism over privacy issues but the flip side of the privacy coin is how generous they are in sharing results, pre-trained models, etc. (even given that publishing great results, data, and code is a great way to attract more potential employees).
BTW, part 2: I know that Facebook and Google get a lot of criticism over privacy issues but the flip side of the privacy coin is how generous they are in sharing results, pre-trained models, etc. (even given that publishing great results, data, and code is a great way to attract more potential employees).
look at word octavian(https://www.octavian.ai/) is doing
Thanks! Interesting stuff.