Machine learning for human rights advocacy: Big benefits, serious consequences by Megan Price.
Megan is the executive director of the Human Rights Data Analysis Group (HRDAG), an organization that applies data science techniques to documenting violence and potential human rights abuses.
I watched the video expecting extended discussion of machine learning, only to find that our old friend, record linkage, was mentioned repeatedly during the presentation. Along with some description of the difficulty of reconciling lists of identified casualties in war zones.
Not to mention the task of estimating casualties that will never appear by any type of reporting.
When Megan mentioned record linkage I was hooked and stayed for the full presentation. If you follow the link to Human Rights Data Analysis Group (HRDAG), you will find a number of publications, concerning the scientific side of their work.
Oh, record linkage is a technique used originally in epidemiology to “merge*” records from different authorities in order to study the transmission of disease. It dates from the late 1950’s and has been actively developed since then.
Including two complete and independent mathematical models, which arose because terminology differences prevented the second one from discovering the first. There’s a topic map example for you!
Certainly an area where the multiple facets (non-topic map sense) of subject identity would come into play. Not to mention making the merging of lists auditable. (They may already have that capability and I am unaware of it.)
It’s an interesting video and the website even more so.
Enjoy!
* One difference between record linkage and topic maps is that the usual record linkage technique maps diverse data into a single representation for processing. That technique loses the semantics associated with the terminology in the original records. Preservation of those semantics may not be your use case, but be aware you are losing data in such a process.