Using Knowledge Base Semantics in Context-Aware Entity Linking

Abstract

Entity linking is a core task in textual document processing, which consists in identifying the entities of a knowledge base (KB) that are mentioned in a text. Approaches in the literature consider either independent linking of individual mentions or collective linking of all mentions. Regardless of this distinction, most approaches rely on the Wikipedia encyclopedic KB in order to improve the linking quality, by exploiting its entity descriptions (web pages) or its entity interconnections (hyperlink graph of web pages). In this paper, we devise a novel collective linking technique which departs from most approaches in the literature by relying on a structured RDF KB. This allows exploiting the semantics of the interrelationships that candidate entities may have at disambiguation time rather than relying on raw structural approximation based on Wikipedia's hyperlink graph. The few approaches that also use an RDF KB simply rely on the existence of a relation between the candidate entities to which mentions may be linked. Instead, we weight such relations based on the RDF KB structure and propose an efficient decoding strategy for collective linking. Experiments on standard benchmarks show significant improvement over the state of the art.

Overview of CLEF HIPE 2020: Named Entity Recognition and Linking on Historical Newspapers

“…For NERC, they improved the non-neural CRF baseline system with additional features such as context tokens, date regex match, ASCII normalization of the focus token, and the 100 most similar words from the HIPE fastText word embeddings provided by the organizers. For EL, a knowledge-base driven approach was applied to disambiguate and link the mentions of their NERC systems and the gold oracle NERC mentions [15]. Their experiments with the HIPE data revealed that collective entity linking is also beneficial for this type of texts-in contrast to linking mentions separately.…”

A Novel Path-Based Entity Relatedness Measure for Efficient Collective Entity Linking

“…A handful of measures rely on RDF KBs [1,18,15,3,25,9]. Such KBs model both data (facts) and knowledge (ontological description of the application domain) using explicit and implicit triples; the latter can be derived through reasoning based on an RDF-specific consequence relation, a.k.a.…”

“…In the sequel, we chose to rely on WSRM to capitalize (i) on properties (R1) and (R2) that WSRM verifies and (ii) on its state-of-the-art performance for collective entity linking, in particular w.r.t. Ref [9].…”

“…Such KBs encode in a formal knowledge graph model, the precise semantics of entities (e.g., types) and of their direct relations called properties (e.g., property names and cardinalities). While the Ref measure [1,18] just provides a binary indicator of whether or not a relation exists between two entities in the RDF KB, the recent WSRM measure [9], which can be viewed as an extension of Ref, further considers the amount of relations between two entities. Though they both have (simple) clear semantics (R1) and are cheap to calculate with edge lookups (R2), they only consider properties between entities to compute relatedness, thus do not meet (R3).…”

“…Importantly, ASRMP m is not tied to WSRM (i.e., another measure could have been used like Ref). We adopt it here because, in addition to satisfying (R1) and (R2), it is currently the relatedness measure showing best performance for collective entity linking in the literature [9]. Our fuzzy logic-based aggregation scheme allows ASRMP m to inherit both (R1) and (R2) from WSRM.…”