Technical Details & Modelling Methodology
databright's politica platform is a first in real-time monitoring and modelling of public discussion and discourse.
System & Data
We collect and compile data in real-time from various online sources, feed that data into our models, and annotate the records with valuable predictions and events.
In doing this we are able to uncover latent, highly valuable, information which is often difficult to observe using more traditional methods.
As the system ingests tens of thousands of new records each day we are able to learn and draw conclusions from samples with a far larger size than conventional techniques.
Models
At present we have three core models: Sentiment, Topic, and Connotation. The foundations of the models have been developed using the latest in machine learning techniques while the structure is comprised of databright's proprietary novel innovations.
The majority of our data comes from Twitter as it is a great source of volumous, and fairly reliable, real-time data. However we also consume data from a variety of additional sources.
We use deep learning coupled with natural language processing techniques and proprietary heuristics to generate these state-of-the-art predictive models.
Our models perform at or above the state-of-the-art performance levels currently found in literature for these types of natural language applications.
Applications
We are currently applying our deep learning platform to the Canadian political space, however it has been designed for easy expansion into other application domains beyond politics.
We currently have plans to apply the platform to the Brexit situation in the United Kingdom including any upcoming general elections, as well as market research campaigns for our clients.
Please don't hesitate to contact databright with any questions or comments you may have.
We are always looking for ways to improve our models as well as partnerships with organizations in both the private and public sectors.