The iKernels group carries out advanced research on machine learning methods for syntactic and semantic processing of natural language. The proposed models aim at improving on the state-of-the-art in Information Search and Retrieval by enriching language models based on bag-of-words, with structural representations and semantics.
In the context of natural language the group owns internationally recognized expertise in the following applications: Question Answering, FrameNet and PropBank Predicate Argument Extraction (Semantic Role Labeling), Relation Extraction, Syntactic and Semantic Parsing, Co-reference resolution, Text Categorization, Textual Entailment Recognition, Word Sense Disambiguation, Entity Recognition and Normalization, Opinion Mining, Speech and Noisy Text Processing, Text Similarity and Summarization.
Regarding Machine Learning, the group is developing new theory and methods for Kernel Machines: Kernel Methods, Structural Kernels, Support Vector Machines, On-line Learning, Structured Output Spaces, Multi-label and Hierarchical Classification and Re-Ranking.
Theory and methods are also applied to broader ICT areas than natural language processing, such as electronic feeder failure detection, bioinformatics, and automatic software processing (e.g., anomaly detection, code classification).