- The document describes a method for understanding city traffic dynamics by utilizing sensor data that measures average speed and link travel time, as well as textual data from tweets and official traffic reports. - It builds statistical models to learn normal traffic patterns from historical sensor data and identifies anomalies, then correlates anomalies with relevant traffic events extracted from tweets and reports. - The method was evaluated on data collected for the San Francisco Bay Area, and it was able to scale to large real-world datasets by exploiting the problem structure and using Apache Spark for distributed processing. Events extracted from social media provided complementary information to sensor data for explaining traffic anomalies.