This thesis investigates the application of Natural Language Processing techniques in the domain of network management, specifically focusing on automated log anomaly detection. As modern network systems generate massive and complex log data, manual analysis becomes impractical, necessitating the development of automated solutions. This research addresses the challenges of log data complexity, volume, and diversity by proposing a comprehensive framework that integrates NLP with deep learning models, particularly LSTM networks, for effective log anomaly detection. The research is structured around four primary objectives: 1) developing a log parsing framework to convert unstructured logs into structured formats, 2) utilizing OpenAI models for feature embedding extraction from parsed logs, 3) implementing an LSTM-based model for anomaly detection, and 4) evaluating the proposed approach against baseline models on a benchmark dataset. The proposed solution leverages advanced NLP techniques, including fine-tuned large language models, to generate high-quality embeddings that capture the semantic meaning of log events. These embeddings serve as inputs for the LSTM model, which is trained to detect anomalies by learning the normal patterns of log sequences. The evaluation results demonstrate that the proposed method significantly outperforms traditional approaches in terms of accuracy, scalability, and real-time processing capability. The research findings highlight the effectiveness of integrating NLP and deep learning for enhancing the automation and efficiency of network management tasks, particularly in anomaly detection. This thesis contributes to the advancement of automated network management systems by offering a novel approach that improves the accuracy and adaptability of log analysis in complex network environments. The proposed methodologies and models are not only applicable to network management but also hold potential for broader applications in fields requiring the analysis of large, unstructured datasets. Future research directions include improving model generalization, optimizing real-time processing, and extending the framework to other domains such as cybersecurity and financial fraud detection.
