This Masters thesis aims to study and implement modern neural network architectures for record matching involving data that contain both structured fields (e.g., names, dates, numerical values) and semi-structured or unstructured information (e.g., descriptions, titles, excerpts, or even images).
The proposed approach is based on designing an architecture that combines:
- Semantic embeddings of structured fields and textual content,
- Attention mechanisms to highlight important features, and
- Information aggregation mechanisms for entire records.
The methodological framework is informed by relevant literature, including:

Leave a Reply
You must be logged in to post a comment.