GLOW System

A GLOW System is a Disambiguation to Wikipedia System that is based on ranking and linking optimization methods.

• AKA: Global Wikification System.
• Context:
  • It solves a GLOW Task by implementing GLOW Algorithms.
  • It was developed by Ratinov et al., 2011.
  • …
• Example(s):
  • …
• Counter-Example(s):
  • Wikipedia-based Semantic Tagging System,
  • TAGME Wikification System (Ferragina & Scaiella, 2010),
  • Wikify System (Mihalcea & Csomai, 2007)),
  • GMRKB-directed Wikification System,
  • WikiM Wikification System,
  • WikiWikiWeb System.
• See: Natural Language Processing System, WikiText Error Correction System, Co-Reference Resolution System, Part-of-Speech Tagging System, Semantic Role Labeling System, Shallow Parsing System, Text Analysis System, Document to Ontology Interlinking System, Wikimedia, Disambiguation to Wikipedia Task.

---

---

References

2011

• (Ratinov et al., 2011) ⇒ Lev Ratinov, Dan Roth, Doug Downey, and Mike Anderson. (2011). “Local and Global Algorithms for Disambiguation to Wikipedia.” In: Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies - Volume 1. ISBN:978-1-932432-87-9
  • QUOTE: The common approach is to utilize the Wikipedia link graph to obtain an estimate pairwise relatedness between titles [math]\displaystyle{ \psi (t_i ,t_j) }[/math] and to efficiently generate a disambiguation context [math]\displaystyle{ \Gamma ' }[/math] , a rough approximation to the optimal [math]\displaystyle{ \Gamma^∗ }[/math] . We then solve the easier problem:
    [math]\displaystyle{ \Gamma^* \approx \arg \underset{\Gamma}{\operatorname{max}} \sum^N_{i=1} [\phi(m_i ,t_i) + \sum_{t_j\in \Gamma '} \psi(t_i ,t_j )]\quad }[/math] (3)

    (...) In this section, we present our global D2W system, which solves the optimization problem in Eq. 3. We refer to the system as GLOW, for Global Wikification. We use GLOW as a test bed for evaluating local and global approaches for D2W. GLOW combines a powerful local model [math]\displaystyle{ \phi }[/math] with an novel method for choosing an accurate disambiguation context [math]\displaystyle{ \Gamma ' }[/math] , which as we show in our experiments allows it to outperform the previous state of the art.

    (...) At a high level, the GLOW system optimizes the objective function in Eq. 3 in a two-stage process. We first execute a ranker to obtain the best non-null disambiguation for each mention in the document, and then execute a linker that decides whether the mention should be linked to Wikipedia, or whether instead switching the top-ranked disambiguation to null improves the objective function. As our experiments illustrate, the linking task is the more challenging of the two by a significant margin.

---