eprintid: 34230
rev_number: 21
eprint_status: archive
userid: 7853
dir: disk0/00/03/42/30
datestamp: 2024-01-26 09:20:23
lastmod: 2024-02-01 19:25:38
status_changed: 2024-01-26 09:20:23
type: doctoralThesis
metadata_visibility: show
creators_name: van den Berg, Esther
title: Corpus-based and Computational Analysis of
Entity Framing
subjects: ddc-004
subjects: ddc-400
subjects: ddc-600
divisions: i-90500
adv_faculty: af-09
keywords: Natural Language Processing, Computerlinguistik, Computational Linguistics, Computational Social Science
cterms_swd: frame
cterms_swd: Maschinelles Lernen
cterms_swd: Soziolinguistik
abstract: Entity Framing is the selection of aspects of an entity to promote a particular viewpoint towards that entity. Compared to issue framing, it has received little attention in Framing research, and it has also received very little attention in Natural Language Processing (NLP). We investigate Entity Framing of political figures on social media and the news through the selection of objectively verifiable attributes like name, title and background information.

Despite indications that they signal the perceived status of the target and/or perceived solidarity with a target, naming and titling have not previously been quantitatively examined in terms of their relation to stance. In this thesis, we collect English and German tweets mentioning prominent politicians and show that naming variation relates positively to stance in a way that is suggestive of a framing effect mediated by respect. We show on the German corpus that this positive relation is impacted by differences in political orientation.

Having provided the first quantitative evidence for the relation between stance and the mentioning of the objectively verifiable attributes name and title, we turn to engineering efforts towards automating detection of the selective mentioning of background information. By nature, whether certain information constitutes an instance of Entity Framing depends on the context in which that information is provided. Nevertheless, previous work on detection of framing through background information has not explored the role of context beyond the sentence. We experiment with computational methods for integrating three kinds of context: context from the same article, context from other publishers’ articles on the same event, and inclusion of texts from the same domain (but potentially different events). We find that integrating event context improves classification performance over a strong baseline. We additionally show through a series of performance tests that this improvement over the baseline holds specifically for instances that are likely to be more difficult to classify, as one would expect from a performance boost that is due to leveraging context.

The result of these studies is a collection of new data sets, methods and findings with respect to Entity Framing, contributed in the hope that further computational research on this topic will be conducted, in order to improve our understanding of Entity Framing in general and of political figures in particular.
date: 2024
id_scheme: DOI
id_number: 10.11588/heidok.00034230
ppn_swb: 1879831732
own_urn: urn:nbn:de:bsz:16-heidok-342307
date_accepted: 2023-02-21
advisor: HASH(0x558ea2549870)
language: eng
bibsort: VANDENBERGCORPUSBASE20221101
full_text_status: public
place_of_pub: Heidelberg
citation:   van den Berg, Esther  (2024) Corpus-based and Computational Analysis of Entity Framing.  [Dissertation]     
document_url: https://archiv.ub.uni-heidelberg.de/volltextserver/34230/1/vdberg_thesis_for_publication.pdf