Proposed compartmentalized architecture for Context-Sensitive Stance Classification.

<p>The white arrow denotes the starting input for each component. The left figure presents the Tweet Level encoding applied to a batch of <i>n</i><sub><i>max</i></sub> tweets from a given user producing a vector of dimensions () per user. For the <i>Tw...

Full description

Saved in:
Bibliographic Details
Main Author: Ramon Villa-Cox (8311131) (author)
Other Authors: Evan M. Williams (21609012) (author), Kathleen M. Carley (2636068) (author)
Published: 2025
Subjects:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1852018964849229824
author Ramon Villa-Cox (8311131)
author2 Evan M. Williams (21609012)
Kathleen M. Carley (2636068)
author2_role author
author
author_facet Ramon Villa-Cox (8311131)
Evan M. Williams (21609012)
Kathleen M. Carley (2636068)
author_role author
dc.creator.none.fl_str_mv Ramon Villa-Cox (8311131)
Evan M. Williams (21609012)
Kathleen M. Carley (2636068)
dc.date.none.fl_str_mv 2025-06-26T17:56:25Z
dc.identifier.none.fl_str_mv 10.1371/journal.pone.0324697.g001
dc.relation.none.fl_str_mv https://figshare.com/articles/figure/Proposed_compartmentalized_architecture_for_Context-Sensitive_Stance_Classification_/29419917
dc.rights.none.fl_str_mv CC BY 4.0
info:eu-repo/semantics/openAccess
dc.subject.none.fl_str_mv Geology
Cancer
Science Policy
Biological Sciences not elsewhere classified
Information Systems not elsewhere classified
including opinion polling
dimensional space using
detection models using
stance detection models
new country contexts
different country contexts
user &# 8217
predict user stances
political stance detection
extrapolate stance predictions
country model performance
impact social applications
extrapolation </ p
political stance
every country
based user
social context
extrapolation power
wide range
tweet encoders
scale weakly
important task
hate speech
future events
existing large
embed users
detecting propaganda
construct transformer
dc.title.none.fl_str_mv Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
dc.type.none.fl_str_mv Image
Figure
info:eu-repo/semantics/publishedVersion
image
description <p>The white arrow denotes the starting input for each component. The left figure presents the Tweet Level encoding applied to a batch of <i>n</i><sub><i>max</i></sub> tweets from a given user producing a vector of dimensions () per user. For the <i>Tweet Embeddings</i> to serve as input for the User Transformer (depicted in the middle panel) they are reshaped and a [<i>CLS</i>] parameter vector is appended at the start of each user’s tweets. In this layer, a batch of <i>b</i><sub><i>size</i></sub> users is processed by <i>L</i><sub><i>x</i></sub> encoder stacks obtaining the final <i>User Embeddings</i>. On the right, we show the heterogeneous GAT model, wherein a user receives information from its neighborhood and their corresponding <i>User Embeddings</i>. The attention mechanism learns different attention weights for which reflect the importance of a neighbor <i>j</i> to the label of the <i>i</i>’th user.</p>
eu_rights_str_mv openAccess
id Manara_e4d73cdcd9c766b3fd43dcbcfe212662
identifier_str_mv 10.1371/journal.pone.0324697.g001
network_acronym_str Manara
network_name_str ManaraRepo
oai_identifier_str oai:figshare.com:article/29419917
publishDate 2025
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
rights_invalid_str_mv CC BY 4.0
spelling Proposed compartmentalized architecture for Context-Sensitive Stance Classification.Ramon Villa-Cox (8311131)Evan M. Williams (21609012)Kathleen M. Carley (2636068)GeologyCancerScience PolicyBiological Sciences not elsewhere classifiedInformation Systems not elsewhere classifiedincluding opinion pollingdimensional space usingdetection models usingstance detection modelsnew country contextsdifferent country contextsuser &# 8217predict user stancespolitical stance detectionextrapolate stance predictionscountry model performanceimpact social applicationsextrapolation </ ppolitical stanceevery countrybased usersocial contextextrapolation powerwide rangetweet encodersscale weaklyimportant taskhate speechfuture eventsexisting largeembed usersdetecting propagandaconstruct transformer<p>The white arrow denotes the starting input for each component. The left figure presents the Tweet Level encoding applied to a batch of <i>n</i><sub><i>max</i></sub> tweets from a given user producing a vector of dimensions () per user. For the <i>Tweet Embeddings</i> to serve as input for the User Transformer (depicted in the middle panel) they are reshaped and a [<i>CLS</i>] parameter vector is appended at the start of each user’s tweets. In this layer, a batch of <i>b</i><sub><i>size</i></sub> users is processed by <i>L</i><sub><i>x</i></sub> encoder stacks obtaining the final <i>User Embeddings</i>. On the right, we show the heterogeneous GAT model, wherein a user receives information from its neighborhood and their corresponding <i>User Embeddings</i>. The attention mechanism learns different attention weights for which reflect the importance of a neighbor <i>j</i> to the label of the <i>i</i>’th user.</p>2025-06-26T17:56:25ZImageFigureinfo:eu-repo/semantics/publishedVersionimage10.1371/journal.pone.0324697.g001https://figshare.com/articles/figure/Proposed_compartmentalized_architecture_for_Context-Sensitive_Stance_Classification_/29419917CC BY 4.0info:eu-repo/semantics/openAccessoai:figshare.com:article/294199172025-06-26T17:56:25Z
spellingShingle Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
Ramon Villa-Cox (8311131)
Geology
Cancer
Science Policy
Biological Sciences not elsewhere classified
Information Systems not elsewhere classified
including opinion polling
dimensional space using
detection models using
stance detection models
new country contexts
different country contexts
user &# 8217
predict user stances
political stance detection
extrapolate stance predictions
country model performance
impact social applications
extrapolation </ p
political stance
every country
based user
social context
extrapolation power
wide range
tweet encoders
scale weakly
important task
hate speech
future events
existing large
embed users
detecting propaganda
construct transformer
status_str publishedVersion
title Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
title_full Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
title_fullStr Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
title_full_unstemmed Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
title_short Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
title_sort Proposed compartmentalized architecture for Context-Sensitive Stance Classification.
topic Geology
Cancer
Science Policy
Biological Sciences not elsewhere classified
Information Systems not elsewhere classified
including opinion polling
dimensional space using
detection models using
stance detection models
new country contexts
different country contexts
user &# 8217
predict user stances
political stance detection
extrapolate stance predictions
country model performance
impact social applications
extrapolation </ p
political stance
every country
based user
social context
extrapolation power
wide range
tweet encoders
scale weakly
important task
hate speech
future events
existing large
embed users
detecting propaganda
construct transformer