Not logged in.
Quick Search - Contribution
Contribution Details
| Type | Master's Thesis |
| Scope | Discipline-based scholarship |
| Title | Neural-Network Architectures and Learning Methods for Financial News Understanding |
| Organization Unit | |
| Authors |
|
| Supervisors |
|
| Language |
|
| Institution | University of Zurich |
| Faculty | Faculty of Business, Economics and Informatics |
| Date | 2018 |
| Abstract Text | In light of the recent success of applying deep neural network technologies in natural language processing, we carried out an in-depth investigation into the use of di erent neural networks for the dicult task of nancial news understanding. In this work, we proposed several novel neural-network architectures and tested a multi-task learning framework aimed to predict the sentiment involved in the nancial news. Our results show that the deep networks we proposed outperform the existing solutions reported in SemEval 2017 Task 5. Meanwhile, we nd that deeper architectures could outperform shallow ones in this task. We present a well-designed architecture that takes companyspeci c information could help the model to understand the news better. Also, by training on the data set labeled by market response, we nd that there are synergy between sentiment prediction task and market response prediction task; these two tasks share a few abstract features at the very beginning. With these well-behaved model at hand, we design a sequential prediction model that summarizes daily news to predict stock returns. The cross-sectional test shows the scores our model predicts could lead to signi cant returns. |
| Export |
BibTeX
|