GroupThink

GroupThink observes the behavior of the ups and downs of the stock market towards the emotional sentiment of real-time tweets.

Using the S&P 500 (SPX) index performance from Aug 10th to Aug 27th stock market correction, we observe the change of emotional sentiments of StockTwit tweets in minute to minute detail. We look at two sets of emotional factors calculated by IBM Watson - the first set being analytical, confidence, tentative, and the second set of data being pure human emotion of cheerfulness, anger, and negativity. The messages were aggregated from pieces of real-time user responses on the S&P 500 stock, taken daily from StockTwits.

What's particularly interesting is the sudden spike of the "anger" emotion in tweets during the August 24th Dow 1000-point drop on Monday. That emotion, however, quickly subsided as Tuesday came.

This project was intended to be a short and simple case study on whether IBM Watson could possibly accurately detect emotions conveyed through text, and whether this idea of emotional sentiment detection could be applied to analyzing real-time feedback of real users.

A few roadblocks were hit while developing this project.

StockTwits limits 30 user messages per request to their stock ticker discussion API. Not only that, only 200 requests can be made per hour, meaning, only a total of 6000 individual tweets could be harvested within an hour, making large-scale harvesting of data difficult. We chose the S&P 500 index since this indice contains a good variety of large-cap U.S. equities and aggregates them. Since only allowing to harvest minimal number of indexes at a time due to API limitation, choosing S&P 500 was obvious.

Next, IBM Watson's Tone Analyzer API requires a self-hosted Node.js application. Due to time limitations, we decided to use the Tone Analyzer demo API page instead to do our data analysis.

Built With

• css
• javascript
• laravel
• php
• python