Semantic Kernel is an entity linking engine based on a cutting-edge machine-learning technology. It has the following three unique features:
End-to-End Entity Linking
Conventional entity linking systems typically depend on named entity recognition system (NER) to detect entity names from documents. Due to limited performance of NER, they often bring poor results, especially in noisy documents such as tweets. Our end-to-end entity linking system detects entity names without NER. Our proposed system has recently won two international competitions1, 2 hosted in well-known academic conferences.
Evaluating Helpfulness of Entities
Wikipedia contains large number of entities that are unlikely to be important for audiences (e.g., bag, vehicle), which causes frequent noises on various tasks such as tagging and text analysis in use of entity linking. We invented a novel method that recognizes whether a detected entity is likely to be important for audiences, which we presented in a leading international conference3.
Recognizing Entity Types
Semantic Kernel accurately recognizes the types of an entity (e.g., Actor, Organization) using DBpedia Ontology Classes. This makes it possible to extract entities that belong to specific entity types from a document.
Ikuya Yamada, Hideaki Takeda, Yoshiyasu Takefuji: An End-to-End Entity Linking Approach for Tweets, WWW 2015 Workshop on Making Sense of Microposts (Florence, Italy), 2015, pp.55-56
Ikuya Yamada, Hideaki Takeda, Yoshiyasu Takefuji: Enhancing Named Entity Recognition in Twitter Messages Using Entity Linking, ACL 2015 Workshop on Noisy User-generated Text (Beijing, China), 2015
Ikuya Yamada, Tomotaka Ito, Shinsuke Takagi, Shinnosuke Usami, Hideaki Takeda, Yoshiyasu Takefuji: Evaluating the Helpfulness of Linked Entities to Readers, 26th ACM Conference on Hypertext and Social Media (Santiago Downtown, Chile), 2014, pp.169-178