INTRODUCTION TO ARTIFICIAL INTELLIGENCE Massimo Poesio Supervised Relation Extraction
RE AS A CLASSIFICATION TASK Binary relations Entities already manually/automatically recognized Examples are generated for all sentences with at least 2 entities Number of examples generated per sentence is NC2 – Combination of N distinct entities selected 2 at a time
GENERATING CANDIDATES TO CLASSIFY
RE AS A BINARY CLASSIFICATION TASK
NUMBER OF CANDIDATES TO CLASSIFY – SIMPLE MINDED VERSION
THE SUPERVISED APPROACH TO RE Most current approaches to RE are kernel- based Different information is used – Sequences of words, e.g., through the GLOBAL CONTEXT / LOCAL CONTEXT kernels of Bunescu and Mooney / Giuliano Lavelli & Romano – Syntactic information through the TREE KERNELS of Zelenko et al / Moschitti et al – Semantic information in recent work
KERNEL METHODS: A REMINDER Embedding the input data in a feature space Using a linear algorithm for discovering non-linear patterns Coordinates of images are not needed, only pairwise inner products Pairwise inner products can be efficiently computed directly from X using a kernel function K:X×X→R
MODULARITY OF KERNEL METHODS
THE WORD-SEQUENCE APPROACH Shallow linguistic Information: – tokenization – Lemmatization – sentence splitting – PoS tagging Claudio Giuliano, Alberto Lavelli, and Lorenza Romano (2007), FBK-IRST: Kernel methods for relation extraction, Proc. Of SEMEVAL-2007
LINGUISTIC REALIZATION OF RELATIONS Bunescu & Mooney, NIPS 2005
WORD-SEQUENCE KERNELS Two families of “basic” kernels – Global Context – Local Context Linear combination of kernels Explicit computation – Extremely sparse input representation
THE GLOBAL CONTEXT KERNEL
THE LOCAL CONTEXT KERNEL
LOCAL CONTEXT KERNEL (2)
KERNEL COMBINATION
EXPERIMENTAL RESULTS Biomedical data sets – AIMed – LLL Newspaper articles – Roth and Yih SEMEVAL 2007
EVALUATION METHODOLOGIES
EVALUATION (2)
EVALUATION (3)
EVALUATION (4)
RESULTS ON AIMED
OTHER APPROACHES TO RE Using syntactic information Using lexical features
Syntactic information for RE Pros: – more structured information useful when dealing with long-distance relations Cons: – not always robust – (and not available for all languages)
Zelenko et al JMLR 2003 TREE KERNEL defined over a shallow parse tree representation of the sentences – approach vulnerable to unrecoverable parsing errors data set: 200 news articles (not publicly available) two types of relations : person-affiliation and organization-location
ZELENKO ET AL
CULOTTA & SORENSEN 2004 generalized version of Zelenko’s kernel based on dependency trees (smallest dependency tree containing the two entities of the relation) a bag-of-words kernel is used to compensate syntactic errors data set: ACE 2002 & 2003 results: syntactic information improves performance w.r.t. bag-of-words (good precision but low recall)
CULOTTA AND SORENSEN (2)
EVALUATION CAMPAIGNS FOR RE Much of modern evaluation of methods is done by competing with other teams on evaluation campaigns like MUC and ACE Modern evaluation campaigns for RE: SEMEVAL (now *SEM) Interesting to look also at the problems of – DATA CREATION – EVALUATION METRICS
SEMEVAL th International Workshop on Semantic Evaluations Task 04: Classification of Semantic Relations between Nominals – organizers: Roxana Girju, Marti Hearst, Preslav Nakov, ViviNastase, Stan Szpakowicz, Peter Turney, Deniz Yuret – 14 participating teams
SEMEVAL 2007: THE RELATIONS
SEMEVAL 2007: DATASET CREATION
SEMEVAL 2007: DATASET CREATION (2)
SEMEVAL 2007 – DATASET CREATION (3)
SEMEVAL 2007 – DATASET CREATION (4)
SEMEVAL 2007: DATASET
SEMEVAL 2007: COMPETITION
SEMEVAL 2007: COMPETITION (2)
SEMEVAL 2007: BEST RESULTS
INFLUENCE OF NER ON RE
INFLUENCE OF NER ON RE (2)
GENERATING CANDIDATES
ACKNOWLEDGMENTS Many slides borrowed from – Roxana Girju – Alberto Lavelli