Probabilistic Bag-Of-Hyperlinks Model for Entity Linking

Abstract

Many fundamental problems in natural language processing rely on determining what entities appear in a given text. Commonly referenced as entity linking, this step is a fundamental component of many NLP tasks such as text understanding, automatic summarization, semantic search or machine translation. Name ambiguity, word polysemy, context dependencies and a heavy-tailed distribution of entities contribute to the complexity of this problem.We here propose a probabilistic approach that makes use of an effective graphical model to perform collective entity disambiguation. Input mentions (i.e., linkable token spans) are disambiguated jointly across an entire document by combining a document-level prior of entity co-occurrences with local information captured from mentions and their surrounding context. The model is based on simple sufficient statistics extracted from data, thus relying on few parameters to be learned.Our method does not require extensive feature engineering, nor an expensive training procedure. We use loopy belief propagation to perform approximate inference. The low complexity of our model makes this step sufficiently fast for real-time usage. We demonstrate the accuracy of our approach on a wide range of benchmark datasets, showing that it matches, and in many cases outperforms, existing stateof-the-art methods.

Joint Entity Recognition and Linking in Technical Domains Using Undirected Probabilistic Graphical Models

“…Their results show that entity co-occurrences and mention-entity pairs provide complementary information to the model. Based on the same sources of information, Ganea et al [4] train a Markov network for the entity linking task, using approximate MAP inference by belief propagation. Both Ganea et al and Hakimov et al perform entity linking in isolation by relying on gold annotations for the recognition problem.…”

“…In each iteration, an annotation span explorer and a concept assignment explorer are consecutively applied in order to generate a set of proposal states which differ from the current state in one atomic change. The annotation span explorer is able to add a new non-overlapping (empty) annotation 4 , remove an existing annotation, or apply a semantic transformation to one token. The concept assignment explorer can assign a concept to an empty annotation, or change or remove one from an non-empty annotation.…”

Semantic similarity analysis and application in knowledge graphs

“…The polysemy problem is caused by the fact that multiple entities in KGs might have the same name, and this is quite common for named entities. The task of addressing the polysemy problem for named entities is called NED, and there is a large body of research techniques that have been proposed for addressing NED automatically (Cucerzan, 2007;Mihalcea and Csomai, 2007;Milne and Witten, 2008;Kulkarni et al, 2009;Mendes et al, 2011;Ganea et al, 2016). However, resolving the polysemy problem is a common challenge whose difficulty is equivalent to solving central problems of AI (Navigli, 2009).…”

“…Following such idea, recent NED works formulate probabilistic framework considering various type of statistical information including mention-entity probability, entity-entity co-occurrence, contextual word-entity statistics from entity annotation dataset (e.g. Wikipedia anchor dataset) (Blanco et al, 2015;Ganea et al, 2016).…”

“…From GERBIL, we retrieve the evaluation results of AGDISTIS (Usbeck et al, 2014), AIDA (Hoffart et al, 2011), Babelfy (Moro et al, 2014), PBOH (Ganea et al, 2016),…”