Automatically detect unexpected changes of topics in the Twitter feeds of several leading US presidential candidates
Semantic anomaly detection is made possible by combining Numenta's Hierarchical Temporal Memory (HTM) machine intelligence platform with Cortical.io's Semantic Fingerprinting technology. Each candidate's Tweets are converted into semantic fingerprints via the Retina API and these streams of fingerprints are used as input for the HTM. The HTM then learns the speech patterns of each candidate and makes constant predictions about their future Tweets. The discrepancies between these predictions and the actual subsequent Tweets are used to calculate anomaly scores for each day, which indicate how unusual the semantic content for a day's Tweets was for that particular candidate. The graphs below show these computed anomaly scores. Spikes in the graphs indicate particularly unexpected or unusual topics for the candidate. The buttons below can additionally be used to filter the Tweets by their similarity to social and economic issues and apply anomaly detection only to those filtered Tweets.
Powered by Numenta's HTM and Cortical.io's Retina Technology