NUPIC for modelling user's behaviour

I wonder if it is possible to use NUPIC to model human’s behaviour. The idea is to insert into the HTM a bag of words: Key, value pairs of activity. For example:
Then I will query HTM engine with the current context i.e bag of words against the previous learned vectors. The main objective that NUBIC will start to predict the next user’s activity. I used a custom SDM implementation before to predict user’s behaviour. But I wish to use NUPIC as that SDM implementation was under NDA of a commercial product.
Please note, this sound like but I do not want any NLP over the vectors of bag of words. I will calculate the Similarity using the hamming distance.
I really wish to know how I can start to index those bag of words into NUPIC any help will be appreciated.
Kind Regards
Basel Magableh

1 Like

How does the bag of words represent user activity?

Hi Matthew,
The generic format of the bag of words is bow:[{sensors_events}, {location}, time, {activities}, {keywords}, {tags}}
for example:
sensors_events would be a dictionary of all sensors values at T for example {temp:22, speed:50}
My original idea was using the Sparse Distributed Model [Pentti Kanerva] 1988 implemented using the Random Index
What we find that the RI start to give predictions of the user’s activity and provide us with similarity between current bag of words and user’s random index
More information could be found in the patent

Check out which is an example app that does something similar with events happening on a computer.

Many thanks, I will try to use it
Kind Regards