ANALYSIS OF NEWS SENTIMENT ON STOCK PRICE DYNAMICS USING LONG SHORT-TERM MEMORY (LSTM) NETWORKS
Complex systems consist of several components that can interact with one another. The system modelling can be approached through scientific research to study the relationships between the components that can cause particular behaviour. In the financial sector, investment is one of the ways to have s...
Saved in:
| Main Author: | |
|---|---|
| Format: | Final Project |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/49108 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | Complex systems consist of several components that can interact with one another. The system modelling can be approached through scientific research to study the relationships between the components that can cause particular behaviour. In the financial sector, investment is one of the ways to have savings in the future. The movement of the stock price is one of the complex things that can be analyzed. In addition to economic indicators, responses of investors based on specific news will also affect the stock price. Therefore, one of the stock prices in the capital market can be affected by investor response based on news sentiments or widely offered information. The method used for sentiment analysis in this study is VADER sentiment analysis. As for the modelling of complex systems, one of the most advanced techniques is Long-Term Memory (LSTM) Networks. The method can learn based on the information stored in memory. This Final Project aims to provide stock price modelling based on news sentiment analysis, understanding the effect of parameter variations in the Long Short-Term Memory (LSTM) Networks model and parameter settings that provide the best modelling based on the variations carried out. Based on the results of modelling, where the significant observed stock price point is increasing and decreasing. The changes interpreted as the news affects circulating in that period. Furthermore, the study has observed the more the amount of sample data used, the better the evaluation scores of the model. Nevertheless, there is no clear relationship between the results of the model evaluation with the dropout parameters. Lastly, the setting that provides the best modelling results with an accuracy score of 0.8 out of 1 is to use dropout value 0.2, and the number of data samples used is 1000. |
|---|