TUGraz OPEN Library

Zingl, Manuel

Nuss, Martin

Bauernfeind, Daniel

Aichhorn, Markus

A real-frequency solver for the Anderson impurity model based on bath optimization and cluster perturbation theory

Röck, Martin

Hollberg, Alexander

Habert, Guillaume

Passer, Alexander

LCA and BIM: Integrated Assessment and Visualization of Building Elements’ Embodied Impacts for Design Guidance in Early Stages

Benítez-Mateos, Ana

Nidetzky, Bernd

Bolivar, Juan Manuel

Lòpez-Gallego, Fernando

English

Single-Particle Studies to Advance the Characterization of Heterogeneous Biocatalysts

Behrndt, Jussi

Gesztesy, Fritz

Nakamura, Shu

Spectral shift functions and Dirichlet-to-Neumann maps

Cody, Brian

Löschnig, Wolfgang

Eberl, Alexander

Operating Energy Demand of Various Residential Building Typologies in Different European Climates

Triebl, Robert

Kraberger, Gernot J.

Mravlje, Jernej

Spin-orbit coupling and correlations in three-orbital systems

Jampour, Mahdi

Moin, Mohammad-Shahram

Yu, Lap-Fai

Bischof, Horst

Mapping Forests: A Comprehensive Approach for Nonlinear Mapping Problems

Lipp, Moritz

Aga, Misiker Tadesse

Schwarz, Michael

Gruss, Daniel

Maurice, Clémentine

Raab, Lukas

Lamster, Lukas

Nethammer

Kowald, Dominik

Lex, Elisabeth

ar
X
iv
:1
80
5.
03
06
7v
1 
 [c
s.I
R]
  8
 M
ay
 20
18
Overcoming the Imbalance Between Tag Recommendation
Approaches and Real-World Folksonomy Structures with
Cognitive-Inspired Algorithms
Dominik Kowald
Know-Center & Graz University of Technology, Austria
dkowald@know-center.at
Elisabeth Lex
Graz University of Technology, Austria
elisabeth.lex@tugraz.at
ABSTRACT
Social tagging systems enable users to collaboratively annotateWeb
resources with freely chosen keywords (i.e., tags). In order to assist
users in this annotation process, tag recommendation algorithms
have been proposed, which suggest a set of tags for a given user and
a given resource [6]. Essentially, tag recommendation algorithms
aim to help not only the individual to find appropriate tags [4] but
also the collective to consolidate the shared tag vocabulary and thus,
to reach semantic stability and implicit consensus [19].
Problem. While there is already a large body of research available
with respect to tag recommendation algorithms, most existing ap-
proaches are designed in a purely data-driven way. Consequently,
they rely on dense (i.e., broad) folksonomy structures, in which mul-
tiple posts per resource exist to have an appropriate amount of tag
assignments available to train the algorithms. In order to ensure
these dense data structures, many researchers use p-core pruning,
which was, however, shown to unnaturally bias the tag recommen-
dation evaluation process [3].
As indicated in Table 1, the folksonomy structures of real-world
social tagging systems are yet typically quite sparse (i.e., narrow),
which is reflected by the average number of posts per resource (i.e,
|P |/|R |). This degree of narrowness (i.e., sparsity) is smaller than
two in four of the six systems. Thus, in these settings, current state-
of-the-art tag recommendation approaches are not performing very
well due to the lack of available training data [10]. This leads to an
imbalance between the way how current approaches are designed
and how the folksonomies of real-world social tagging systems are
structured.
Approach. To overcome this imbalance, we propose to build upon
cognitive-inspired tag recommendation algorithms. We ground our
approach on related research on the underlying cognitive mecha-
nisms of social tagging, which has shown that the way users choose
tags for annotating resources corresponds to processes and struc-
tures in human memory [2]. In this respect, a prominent example is
the activation equation of the cognitive architecture ACT-R, which
