NewsReader Data Model ontology 0.2

Namespace Document 09 Jan 2014

Document URL:

http://dkm.fbk.eu/ontologies/newsreader.html (HTML, PDF)

Vocabulary URL:

http://dkm.fbk.eu/ontologies/newsreader.owl (RDF/XML, TURTLE, Manchester Syntax)

Authors:

Marco Rospocher, Luciano Serafini (FBK-Irst)
Francesco Corcoglioniti

Copyright © 2014 FBK-Irst

Abstract

Status of This Document

Table of Contents

• Overview
• Resource layer
• Mention layer
• Entity layer
• Terms reference

Overview

The following UML class diagram informally presents an overview of the ontology. Classes are rendered as UML classes, datatype properties as attributes and object properties as UML relations; minimum and maximum cardinalities and expected datatypes are also shown. The components of the ontology are detailed in the following sections.

The vocabulary namespace is http://dkm.fbk.eu/ontologies/newsreader. The suggested prefix for referencing the vocabulary is nwr:.

A list of classes and properties is reported below, with links to their reference documentation:

Resource layer

For each processed news, two resources are stored in the KnowledgeStore: (i) a News resource for the news itself, containing its metadata and, optionally, its textual content (depending on availability and copyright agreements); and (ii) a NAFDocument resource storing the NAF document generated for the news.

News are described using metadata from the Dublin Core Metadata Terms vocabulary (dct:* attributes), augmented with NewsReader-specific attributes to keep track of the external source document the news has been imported from: originalFileName, originalFileFormat, originalPages (as defined in NAF).

NAF documents are described with the subset of metadata from the NAF header that is most relevant for selecting NAF documents in the KnowledgeStore. This subset comprises: the NAF version (attribute version); the publicId of the NAF document (attribute dct:identifier); the NAF layers available in the NAF document, e.g., text, terms, deps (attribute layer); the NAF processors used (attribute dct:creator) and the language of the processed document (attribute dct:language); complete metadata and all the produced linguistic annotations are available in the stored XML content of the NAF document.

Mention layer

Based on the markables of WP3 annotation guidelines, four main types of mentions are formalized in this ontology: Entity mentions, Relation mentions, Signal and CSignal mentions and Value mentions; for all of them, the position of the mention in the news is encoded with numerical character offsets based on the NLP Interchange Format (NIF) vocabulary (attributes nif:beginIndex, nif:endIndex, nif:anchorOf), so to enable interoperability with tools consuming NIF data.

Entity mentions (class EntityMention) denote entities in the domain of discourse (linked with ks:refersTo). An optional localCorefID attribute can be used to group mentions coreferring within a document (intra-document coreference). Entity mentions are further characterized based on the type of entity:

• Object mentions (class ObjectMention) refer to persons, locations, organizations, products, financial objects (e.g., "NASDAQ Index") and mixed entities (e.g., "the CEO and his company"), discriminated via attribute entityType and enumeration EntityType; the types considered are those proposed in the revised annotation guidelines of WP3. Object mentions are described by: a syntactic head (attribute syntacticHead); a syntactic type, e.g., name, nominal or pronoun (attribute syntacticType, enumeration SyntacticType); and a linguistic entity class, e.g., specific referential (attribute entityClass, enumeration EntityClass).
• Time mentions (class TimeMention) are described using the subset of TIMEX3 properties selected in NAF and in the annotation guidelines. These properties include: the TIMEX3 type, e.g., date, time, duration (attribute timeType, enumeration TIMEX3Type); the normalized time value (attribute value); the function within the document, e.g., document creation time (attribute functionInDocument, enumeration FunctionInDocument); relations with other time mentions (attributes beginPoint, endPoint, anchorTime, valueFromFunction); the optional quantifier (e.g., "every"), frequency (e.g., "twice-a-month") and modifier (e.g., "approx") characterizing the expression and whether it is used as a temporal function (respectively, attributes quant, freq and mod, with corresponding enumeration TIMEX3Modifier).
• Event mentions (class EventMention) are characterized using a number of attributes: the linguistic class of the event, e.g., speech-cognitive (attribute eventClass, enumeration EventClass); the lemma of the token conveying the event (attribute pred); the part-of-speech (attribute pos, enumeration PartOfSpeech), e.g., adjective, noun or verb; the certainty and factuality of the event, together with a factuality confidence (attributes certainty, factuality and factualityConfidence, with enumerations Certainty and Factuality); the tense, aspect, polarity (e.g., positive) and modality (e.g., ``should'') of the verbal form used (respectively, attributes tense, aspect, polarity and modality, with enumerations Tense, Aspect and Polarity). In addition, references to external resources further specifying the type of event are stored using attributes framenetRef, verbnetRef, propbankRef and nombankRef.

Relation mentions (class RelationMention) express relations between two entities, whose mentions are identified by source and target links. Different kinds of relation mentions are stored:

• Causal links (class CLink) express a causal relation between two events.
• Temporal links (class TLink) denote a certain temporal relation (attribute relType, enumeration TLinkType), e.g., before, include, simultaneous, among two events or time expressions.
• Subordinate links (class SLink) express certain structural relations among events.
• GLinks (class GLink) express grammatical relations among events, as in ``the share drop came on the same day'', with ``drop'' and ``came'' being events.
• Participation mentions (class Participation) denote the participation of an entity to an event in a certain thematic role (attribute thematicRole), possibly further specified by references to external resources (attributes framenetRef, verbnetRef, propbankRef and nombankRef).

Signal and CSignal mentions (respectively, classes SignalMention and CSignalMention) identify pieces of text supporting the existence of a temporal or causal relation, to which they are linked by relations signal and csignal.

Value mentions (class ValueMention) are numerical expressions used for quantities (cardinal numbers in general), percentages and monetary expressions; the type of value is expressed by attribute valueType, enumeration ValueType.

Entity layer

Different kinds of entities are stored, including persons, organizations, geo-political entities or locations, events, points and intervals in time extracted from text; the type of entity is conveyed by an rdf:type axiom.

The context in which an axiom holds is described and identified in terms of temporal validity (attribute sem:hasTimeValidity) and time-referenced point of view (attribute sem:hasPointOfView), e.g., ``Financial Times'' point of view expressed on 2013/12/15; the Simple Event Model (SEM) and the OWL Time vocabularies are used to that purpose.

Axiom metadata consists of a confidence value (attribute confidence), a provenance indication (attribute dct:source) and a crystallized flag (attribute crystallized). Confidence is represented on a 0.0 - 1.0 scale and quantifies how reliable an extracted statement is. Provenance is stored for background knowledge axioms and denote the external sources they have been imported from, e.g., DBPedia (note that he adoption of a provenance model to track sources, authority, and tool processing activities, is still under definition at project level). The crystallized flag is set for axioms belonging to background knowledge or assimilated to it after repeated extraction of the conveyed information, according to some crystallization algorithm.

Terms reference