Experimentation of a semantic search through natural language on structured data relating to incident reports within an IT infrastructure

This thesis explores the development and experimentation of a new semantic search procedure designed to enhance the retrieval of incident reports within an IT infrastructure. Leveraging Elasticsearch, the study focuses on integrating real-time data updates from an incident report database within InfoCamere company. The research introduces a new approach where incident data, stored in Elasticsearch, is processed through a specific module that creates a vector database. This vector database facilitates advanced semantic search capabilities by interpreting natural language queries. The methodology involves the creation and fine-tuning of a software system that interprets user queries in natural language, interacts with the vector database, and provides contextually accurate responses. This system was rigorously tested to evaluate its effectiveness in retrieving relevant incident reports compared to traditional keyword-based search methods. The results demonstrate a significant improvement in search accuracy and relevance, highlighting the system’s ability to understand user intent and deliver precise information. This research contributes to the field of IT infrastructure management by proposing a robust framework for implementing real-time, NLP-driven semantic search using Elasticsearch. The findings suggest that this approach can significantly enhance the efficiency of incident management processes, improving decision-making and reducing response times within IT operations.