Document Influence Model

[bhargavvader]

Document Influence Model is a modelling technique described here which allows us to identify which documents most influenced a topic.

The C++ implementation for the same is here, same place where the Dynamic Topic Modelling code is.

During my implementation of python DTM, I sort of 'set up' the code for DIM.
For example, in line 249 of the DTM code, if the model type was DIM, it would call the appropriate method for DIM. The methods are usually quite similar to the DTM version, which was translated to python from the C++ code. By referring the C++ code and the DTM python code it's possible to implement DIM by simply coding up and plugging in the appropriate methods with a few changes.

As a bonus, if one can find a better way to vectorise or speed up the DTM code while doing this, it'll be awesome!

Would be happy to discuss ideas or ways to go about doing this on this thread if anyone wishes to take this up.

[bhargavvader]

@tmylk , could you add the appropriate labels for this?

[anmolgulati]

I would like to take this up.
@bhargavvader Yes you have already setup many of the method calls for DIM. I'll start with first implementing methods for DIM. Then we could look at speeding up DTM. Where exactly do you think it could be optimised?
[Edit]: If someone else wants, could take this up. I'm trying to work on some other issue presently.

[bhargavvader]

@anmol01gulati , have a look at #840 and tell me if anything interests you!

[mjawa]

I am taking a shot on this.

[bhargavvader]

Awesome!

[bhargavvader]

@mjawa , any updates?

[mjawa]

It would be another two weeks before I can take a stab on this.

[devashishd12]

@mjawa any updates on this? Can I take this up?

[piskvorky]

@dsquareindia please go ahead, @mjawa excused himself due to private issues.

[bhargavvader]

Awesome, can't wait to see the PR, @dsquareindia 😉

[kris-singh]

I would work on this if its okay. But i need a to read the paper first. Any other blog that expalins this.

[bhargavvader]

What exactly do you want explained?

[walfly]

Is anyone working on this?

[menshikh-iv]

Ping @dsquareindia @bhargavvader @kris-singh, what's status here?

[bhargavvader]

I'll be glad to help anyone willing to do this.

Note: refer to #840 as well for more links and information.