beamandrew / cui2vec Goto Github PK

In the section 2.4 of the paper,

a CUI-CUI co-occurrence matrix is constructed, ... For nonclinical text data (e.g., journal articles), it is first preprocessed (see Section 3) and chunked into fixed length windows of 10 words, and a co-occurrence is counted as the appearance of
a CUI-CUI pair in the same window. For claims data, ICD-9 codes are mapped to UMLS CUIs and a co-occurrence is counted as the number of patients in which two CUIs appear in any 30-day period. Finally, for the clinical notes, we counted a co-occurrence as two CUIs appearing in the same 30-day ‘bin’

The co-occurance matrix created on these 3 separate sources - would you be able to kindly provide access to it? It is very powerful data-structure and can lead to further investigations (we already hold UMLS license, and if required can reach out to you privately to get the download access, if it cannot be publicly released).

Thank you

In R/word2vec_fit.R # construct_pmi()
do we really need to convert input sparse matrix into a regular matrix?
As more and more CUIs are considered, this will lead to exceeding the memory limit of the m/c.

I have made changes in my forked repo to construct pmi directly using sparse matrix:
kaushikacharya@c33492d

If you want I can create a pull request.

Hi Andrew,

Do you have a lookup dictionary for the pretrained embeddings? I saw in the embedding file, the "medical concepts" are in format of "CXXXX", not sure if they are ICD codes, procedure codes or something else.

Thanks!