formalizes activation processes in human memory [1]. Specifically,
the activation equation determines the probability that a specific
Permission to make digital or hard copies of part or all of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored.
For all other uses, contact the owner/author(s).
Conference’17, Washington, DC, USA
© 2016 Copyright held by the owner/author(s). 978-x-xxxx-xxxx-
x/YY/MM. . . $15.00
DOI: 10.1145/nnnnnnn.nnnnnnn
Dataset |U | |R | |T | |Y | |P | |P |/ |R |
Flickr 9,590 856,755 125,119 3,328,590 856,755 1.000
CiteULike 18,474 811,175 273,883 3,446,650 900,794 1.110
BibSonomy 10,179 683,478 201,254 2,986,396 772,108 1.129
Delicious 15,980 963,741 184,012 4,266,206 1,447,267 1.501
LastFM 1,892 12,522 9,748 186,474 71,062 5.674
MovieLens 4,009 7,601 15,238 95,580 55,484 7.299
Table 1: Statistics of publicly available datasets for Flickr, Ci-
teULike, BibSonomy, Delicious, LastFM and MovieLens. Here,
|U | is the number of users, |R | is the number of resources, |T |
is the number of tags, |Y | is the number of tag assignments, |P |
is the number of posts and |P |/|R | accounts for the degree of
narrowness (i.e., sparsity).
memory unit (e.g., a word or tag) will be needed in a specific con-
text. Thus, such an approach should enable us to efficiently model
the tag vocabulary of a user in a cognitive-plausible way without
the need for dense data structures associated with the resource to be
tagged.
Specifically, in [8, 11, 14, 18], we have implemented a novel
tag recommendation algorithm based on the activation equation of
ACT-R. This algorithm ranks a user’s tags based on three factors:
(i) past usage frequency, (ii) past usage recency, and (iii) similar-
ity with current semantic context cues. We have evaluated our ap-
proach using six real-world folksonomy datasets (i.e., Flickr, CiteU-
Like, BibSonomy, Delicious, MovieLens and LastFM) and com-
pared it to various state-of-the-art approaches using our TagRec
framework [7, 9], which is freely-available on Github1. This evalu-
ation included popularity-based approaches (i.e, MPr and MPu,r ),
Collaborative Filtering (CF), Latent Dirichlet Allocation (LDA), Folk-
Rank (FR), an alternative recency-based approach (GIRPTM) [20]
as well as our cognitive-inspired algorithm (ACT-R).
Results. The results of this evaluation by means of the nDCG@10
metric are given in Table 2 [5, 10]. We see that the performance of
the classic, data-driven tag recommendation approaches (i.e., MPr ,
MPu,r , CF, LDA, PITF and FR) strongly depends on the sparsity
of the datasets. While their performance is quite weak for sparse
(i.e., narrow) ones such as Flickr, CiteULike, BibSonomy and De-
licious, they perform very well in dense (i.e., broad) settings such
as LastFM and MovieLens. We observe a different behavior with
respect to the time-dependent approach GIRPTM. Compared to the
data-driven algorithms, this approach achieves the best results in the
narrow folksonomy Flickr and the worst in the broad folksonomies
LastFM and MovieLens. The overall best results are reached by our
cognitive-inspired ACT-R approach.
1https://github.com/learning-layers/TagRec
Dataset MPr MPu,r CF LDA PITF FR GIRPTM ACT-R
Flickr - .569 .666 .280 .535 .561 .686 .711
CiteULike .063 .392 .359 .138 .294 .392 .422 .438
BibSonomy .091 .407 .369 .219 .327 .408 .409 .434
Delicious .187 .358 .356 .271 .302 .292 .393 .431
LastFM .283 .386 .317 .388 .414 .399 .397 .425
MovieLens .271 .328 .254 .296 .324 .319 .326 .338
Table 2: Evaluation results for various tag recommendation al-
gorithms in real-world folksonomy structures by means of the
nDCG@10 metric. Our cognitive-inspired approach (i.e., ACT-
R) outperforms data-driven methods in all datasets.
Discussion. Our results demonstrate that there indeed exists an
