TonalityTech co-founder wins Award for Best Paper at the HICSS 2016
March 28, 2016
The academic research of TonalityTech's CIO and co-founder Nicolas Pröllochs was recently distinguished with a Best Paper Award at the renowned Hawaii International Conference on System Sciences 2016 (HICSS), in Kauai, Hawaii, USA. In this paper, published with Professor Dr. Dirk Neumann and TonalityTech co-founder and CEO Dr. Stefan Feuerriegel, Nicolas Pröllochs applies reinforcement learning to detect negations in financial news sentiment to enhance the measured accurateness of sentiment.
"A positive tone is sometimes difficult to measure as positive words can easily be negated in ways which are difficult to identify with standard algorithms. Since sentiment analysis in financial news is highly vulnerable to negations, it is important to accurately identify the negated parts of a statement", as TonalityTech's CIO Nicolas Pröllochs highlights.
The authors point out how simple negations can invert not only the meaning of single words but of complete phrases and whole sentences – thus they develop machine learning-based reinforcement learning algorithms and benchmark these against conventional rule-based algorithms with the aim of better predicting negated parts.
Applying the reinforced learning methods in fact outperforms conventional rule-based approaches and yields striking results. "With less than 5 percent of sentences in our data base containing negations, our approach achieved an overall 9.2 percent improvement in the correlation of news sentiment with stock market returns when accounting for negations", explains Nicolas Pröllochs. Given the restricted set of sentences with negations, this approach of the researchers from TonalityTech and the University of Freiburg leads to an astonishingly strong improvement of the overall predictive power of the algorithm.Back to news overview