Domain-specific Named Entity Disambiguation in Historical Memoirs

Abstract

This paper presents the results of the extraction of named entities from a collection of historical memoirs about the italian Resistance during the World War II. The methodology followed for the extraction and disambiguation task will be discussed, as well as its evaluation. For the semantic annotations of the dataset, we have developed a pipeline based on established practices for extracting and disambiguating Named Entities. This has been necessary, considering the poor performances of out-of-the-box Named Entity Recognition and Disambiguation (NERD) tools tested in the initial phase of this work.

Index-Driven Digitization and Indexation of Historical Archives

“…Even if in principle it might be useful to capture all the components of every entry of every index of interest into the meta-index, in practice only the indexation units will normally be part of the meta-index as mentions of a thing of interest. The task of performing entity/record linkage or named entity disambiguation on historical texts is an open and challenging research area, beyond the intended scope of this paper (Piotrowski, 2012;Olieman et al, 2017;Rovera et al, 2017).…”

A Semantic Web Approach to Enable a Smart Route to Historical Archives

“…Furthermore, a significant research effort has been devoted to the task of automatically extracting information about events from texts; see, for instance, [12,27,39]. In particular, several works showed that historical texts represent a peculiar domain, where IE/NER tools show quite low performances compared to other domains; see, for example [7,14,33,36]. We do not analyze the related work within these two fields in depth, since these aspects fall outside the scope of this paper.…”

DeezyMatch: A Flexible Deep Learning Approach to Fuzzy String Matching

“…While these choices allow fast run-time, they generally rely on the assumption that all surface forms of each entity are present as aliases in the KB. The performances of these systems degrade when dealing with domain-specific vocabulary (Munnelly and Lawless, 2018), local variations (Rovera et al, 2017), historical materials (Olieman et al, 2017; and, in general, challenges that emerge when performing EL on non-standard documents. 2 This subtask of EL, often referred to as candidate ranking (and selection), is mostly ignored when designing downstream systems, even though its significant impact on downstream NLP pipelines has been shown previously (Quercini et al, 2010;Hachey et al, 2013).…”