A supervised learning framework for news-based sentiment scores conditioned to financial returns
Tutor / directorArratia Quesada, Argimiro
Document typeMaster thesis
Rights accessOpen Access
Finance is about managing money. There are several instruments for dealing with money. There are defined strategies more or less complex, mathematical models for optimal selection of stocks combination and for prediction their assets future values. Our work in this project is to bring a system able to predict if an asset will increase or decrease its future price and help investors to select the best strategy, all based on the sentiment of news articles about this asset. During the path to achieve this system, we have made three key contributions. First, give a full implementation of a recently published novel text-mining methodology for sentiment extraction that does not requires a pre-defined dictionary of sentimental terms. We take care of all technical details, starting from adequately pre-processing the source text-data, design of appropriate data structures and a schema for testing and measuring the model performance. Achieving as much as possible a scalable infrastructure and open to change. Second test the model with another variable than financial returns. This has given us a key aspect of how has to be the shape of input data and helped to understand better which are the limitations of the system. Also, this has been the proof that this model is flexible to be used beyond return financial context, in any kind of sentiment analysis problem. Finally test the model in a real investment environment, to see how good is the system for in- vestment. This procedure includes, developing the core of a suitable investment strategy. Think how the model results could be included in the strategy, and implement a simulator of a real stock broker.
DegreeMÀSTER UNIVERSITARI EN INNOVACIÓ I RECERCA EN INFORMÀTICA (Pla 2012)
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder