Upload
silvio-peroni
View
409
Download
4
Tags:
Embed Size (px)
DESCRIPTION
This work presents some experiments in letting humans annotate citations according to CiTO, an OWL ontology for describing the function of citations. We introduce a comparison of the performance of different users, and show strengths and difficulties that emerged when using that particular model to characterise citations of scholarly articles.
Citation preview
http://creativecommons.org/licenses/by-sa/3.0
Characterising Citations in Scholarly Articles: an Experiment
Paolo Ciancarini – [email protected] Di Iorio – [email protected]
Andrea Giovanni Nuzzolese – [email protected] Peroni – [email protected]
Fabio Vitali – [email protected] Workshop on Artificial Intelligence and Cognitionco-located with XIII Conference of the Italian Association for Artificial IntelligenceTurin (Italy), December 3, 2013
What and Why
What and Why
• What: to analyse how humans use a particular citation model for the annotation of citations in scientific article
✦ Citation model under investigation: the Citation Typing Ontology (CiTO), an OWL-2 DL ontology describing various functions of citation (e.g. uses method in, extends, critiques)
What and Why
• What: to analyse how humans use a particular citation model for the annotation of citations in scientific article
✦ Citation model under investigation: the Citation Typing Ontology (CiTO), an OWL-2 DL ontology describing various functions of citation (e.g. uses method in, extends, critiques)
• Why: final aims (of future works): ✦ To improve CiTO by adding new properties, and by creating cluster of
semantically-similar properties (e.g. disagrees with and critiques) by analysing how humans use them when dealing with concrete tasks of annotation
✦ To improve CiTalO, a web application that tries to infer citation functions from sentences containing a citation, so as to reflect human behaviour when annotating citations
Who
users
Who
users
The metal m
odelsw
ere derived from
Who
citation model (CiTO)mentalmodel
mentalmodel
mentalmodel
mentalmodel
mentalmodel
users
The metal m
odelsw
ere derived from
Who
citation model (CiTO)mentalmodel
mentalmodel
mentalmodel
mentalmodel
mentalmodel
users
author
“It extends the research outlined in earlier work [3]”
Interpretation of author’s
text
The metal m
odelsw
ere derived fromMapping personal interpretation with citation functions
Who
citation model (CiTO)mentalmodel
mentalmodel
mentalmodel
mentalmodel
mentalmodel
users
author
“It extends the research outlined in earlier work [3]”
Interpretation of author’s
text
How, When and Where
How, When and Where
• How: preliminary user testing session with users to whom we asked to assign CiTO properties to the citations in the Proceedings of Balisage 2011:
✦ 18 papers✦ 104 citations selected (out of 377)✦ 5 users
How, When and Where
• How: preliminary user testing session with users to whom we asked to assign CiTO properties to the citations in the Proceedings of Balisage 2011:
✦ 18 papers✦ 104 citations selected (out of 377)✦ 5 users
• When: well, few months ago, but we are still experimenting involving a larger set of users and using different experimental settings
How, When and Where
• How: preliminary user testing session with users to whom we asked to assign CiTO properties to the citations in the Proceedings of Balisage 2011:
✦ 18 papers✦ 104 citations selected (out of 377)✦ 5 users
• When: well, few months ago, but we are still experimenting involving a larger set of users and using different experimental settings
• Where: the whole test was/is performed online, without any supervision
How, When and Where
• How: preliminary user testing session with users to whom we asked to assign CiTO properties to the citations in the Proceedings of Balisage 2011:
✦ 18 papers✦ 104 citations selected (out of 377)✦ 5 users
• When: well, few months ago, but we are still experimenting involving a larger set of users and using different experimental settings
• Where: the whole test was/is performed online, without any supervision
By the way, would you like to help us?
How, When and Where
• How: preliminary user testing session with users to whom we asked to assign CiTO properties to the citations in the Proceedings of Balisage 2011:
✦ 18 papers✦ 104 citations selected (out of 377)✦ 5 users
• When: well, few months ago, but we are still experimenting involving a larger set of users and using different experimental settings
• Where: the whole test was/is performed online, without any supervision
By the way, would you like to help us?
Please ... ;-( ...
Results
• Users have selected 34 different CiTO properties over 40✦ average: 22.4 properties per user
• Properties used many times✦ cites for information (110)✦ cites as related (39)✦ cites as data source (38)
• Low positive agreement for the 5 raters over all 104 subjects✦ k = 0.16
• Moderate positive local agreement on ✦ cites as data source: k = 0.5✦ cites as potential solution: k = 0.45✦ cites as recommended reading: k = 0.34✦ includes quotation from: k = 0.49
Results
• Users have selected 34 different CiTO properties over 40✦ average: 22.4 properties per user
• Properties used many times✦ cites for information (110)✦ cites as related (39)✦ cites as data source (38)
• Low positive agreement for the 5 raters over all 104 subjects✦ k = 0.16
• Moderate positive local agreement on ✦ cites as data source: k = 0.5✦ cites as potential solution: k = 0.45✦ cites as recommended reading: k = 0.34✦ includes quotation from: k = 0.49
Summarising: it is a difficult task for
humans
Results
• Users have selected 34 different CiTO properties over 40✦ average: 22.4 properties per user
• Properties used many times✦ cites for information (110)✦ cites as related (39)✦ cites as data source (38)
• Low positive agreement for the 5 raters over all 104 subjects✦ k = 0.16
• Moderate positive local agreement on ✦ cites as data source: k = 0.5✦ cites as potential solution: k = 0.45✦ cites as recommended reading: k = 0.34✦ includes quotation from: k = 0.49
Summarising: it is a difficult task for
humans
Which reminds me...
... that we are looking for users for
additional tests
Results
• Users have selected 34 different CiTO properties over 40✦ average: 22.4 properties per user
• Properties used many times✦ cites for information (110)✦ cites as related (39)✦ cites as data source (38)
• Low positive agreement for the 5 raters over all 104 subjects✦ k = 0.16
• Moderate positive local agreement on ✦ cites as data source: k = 0.5✦ cites as potential solution: k = 0.45✦ cites as recommended reading: k = 0.34✦ includes quotation from: k = 0.49
Summarising: it is a difficult task for
humans
Which reminds me...
Would you like to help us?Please ... ;-( ...
... that we are looking for users for
additional tests
Thanks for your attention