Interested in the similarity(phrase1, phrase2) function

[Michelle Shumate]

Hi,
I am working on a project that I think could benefit from the WikiBrain platform. I'm interested in the similarity(phrase1, phrase2) function in particular. I am interested in using the AtlasifySr+E (or the ensemble measure of similarity). However, I ran across an issue in the tutorial. On http://shilad.github.io/wikibrain/tutorial/sr.html it states:

The inlink metric is a fast but relatively inaccurate SR metric. You can also build the "ensemble" metric that provides a linear combination of four other metrics. Beware that training the ensemble is costly. It takes about 10 minutes on Simple English Wikipedia, and a little over a day on the full Wikipedia. Most of the model-building time supports the mostSimilar() call, so you can speed up model building if you only need similarity(). TODO: explain how to do this.

I am hoping someone can help explain how to do this.

Michelle Shumate
Associate Professor
Northwestern University

[Shilad Sen]

Michelle,

Yes! I've been meaning to do this for some time. 

Unfortunately, the optimization options are fairly complicated, at the moment. Can you tell me a little more about your exact use case. What kind of performance do you need? Do you have a predefined set of phrases, etc.

Thanks for your interest in WikiBrain!

-Shilad

--
You received this message because you are subscribed to the Google Groups "wikibrain" group.
To unsubscribe from this group and stop receiving emails from it, send an email to wikibrain+...@googlegroups.com.
To post to this group, send email to wiki...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/wikibrain/644c9fa7-aa19-4be0-9135-29788f32636e%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Shilad W. Sen
Associate Professor
Mathematics, Statistics, and Computer Science Dept.
Macalester College
ss...@macalester.edu

http://www.shilad.com

https://www.linkedin.com/in/shilad

651-696-6273

[Michelle Shumate]

Hi Shilad,
Sure. We are having participants in an experiment generate concept maps for organizations. Depending upon their condition, they will get different pairs of nonprofit and corporations. We want to use the measure to assess how well the concept maps of the two organizations fit. We would generate all the word pairs and submit them to the similarity function.

Here's the write up of how we hope to do it (attached).

Happy to chat more about the project.

BTW: Darren Gergle has offered to let us use his setup in the CollabLab, but he deferred to you guys since you have been working developing WikiBrain for the questions.

Michelle