This study at the following link is a very interesting study, which was conducted by University College London and published as a whitepaper last week.
Link to study found here:
In this study, the researcher (Thársis T.P. Souza, studying under professor Dr. Tomaso Aste at University College London Computer Science Department) investigates using some deeper methodologies than previous studies to find out exactly why and how the online chatter from Twitter, StockTwits and private institutional chatrooms can consistently predict the market's behavior. The groundbreaking study is important because it goes beyond previous studies which confirmed weak causality, but were previously unable to explain the reasons why the sentiment analysis was predicting the market behavior. The reason for these previous limitations were because previous studies were limiting their scope to linear activity only, and failing to use more sophisticated methodologies which this study employs and describes.
For reference, the way that PsychSignal's technology works, is by applying a proprietary Natural Language Processing Engines to determine the bullishness and bearishness of the trading public, using the raw 100% firehose feeds from multiple sources like Twitter. The PsychSignal dataset consistently outperforms all other sentiment analysis products because it delivers simultaneous Bull and Bear, to 2 decimal places on a scale of 0 to 4, and can be understood using non-linear causal relationships as investigated in this paper.
For those seeking to learn more and try out market sentiment using PsychSignals datafeed, are encouraged to try it out with a starter kit feed through Quantopian here at the following LINK: