This pipeline was used in part for predicting how voters felt about the different candidates during the 2016 United States Presidential Election. A flat file of raw tweets in JSON format (a very small subset from the original paper) is passed (line by line) to a sentiment prediction model written in Keras, which is then stored in a flat file.
This model also works well with a streaming Twitter API connection, so if you want to try that out, you can change this config to have a "StreamingAPI" type data source and feed it through the same pipeline. Just make sure to remember to include your API credentials! We also included a tweets preprocessor so you can see how easy it is to clean data in the pipeline in streaming fashion.