System and method for forming predictions using event-based sentiment analysis
In embodiments, a sentiment analyzer identifies a first event and accesses a first set of messages. The sentiment analyzer associates the first set of messages with the first event and analyzes the messages to identify a set of sentiment features. The set of sentiment features is used to analyze a second set of messages to form a prediction associated with a second event. The prediction may be used to facilitate an event-related service.
1. A computer-implemented method for forming and applying a prediction associated with an event, the method comprising:
building a dynamic dictionary by:
accessing, using a computing device having a processor and a memory, event information from an event information source,
identifying, using the processor and the event information, a first event, wherein the first event comprises either a gain event associated with a stock price of a stock or a loss event associated with the stock price, wherein the first event is a gain event when the event information indicates a relative increase of the stock price that exceeds a predetermined return associated with a market index, and wherein the first event is a loss event when the event information indicates a relative decrease of the stock price that exceeds a predetermined loss associated with the market index,
accessing, using the computing device, a first set of messages from a message source, wherein the first set of messages is generated by a plurality of messaging users before the first event occurred,
associating, using the processor, the first set of messages with the first event,
assigning a label to each of the first set of messages to create a first set of labeled messages, wherein the label corresponds to a predetermined sentiment value,
training a classifier using the first set of labeled messages,
determining, using the processor and the classifier and based on whether the first event is a gain event or loss event, that a text feature of each message associated with the first event includes a positive or negative sentiment, respectively, and
storing the sentiment-associated text features in the dynamic dictionary in the memory;
accessing, using the computing device, a second set of messages associated with a second event from the message source;
computing, using the processor and dynamic dictionary of sentiment-associated text, an aggregate sentiment score for the second set of messages;
predicting, using the processor and aggregate sentiment score, a positive or negative event;
taking a long position with respect to the stock if the second event comprises a gain event; and
taking a short position with respect to the stock if the second event comprises a loss event.