petrarch

petrarch copied to clipboard

Hi,

We have been running Hypnos with good success. Thank you very much. However, it seems the majority of events Hypnos returns are classified as Political, as far as the cameo codes. Is there a reason for this or do you think we am doing something wrong? If this is known functionality, would there be a way for me to add all cameo codes and therefore all event types?

Thanks again, Scott

The coding dictionaries for PETRARCH were derived from earlier NSF-funded work in political science that was generally focused initially on international mediation, and later on conflict forecasting. Consequently some parts of the ontology (particularly those dealing with political violence and international negotiation) have had extensive development, but other parts have had very little. The dictionaries, of course, are open source so you can easily add additional vocabulary, and we've also very recently started a new NSF-project that will be extending CAMEO (e.g. to democratic processes such as elections and budget debates) but those new dictionaries will probably not be available for at least six months.

First, a question. You're using PETRARCH and not PETRARCH2, correct? Second, echoing @philip-schrodt's comments, there tends to be a pretty heavy skew towards the "verbal" CAMEO codes. You can see that in the attached plot.

Thank you both for the very helpful comments. I have a couple of follow up questions.

Philip, When you mention open source dictionaries of additional vocabularies. If we look to add these, would this likely produce a more balanced distribution of event classifications? I am sorry I am not familiar with these dictionaries and how they integrate with Petrarch.

John, we are using the version of Petrarch that comes with Hypnos that we installed a few months ago. I am not sure which version of Petrarch this runs. If it uses the initial version and we were to use version 2, would this improve the distribution?

Thanks again.

There's only a limited amount that can be done to "improve" the distribution. To a large degree it is what it is; people make statements far more often than people engage in combat. Phil's point about some dictionary categories being less fleshed out is salient, though.

Finally, on PETR vs. PETR2, PETR2 would actually skew things more heavily towards the lower-level categories due to some things it picks up differently.

Unfortunately, the additional dictionaries that are available at the moment from ICEWS only cover actors, not events (their event dictionaries and coding engine are apparently considered proprietary even though this was all done with public money...some project manager apparently wasn't paying attention...). So those won't change the distribution much.

What some people have done on other projects is enhance the dictionaries so that they pick up existing types of behavior they are specifically interested in (e.g. one project was working just on piracy as I recall). That could work provided you need to monitor a fairly limited amount of behavior.

[hmmm, looks like I'm currently logged in on github as civet-software rather than philip-schrodt.]

Thank you both for the feedback and suggestions. It is very helpful. I will let you know if we have any follow-up. I think at this point we may go the route of enhancing the dictionaries. Although I'm not sure when we will get to it. Thanks again.