No context in vocabulary when setting an positive bow_size

[MichaelZhouwang]

Hi,
When I set an positive non-negative window_size, the contexts extracted are done by method 'get_neighbors' and the words returned are just words, without context prefix such as 'acompI_', which results in no contexts in vocabulary because the context vocabulary are all prefixed. Is the model only built for syntactic dependency contexts? Or I could simply average the context embedding for all syntactic to obtain an 'average context embedding' ? Look forward to your reply. Thanks !

[orenmel]

When using bow>0, the code expects to get standard word embeddings, i.e. without syntactic prefixes. You could try to do averaging over prefixes or simply learn a standard word2vec model and dump the target and context embeddings into two separate files. I think you can find instructions on how to do that on the web.

[MichaelZhouwang]

When using bow>0, the code expects to get standard word embeddings, i.e. without syntactic prefixes. You could try to do averaging over prefixes or simply learn a standard word2vec model and dump the target and context embeddings into two separate files. I think you can find instructions on how to do that on the web.

Thank you very much!