Embracing Uncertainty in Entity Linking

Information Systems

One of the key challenges to realize automated processing of the information on the Web, which is the central goal of the Semantic Web, is related to the entity matching problem. There are a number of tools that reliably recognize named entities, such as persons, companies, geographic locations, in Web documents. The names of these extracted entities are, however, non-unique; the same name on different Web pages might or might not refer to the same entity. The entity matching problem concerns of identifying the entities, which are referring to the same real-world entity. This problem is very similar to the entity resolution problem studied in relational databases, however, there are also several differences. Most importantly Web pages often only contain partial or incomplete information about the entities.

We introduce Dexter, an open source framework for entity linking. The entity linking task aims at identifying all the small text fragments in a document referring to an entity contained in a given knowledge base, e.g., Wikipedia. The annotation is usually organized in three tasks. Given an input document the first task consists in discovering the fragments that could refer to an entity. Since a mention could refer to multiple entities, it is necessary to perform a disambiguation step, where the correct entity is selected among the candidates. Finally, discovered entities are ranked by some measure of relevance. Many entity linking algorithms have been proposed, but unfortunately only a few authors have released the source code or some APIs. As a result, evaluating today the performance of a method on a single subtask, or comparing different techniques is difficult. In this work we present a new open framework, called Dexter, which implements some popular algorithms and provides all the tools needed to develop any entity linking technique. We believe that a shared framework is fundamental to perform fair comparisons and improve the state of the art.

Proceedings of the first international workshop on Entity recognition & disambiguation - ERD '14, 2014

This paper describes our system for the Entity Recognition and Disambiguation Challenge 2014. There are two tasks: one to find entities in queries (Short Track), the other to find entities in texts from web pages (Long Track). We have participated in both tracks with the same system tuned to each of the tasks. On the final test set, we reached the f-measure of 71.9% on the Long Track and of 66.9% on the Short Track. We describe our system and its components in depth, together with their influence on performance. The specifics of each of the tasks are also discussed.

2011

Abstract. The Entity Resolution problem has been widely addressed in the literature. In its simplest version, the problem takes as input a knowledge base composed of records describing real world entities and outputs the sets of records judged to correspond to the same real world entity. More elaborated versions take into account links amongst records representing relationships between the entities which represent. However, none of the approaches in the literature question the validity of certain links between records.

For unsupervised clustering, traditional accuracy metrics based on the constituent records do not often reflect the accuracy at the cluster level. For a specific example, consider entity resolution where the goal is to cluster records across multiple, heterogeneous data sources into "entities." Measuring the accuracy of entity resolution is not as simple as applying the well known record level metrics of precision and recall. Rather than using traditional tuple-based metrics for accuracy, we posit that new, entity-based metrics should be defined instead. Defining entitylevel metrics gains users a less source biased, yet deeper insight into entity resolution performance. We show that traditional record linkage metrics are not appropriate, and offer some early thoughts on entity-centric measurements that are more so.

2012

Several research results have shown that specifying the information about certain entities is the most common information demand of information retrieval users. The needs should be answered by returning specific entities, their properties or related concepts instead of just any type of documents. While some search engines are capable of recognizing specific types of entities, true entity-oriented search still has a long way to go because of the high ambiguity in names across documents. Entity linking (EL) goes beyond the entity recognition task by linking a textual named entity mention to a knowledge base entry. It is a difficult task involving several challenges. This paper gives a survey of the EL tasks in the general and the biomedical domain. In addition, results of our latest EL work are provided for reference, which uncover new EL challenges found in biomedical text mining, along with discussions regarding their possible solutions.

2018 IEEE 34th International Conference on Data Engineering (ICDE), 2018

Entity Resolution (ER) aims to identify different descriptions in various Knowledge Bases (KBs) that refer to the same entity. ER is challenged by the Variety, Volume and Veracity of descriptions published in the Web of Data. To address them, we propose the MinoanER framework that fulfills full automation and support of highly heterogeneous entities. MinoanER leverages a token-based similarity of entities to define a new metric that derives the similarity of neighboring entities from the most important relations, indicated only by statistics. For high efficiency, similarities are computed from a set of schema-agnostic blocks and processed in a non-iterative way that involves four threshold-free heuristics. We demonstrate that the effectiveness of MinoanER is comparable to existing ER tools over real KBs exhibiting low heterogeneity in terms of entity types and content. Yet, MinoanER outperforms state-of-the-art ER tools when matching highly heterogeneous KBs.

ACM Transactions on Database Systems, 2016

We introduce and develop a declarative framework for entity linking and, in particular, for entity resolution. As in some earlier approaches, our framework is based on a systematic use of constraints. However, the constraints we adopt are link-to-source constraints, unlike in earlier approaches where source-to-link constraints were used to dictate how to generate links. Our approach makes it possible to focus entirely on the intended properties of the outcome of entity linking, thus separating the constraints from any procedure of how to achieve that outcome. The core language consists of link-to-source constraints that specify the desired properties of a link relation in terms of source relations and built-in predicates such as similarity measures. A key feature of the link-to-source constraints is that they employ disjunction, which enables the declarative listing of all the reasons two entities should be linked. We also consider extensions of the core language that capture collec...

2010

Abstract. We propose a method for consolidating entities in RDF data on the Web. Our approach is based on a statistical analysis of the use of predicates and their associated values to identify “quasi”-key properties. Compared to a purely symbolic based approach, we obtain promising results, retrieving more identical entities with a high precision. We also argue that our technique scales well—possibly to the size of the current Web of Data—as opposed to more expensive existing approaches.

2014

Data Integration is the problem of combining data in various data sources and providing a user with a unified view over these sources. Building an automatic data integration system that can process large, semi-structured data sources has emerged as an important problem. An automated data integration system requires automatic population of an Entity Name System (ENS). An ENS is a thesaurus for entities and is used to serve instance matching needs across data sources. Resource Description Framework (RDF) is a graph-based data model used to publish data on the Web as linked data. To build and populate an ENS over linked data, the fundamental problem of data matching needs to be solved. Traditionally, data matching concerned identifying pairs of logically equivalent entities across one or more structurally homogeneous data sources, and required a human in the loop. Additionally, most systems run on serial architectures. These assumptions cannot be expected to hold for linked data. Given...