Sometimes it can be hard to visualise what text analytics can really mean - a lot of the time our brains seem to stop at keyword counting. Here's one way of taking things a little further.
A client had an interest in determining if relationships existed between various corporate entities. We can easily see applications for this type of research; a simple example might be M&A.
As part of the research phase we read select sections of newspapers for a period of 2 years. That is, we read the papers from 2 years ago until the day in question and ended up with some 20,000 stories.*
Named Entity Extraction
Given the nature of the brief we then extracted the named entities referenced in each story.
So for each story we now had entities and the idea of a relationship between those entities. We didn't put any thought into what the nature of that relationship was, we just took it as sufficient that a journalist thought to mention the entities together. Then, over periods of time we could look to see the evolution of those relationships.
NER - as it's known - is by no means a perfect process, but the results can be surprisingly good, particularly where you have a defined set of counterparties you know you are interested in.
The graphs below should give you an insight as to how powerful the analytics can be.
* That's no mean feat by the way - particularly if you want to do it for free.