Test Results on Different Glove Embeddings - For Interest

[kpoots]

I generated my own Glove embeddings, but restricted those embeddings to only be what is in my dataset vocabulary. The results were worse it seems, although I did get something for queries which previously didn't have any results.

It seems that using an off-the-shelf collection of Glove embeddings - effectively - adds to the vocabulary of the queries. This means that Glove expansion is somewhat dependent on the collection of Glove embeddings - where it came from. This is interesting. Might lead to a variation on expanders. If you find a giant collection of Glove embeddings, the results could improve.

Ok, I can dream of electric sheep https://en.wikipedia.org/wiki/Do_Androids_Dream_of_Electric_Sheep%3F . And giant collections of Glove embeddings.

Sorry its late, and I'm tired, but I am trying.