imbalance between current state-of-the-art tag recommendation al-
gorithms and the folksonomy structures of real-world social tag-
ging systems. While these algorithms are designed for dense struc-
tures, most social tagging systems exhibit a sparse nature. To over-
come this imbalance, we showed that cognitive-inspired algorithms,
which model the tag vocabulary of a user in a cognitive-plausible
way, can be helpful. Our present approach does this via imple-
menting the activation equation of the cognitive architecture ACT-
R, which determines the usefulness of units in human memory (e.g.,
tags).
Apart from that, in [12], we have shown that ACT-R can be gen-
eralized for related use cases such as hashtag recommendations
in Twitter. For future work, we plan to build upon this results
in order to propose cognitive-inspired recommender systems (e.g.,
for resource recommendation) as an alternative to data-driven ones
[13, 15–17]. In this sense, our long-term goal is to design hybrid
approaches, which combine the advantages of both worlds in order
to adapt to the current setting (i.e., sparse vs. dense ones).
Acknowledgments. The authors would like to thank Stefanie Lind-
staedt, Paul Seitlinger, Simone Kopeinik and Tobias Ley for sup-
porting this research. This work is funded by the Know-Center Graz
and the H2020 project AFEL (grant agreement: 687916).
KEYWORDS
Tag Recommendations; Data Sparsity; Folksonomy Structures; ACT-
R; Recommender Evaluation
REFERENCES
[1] John R Anderson, Daniel Bothell, Michael D Byrne, Scott Douglass, Christian
Lebiere, and Yulin Qin. 2004. An integrated theory of the mind. Psychological
review 111, 4 (2004), 1036.
[2] Ulrike Cress, Christoph Held, and Joachim Kimmerle. 2013. The collective
knowledge of social tags: Direct and indirect influences on navigation, learning,
and information processing. Computers & Education 60, 1 (2013), 59–73.
[3] Stephan Doerfel and Robert Ja¨schke. 2013. An analysis of tag-recommender
evaluation procedures. In Proceedings of the 7th ACM conference on Recom-
mender systems. ACM, 343–346.
[4] Robert Ja¨schke, Leandro Marinho, Andreas Hotho, Lars Schmidt-Thieme, and
Gerd Stumme. 2008. Tag recommendations in social bookmarking systems. Ai
Communications 21, 4 (2008), 231–247.
[5] Simone Kopeinik, Dominik Kowald, and Elisabeth Lex. 2016. Which Algo-
rithms Suit Which Learning Environments? A Comparative Study of Recom-
mender Systems in TEL. In Proceedings of the 11th European Conference on
Technology Enhanced Learning (ECTEL ’16). Springer, 124–138.
[6] Dominik Kowald. 2015. Modeling Cognitive Processes in Social Tagging to
Improve Tag Recommendations. In Proceedings of the 24th International Con-
ference on World Wide Web (WWW ’15 Companion). International World Wide
Web Conferences Steering Committee, Republic and Canton of Geneva, Switzer-
land, 505–509. DOI:http://dx.doi.org/10.1145/2740908.2741746
[7] Dominik Kowald, Simone Kopeinik, and Elisabeth Lex. 2017. The TagRec
Framework As a Toolkit for the Development of Tag-Based Recommender Sys-
tems. In Adjunct Publication of the 25th Conference on User Modeling, Adapta-
tion and Personalization (UMAP ’17). ACM,New York, NY, USA, 23–28. DOI:
http://dx.doi.org/10.1145/3099023.3099069
[8] Dominik Kowald, Simone Kopeinik, Paul Seitlinger, Tobias Ley, Dietrich Albert,
and Christoph Trattner. 2015. Refining frequency-based tag reuse predictions by
means of time and semantic context. In Mining, Modeling, and Recommend-
ing’Things’ in Social Media. Springer, 55–74.
[9] Dominik Kowald, Emanuel Lacic, and Christoph Trattner. 2014. TagRec: To-
wards a Standardized Tag Recommender Benchmarking Framework. In Pro-
ceedings of the 25th ACM Conference on Hypertext and Social Media (HT
’14). ACM, New York, NY, USA, 305–307. DOI:http://dx.doi.org/10.1145/
