Advancing Deep-Learning in NLP: From Network Logs to Text Classification

This PhD Symposium Paper explores deep-learning based sequential modeling methods for textual data by investigating two applications: network security log analysis and financial forecasting. The network log analysis was performed using a hybrid approach that combines traditional pattern matching and transformer architectures. This method successfully identified common log patterns and achieved high accuracy in classifying new log types. For economic forecasting, economic news articles and retail data are used; transformer models demonstrated superior performance in sentiment analysis and item classification. Sentiment indices showed predictive power during economic downturns. These findings suggest that the integration of novel methods with traditional approaches can lead to improved models and robust applications that bridge qualitative and quantitative data.