Bat Algorithm for Complex Event Pattern Detection in Sentiment Analysis
Complex event pattern detection has become an emerging research area in various monitoring applications. For learning and predicting the event patterns, dynamic Bayesian network (DBN) with Hidden Markov Model (HMM) and heuristic search learning algorithms have been a popular technique used in which...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Monograph |
| Language: | English |
| Published: |
UUM
2021
|
| Subjects: | |
| Online Access: | https://repo.uum.edu.my/id/eprint/29643/1/13064.pdf https://repo.uum.edu.my/id/eprint/29643/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Complex event pattern detection has become an emerging research area in various monitoring applications. For learning and predicting the event patterns, dynamic Bayesian network (DBN) with Hidden Markov Model (HMM) and heuristic search learning algorithms have been a popular technique used in which structure learning is trained to classify complex events pattern. Regardless the success of DBN with HMM in detecting complex event patterns, learning the spatial and temporal features from large-scale events is a challenging task. This is due to the difficulties in learning DBN structure which is NP-hard as number of possible graphs increases super-exponentially with the number of events. Thus, this study proposed a Bat Algorithm (BA) to address the complex learning structure of DBN in detecting sentiment patterns. This algorithm mainly based on the echolocation behavior of bats. Two main characters are highlighted in this learning, which is the velocity and frequency. Bats fly randomly with certain velocity at a fixed frequency to search for prey. They can automatically adjust the frequency of their emitted pulses and modify the rate of pulse emission, depending on the proximity of their target. Such behavior may be suitable to analyze sentiment expression. However, BA needs to be coupled with appropriate learning scheme in determining the tradeoff between global explorative search with local exploitative search. In addressing this matter, the study includes several phases such as designing a DBN learning scheme, determining spatio-temporal sentiment features and creating BA in analyzing large-scale sentiment expressions. The outcome of this research is to detect sentiment polarity patterns (whether the polarity is positive, neutral or negative based on subjectivity lexicons) from spatio-temporal activities on news. The pattern can be used to monitor and analyze emerging issues in the setting of public sector as direct response to the surge of interest. |
|---|