Named Entity Recognition is some which is helping lot of developers to get around context based entity identification. Entity can be any thing - First Name, Last Name , Location, It can be combination of certain pixels and more.
This is now evolved a lot to recognize like our brains using DNN, RNN, over LTSM and BIDIRECTIONAL LTSM and all that
Can anybody in this community put help on using nupic to use NER. In my opinion - Nupic is the solution for many future ready applications and we want to see Nupic in helping community to recognize Named entities of our choice. Ofcource we should teach the system with samples about Entity first. But, can we use NuPic if so how ?
Appreciate your help ipython or something to start with.
Can you explain this statement a little better? I’m not sure I see how NER and HTM would be used together. Does NER process language as if it were a temporal stream?
If I understand NER correctly, it is essentially classification/labeling of words in a sentence (sentences can be represented as a temporal stream of words).
HTM could potentially be applied to the first part of this problem to generate representations for words in the context of the sentence they are in. One approach would be to run word SDRs (such as from cortical.io) through the temporal memory process.
But the desired output of NER is not prediction or anomaly detection. It is classification of the word in context. This would require a classifier which takes the SDR output of temporal memory and labels it (as “person”, “organization”, “location”, etc). I’m not familiar enough with NuPIC myself to describe how much development would be required for this step.
My sense is that HTM is not particularly well suited to this type of problem, but I’ll let others comment who might have a different view.
Thanks for the explanation. I could see where an HTM might help identify contextual meaning of words in sentences in combination with cortical.io’s semantic fingerprints, but this is an old idea that really requires temporal pooling of sequences.