Societe Generale Global Solution Centre (SG GSC) hosted the second edition of its annual hackathon, Brainwaves 2015, on September 12 and 13. The event aimed at bringing together college students, working professionals and SG GSC employees together for a 30-hour build session which saw 165 participants across 45 teams showcase their coding abilities to create applications.
This application was awarded first place
An app that analyses stock portfolio, recommends securities based on the user profile and also verifies recommendations using sentiment analysis from social media.
A python microframework called Flask was used to build a RESTful connection between Client (Front-end) and Server (Back-end).
Back-end was deployed using Apache Spark for big data munging with the help of pyspark and mllib.
Collaborative Filtering is a method of making automatic predictions (filtering) about the interests of a user by collecting preferences or taste information from many users (collaborating). It is been vastly used by "e-commerce gaints" in order to provide its clients a highly personalized recommendations based on thier past transactions and common interests with other people with similar purchases.
A sample collaborative filtering dataset by Jester was used to train the model.
Accuracy of the model
Here,
- Green indicates the "Predicted" metric of all the stocks for a given user.
- Blue indiciates the "Actual" pre-calculated metric of all the stocks for a given user.
- x-axis List of all stocks index from 0 to 40 [which are in our consideration]
- y-axis Represents the likeliness that the given person would go for that stock. More positive the value, more likely he is to buy that stock.
This prediction was further verified using Sentiment Analysis of public data (twitter/facebook/articles) on that stock (Opinion Mining) to help user make an INFORMED decision.