2631775.2631781
[10] Dominik Kowald and Elisabeth Lex. 2015. Evaluating Tag Recommender Al-
gorithms in Real-World Folksonomies: A Comparative Study. In Proceedings
of the 9th ACM Conference on Recommender Systems (RecSys ’15). ACM, New
York, NY, USA, 265–268. DOI:http://dx.doi.org/10.1145/2792838.2799664
[11] Dominik Kowald and Elisabeth Lex. 2016. The Influence of Frequency, Rec-
ency and Semantic Context on the Reuse of Tags in Social Tagging Systems. In
Proceedings of the 27th ACM Conference on Hypertext and Social Media (HT
’16). ACM, New York, NY, USA, 237–242. DOI:http://dx.doi.org/10.1145/
2914586.2914617
[12] Dominik Kowald, Subhash Chandra Pujari, and Elisabeth Lex. 2017. Tempo-
ral Effects on Hashtag Reuse in Twitter: A Cognitive-Inspired Hashtag Recom-
mendation Approach. In Proceedings of the 26th International Conference on
World Wide Web (WWW ’17). International World Wide Web Conferences Steer-
ing Committee, Republic and Canton of Geneva, Switzerland, 1401–1410. DOI:
http://dx.doi.org/10.1145/3038912.3052605
[13] Dominik Kowald, Paul Seitlinger, Simone Kopeinik, Tobias Ley, and Christoph
Trattner. 2015. Forgetting the words but remembering the meaning: Modeling
forgetting in a verbal and semantic tag recommender. In Mining, Modeling, and
Recommending’Things’ in Social Media. Springer, 75–95.
[14] Dominik Kowald, Paul Seitlinger, Christoph Trattner, and Tobias Ley. 2014.
Long Time No See: The Probability of Reusing Tags As a Function of Frequency
and Recency. In Proceedings of the 23rd International Conference on World Wide
Web (WWW ’14 Companion). International World Wide Web Conferences Steer-
ing Committee, Republic and Canton of Geneva, Switzerland, 463–468. DOI:
http://dx.doi.org/10.1145/2567948.2576934
[15] Emanuel Lacic, Dominik Kowald, Lukas Eberhard, Christoph Trattner, Denis
Parra, and Leandro Balby Marinho. 2015. Utilizing online social network and
location-based data to recommend products and categories in online market-
places. In Mining, Modeling, and Recommending’Things’ in Social Media.
Springer, 96–115.
[16] Emanuel Lacic, Dominik Kowald, Paul Seitlinger, Christoph Trattner, and De-
nis Parra. 2014. Recommending Items in Social Tagging Systems Using Tag
and Time Information. In Proceedings of the 1st International Workshop on So-
cial Personalisation co-located with the 25th ACM Conference on Hypertext and
Social Media (HT ’14). CEUR-WS.
[17] Paul Seitlinger, Dominik Kowald, Christoph Trattner, and Tobias Ley. 2013. Rec-
ommending tags with a model of human categorization. In Proceedings of the
22nd ACM international conference on Conference on information and knowl-
edge management (CIKM ’13). ACM, New York, NY, USA, 2381–2386. DOI:
http://dx.doi.org/10.1145/2505515.2505625
[18] Christoph Trattner, Dominik Kowald, Paul Seitlinger, Tobias Ley, and Simone
Kopeinik. 2016. Modeling Activation Processes in Human Memory to Predict
the Use of Tags in Social Bookmarking Systems. The Journal of Web Science 2,
1 (2016), 1–18. DOI:http://dx.doi.org/10.1561/106.00000004
[19] Claudia Wagner, Philipp Singer, Markus Strohmaier, and Bernardo A Huberman.
2014. Semantic stability in social tagging streams. In Proceedings of the 23rd
international conference on World Wide Web. ACM, 735–746.
[20] Lei Zhang, Jian Tang, and Ming Zhang. 2012. Integrating temporal usage pattern
into personalized tag prediction. In Web Technologies and Applications. Springer,
354–365.


Overcoming the Imbalance Between Tag Recommendation Approaches and Real-World Folksonomy Structures with Cognitive-Inspired Algorithms

Sorantin, Max Erich

Arrigoni, Enrico

Dorda, Antonius

Held, Karsten

Impact ionization processes in the steady state of a driven Mott insulating layer coupled to metallic leads

Graz University of Technology

Not a member yet