Supervisors info:
Μανόλης Κουμπαράκης, Καθηγητής, Τμήμα Πληροφορικής και Τηλεπικοινωνιών, Εθνικό και Καποδιστριακό Πανεπιστήμιο Αθηνών
Summary:
The integration of Retrieval-Augmented Generation (RAG) with geospatial
search presents a novel approach to enhancing the accuracy and relevance of
AI-generated responses in location-based queries. While traditional generative
models rely solely on pre-trained knowledge, RAG dynamically retrieves and
incorporates external information, ensuring more contextually grounded and
up-to-date outputs. This research focuses on optimizing geospatial retrieval
mechanisms within the RAG framework, leveraging vector-based search,
neural embeddings, and spatial indexing techniques to improve retrieval
precision in spatial datasets.
To achieve this, the study implements geospatial indexing methods, such as
2dsphere indexing, facilitating efficient proximity-based searches and spatial
filtering. Additionally, natural language processing (NLP) techniques are
employed to interpret user queries, ensuring seamless integration between
textual and geographic data. The proposed system enables the retrieval of
structured location-based information, ranking results based on both semantic
similarity and geospatial relevance.
This work demonstrates the efficiency of embedding-based retrieval and vector
search in geospatial contexts, highlighting improvements in query processing
speed, spatial accuracy, and response coherence. These technologies
contribute to the advancement of location-aware AI applications, including
geographic information systems (GIS), recommendation systems, and spatial
data analytics. Future research will explore scalability, adaptive retrieval
mechanisms, and real-time processing for dynamic location-based search
environments.
Keywords:
Retrieval-Augmented Generation, Geospatial Search, Vector Search, 2dsphere Indexing, NLP, Spatial Data Retrieval
