BSc thesis · NLP · 2021
Named-entity recognition for leetspeak
Rule-based pipeline that detects leetspeak words in a text ("l33t" → "leet") and maps them to their real equivalents, built for NLPA, the NLP framework of the SING research group at the University of Vigo.
Highlights
- Replaced a dependency on a heavy Stanford NLP model (high RAM usage) with an efficient rule-based implementation with no third-party models.
- Tackled the combinatorial explosion of symbol substitutions with multilingual word dictionaries and a leet-equivalence dictionary released to the community.
- Shipped a Spring Boot web interface so non-programmers could encode and decode leetspeak.
Gallery
Technologies
- NLP
- Named-entity recognition
- Java
- Spring Boot
- Pipelines
- Scrum