Earlier this week we announced that we were providing news feeds for tracking current events. As we continue delivering this service, we started working to train a model for sentiment. Back in 2018, Jigsaw Security's CTO started working on intelligence and sentiment tracking as part of some ongoing intelligence activities. The new model being worked on now finally has enough data to be useful.
"One of the hardest things to do when tracking sentiment is understanding the use and context of how a word is being used in a written work" said Kevin Wetzel, CEO of Jigsaw Security. "This challenge is one that intelligence agencies have worked at for years, the goal in developing this model is to provide a list of word combinations that can accurately track sentiment at over 95% accuracy, of course sometimes human language evolves and we can't account for absolutely everything". The new sentiment module is useful for finding the public's view on current topics.
Some examples of how this get's tricky
When somebody says "beautiful tragedy" this could be either positive or negative, so in working through this analytic process, analyst can read a passage, mark it, and the analytic can be adjusted for real word news by actual analyst that can evaluate what is being said. Just looking at the word beautiful could be a positive, tragedy for sure would be negative but the two words together depending on how they are used could be either. It's tricky to get this right 100% of the time, but if we continually evaluate the usage, we can get extremely accurate over time.
Uses for this model
One of the biggest use is to determine if companies or individuals are being mentioned in news or articles, peer review and scientific documentation. To determine how something is viewed we have to look at the reviews and comments on a topic. As we continually evaluate from real intelligence information, we teach the analytics that things can be weighted in such a way that accuracy improved.
How can this model be used?
Determine if a story is being perceived as positive or negative on a topic
Determine over time if a companies reputation is improving or getting worse
Determine how employees view their employers
Track how an organization views a political candidate or topic
Determine how a news segment is being portrayed by the new media
Evaluate intelligence reports to determine if a situation is improving or deteriorating
Figure out if an ad campaign is effective
These are just some of the use cases and it's always fun to play with large data sets in this manner. For more information or to implement this type of model for your company please reach out to the Jigsaw Security team. The use of this analytic model requires the Jigsaw Security analytic platform as well as a feed or data source to be evaluated.