Illinois Cognitive Computation Group UI-CCG TAC 2013 Entity Linking and Slot Filler Validation Systems

In this paper, we describe the University of Illinois (UI CCG) submission to the 2013 TAC KBP English Entity Linking (EL) and Slot Filler Validation (SFV) tasks. We developed two separate systems. Our Entity Linking system integrates an improved version of the Illinois Wikifier with additional functionality to identify and cluster entity mentions that do not correspond to entries in the reference knowledge base. Our Slot Filler Validation system follows an entailment formulation that evaluates each candidate answer based on the evidence present in the source document it refers to. 1 Illinois Entity Linking System The goal of the TAC KBP English Entity Linking (EL) task is to cluster name entity mentions in a document and either link them to entities in a knowledge base (KB), or assign them to a non-KB entry (NIL) with a unique NIL ID. Our Entity Linking system has two main components: (1) the Wikifier1 (Cheng and Roth, 2013; Ratinov et al., 2011) as the underlying knowledge base linking engine, linking all mentions in a given text to the entire Wikipedia (a superset of the TAC KBP Knowledge Base); and (2) a cross-document coreference resolution system based on the Best Latent Left-Linking (L3M ) approach (Samdani et al., 2012; Chang et al., 2013). During testing, the system runs the linking and clustering components separately and then combines the results. http://cogcomp.cs.illinois.edu/page/ software_view/33 It is important to note that our Wikifier component is not retrained on TAC data and our L3M clustering is trained and tuned on the 2012 Entity Linking queries, optimizing for the B3 F1 metric. 1.1 EL System Description The UI CCG EL system first preprocesses the query mentions and documents. Then it applies the Wikifier to link the mentions to Wikipedia entities and applies cross-document coreference resolution system to cluster mentions into groups. The final decision is provided by a voting scheme based on the clustering and wikification results. Figure 1 shows the overall system architecture. We next describe each stage in detail.