Tools
Tools
Thapa, Abhinav (2023) Sentiment Analysis On Juvenile Delinquency Using BERT Embeddings. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (2MB) | Preview |
Preview |
PDF (Configuration manual)
Download (1MB) | Preview |
Abstract
Criminal Juvenile Delinquency is one of the most prevalent problems in the modern society. It generally occurs because of poverty, poor education, lack of awareness, insufficient social infrastructure, and unstable family conditions. Although there are juvenile justice systems in place, few offenders are penalized due to their young age under the guise of teenage expression for less severe crimes. However, sometimes benign crimes eventually propagate over the years into full-fledged antisocial behavior recovering from which becomes virtually impossible. Such offenders can be corrected with effective policy making. But it is difficult to draw policies balancing social conduct and human rights in case of juveniles due to which identifying public opinion in such cases is helpful in decision making. Hence, this project aims at performing sentiment analysis by fine-tuning BERT transformer models on twitter posts dataset to gauge public sentiment towards juvenile delinquency. We also propose a contrast between a BERT fine-tuned model and Machine Learning (Random Forest and Support Vector Classifier) model with BERT embeddings for sentiment classification to determine whether fine-tuning is worth the effort for this problem domain. The feature engineering approach using BERT features (100% Class Accuracy) outperformed the fine-tuned BERT model (77% Class Accuracy) on a benchmark of twitter post dataset on juvenile delinquency proving that in terms of design complexity, execution time, and performance, features engineered directly from the primary dataset using BERT has higher utility when compared with fine-tuned transfer learning approach.
Actions (login required)
 |
View Item |