Embed Size (px)
Citation preview
We additionally investigated combinations of the above hypotheses, such as user-tfidf, where users navigate semantically on their own resources. Legend: 1. Colored items indicate ownership. 2. : user page; : resource page; : tag page; an arrow indicates allowed navigation.
Interpreting Navigation Behavior
in a Social Tagging System1 1 1 2 1,3Thomas Niebler , Martin Becker , Daniel Zoller , Stephan Doerfel , Andreas Hotho
(1)Data Mining and Information Retrieval Group, University of Würzburg, Germany ,(2) (3)
KDE Group, University of Kassel, Germany , L3S Research Center, Hanover, Germany
IRDM
NFOR-
ETRIEVALATAININGGROUP
MATION
FolkTrails
Results
Conclusion
Hypotheses about NavigationUniform Hypothesis (uniform)
Similar to the hypothesis above, another incentive for navigating in one‘s own pages might be the search for already saved resources, which are semantically similar.
The Uniform Hypothesis serves as the baseline.It assumes that users navigate randomly.
• All of the basic hypotheses explain the observed transitions better than the baseline, which indicates at least some structural properties explaining the observed transitions.
• The good performance of the semantic hypotheses indicates that semantic similarity of pages is a strong factor for navigation in BibSonomy. However, users tend to mostly navigate on their own pages.
• Overall, the combination of the user consistent and the semantic hypothesis performs best, indicating that navigation on BibSonomy can mainly be explained by semantic navigation within the resources of a specific user.
Overall Request Log Dataset
Request Subset: Usage Continuity
Request Subset: Inside vs. Outside Navigation
MotivationThe Navigation Behaviour of users on Web 2.0 systems is largely unexplored. This is however important to effectively analyze and e.g. improve website structures for better content provision.
In this work, we focus on the analysis of navigation in Social Tagging Systems, specifically in BibSonomy. Social Tagging Systems have been in the focus of research for many years now and interesting results about the generated content entities have been presented, exhibiting the great semantic exploitability of such systems. Still, how users navigate such systems is largely unknown.
We provide and analyze several hypotheses about the incentives of user navigation in the Social Tagging System BibSonomy, thus shedding light on how users consume the provided content.
Technical BackgroundHypTrails is a Bayesian approach for comparing hypotheses about human trails on the web. It models transition data in first-order Markov chains and represents a transition dataset as a matrix of transition counts.
Folksonomies are the tripartite structures behind social tagging systems and are comprised of users, resources and tags as well as a set of tag assignments.
lll
l
l
l
l
l
l
l
l
l
3e+06
2.5e+06
2e+06
1.5e+06
0 1 2 3 4 5
K
Marg
inal l
ikelih
ood
l
l
cat
folk
folk−user
folk−tfidf
user
user−tfidf
page
tfidf
uniform
ll l
l
l
l
l
l
l
l
l
l
5e+05
4e+05
3e+05
2e+05
0 1 2 3 4 5
K
Ma
rgin
al l
ike
liho
od
l
l
cat
folk
folk−user
folk−tfidf
user
user−tfidf
page
tfidf
uniform
lll
l
l
l
l
l
l
l
l
l
5e+05
4.5e+05
4e+05
3.5e+05
0 1 2 3 4 5
K
Ma
rgin
al l
ike
liho
od
l
l
cat
folk
folk−user
folk−tfidf
user
user−tfidf
page
tfidf
uniform
• In this work, we studied a large dataset of webserver logs from the social tagging system BibSonomy in order to analyze and understand components of user navigation in social tagging systems.
• Our results show that there is a strong semantic component inherent in user navigation. On subsets of the set of request logs, we can also show that users navigating outside their own resources largely follow the folksonomy structure.
• Overall, we were able to gain new insights into the underlying processes of navigation in tagging systems, which can be extended and leveraged in the future, for example, by considering new hypotheses, improving navigation experience or extracting the latent semantic information.
Motivated by the fact that users often navigate on their own pages, we investigate whether users behave differently when they are browsing the folksonomy outside of their own pages.
• The hypothesis which best explains the behaviour outside one‘s own resources is the folksonomy consistent & semantic navigation hypothesis.
• This indicates that users utilize the folksonomy structure combined with a semantic incentive, i.e. rather following semantically related pages, when discovering new resources.
Since we expect users to adapt to systems they are using, we investigate if their navigation behaviour changes over time.
• The semantic hypothesis performs significantly better than the folksonomy hypothesis. We may thus observe a learning process: Short-term users are not as adapted to the folksonomy structure as long-term users and thus rely on their semantic intuition.
• The page consistent and folksonomy consistent & user consistent hypotheses explain navigation equally well. This might be due to increased use of pagination as well as a lack of personal resources.
User Consistent Hypothesis (user) Page Consistent Hypothesis (page) Category Consistent Hypothesis (cat)
Folksonomy Consistent Hypothesis (folk) Semantic Navigation Hypothesis (tfidf)
Semantic Navigation& User Consistent Hypothesis
Semantic Navigation& Folksonomy Consistent HypothesisBecause the fo lksonomy structure itself does contain large semantic value, we expect this hypothesis to improve to the pure folksonomy consistent hypothesis.
Because tagging data contain a lot of semantic information, we expect users to prefer pages, which are semantically stronger related.
Because BibSonomy is based on the Folksonomy structure, we expect users to only follow links provided by the Folksonomy.
A relaxation of the page consistent hypothesis, we assume that users tend to stay on pages with the same category, e.g. only on user pages.
In previous studies it was found that users often stay on the same pages, e.g. due to pagination effects.
Many people make use of BibSonomy as a place to save their resources. Thus, they navigate mainly on their own pages.
DatasetsWe analyze a large set of content and log data from the Social Tagging System BibSonomy, spanning from 2006 to 2012.
User and Content DatasetWe use the Folksonomy data from nonspammers with their respective resources and tags. 17,932 users were explicitly classified as nonspammers. They created 456,777 bookmark posts and 2,410,844 publication posts using 65,228 distinct tags, which have been at least used twice.
Request Log DatasetThe BibSonomy log files include all HTTP requests to the system. After filtering, the remaining dataset contains 103,415 distinct visited content pages, i.e. pages which show at least one resource. We recorded 327,060 transitions between these pages. 123,452 transitions were self-transitions (i.e., transitions from a page to itself) and 261,300 were transitions, where the logged in user owns both the source and the target page.
Request Log SubsetsWe additionally investigated outside navigation, i.e., where users navigate outside their own pages. The subset of those requests counts 42,193 requests. and effects of usage continuity. The subset of requests by short-term users, i.e. who used the system less than half a year, is comprised of 48,221 requests.
As navigating on one‘s own pages resembles navigation in a res t r i c ted sys tem, we assumed that users utilize the folksonomy structure there.
Folksonomy Consistent& User Consistent Hypothesis
increasing concentration factor /“believe” (k)
mar
gin
al li
kelih
oo
d /
ev
iden
ce (
e)
Bayes factor
Marginal likelihood / evidence curves
User Tags
Resource
Marginal likelihood
We analyzed navigation on the overall request log dataset and on two subsets: Outside navigation and short-term usage.
0.8
0.1
0.2
0.2
0.5
0.60.9
0.5
0.1
1.0
1.0
1.0
1.0 1.0
1.0
Thomas Niebler [email protected] http://dmir.org/staff/niebler @thomasniebler http://www.bibsonomy.org/user/thoni
Martin Becker [email protected] http://dmir.org/staff/becker @mgbckr http://www.bibsonomy.org/user/becker
Daniel Zoller [email protected] http://dmir.org/staff/zoller @nosebrain http://www.bibsonomy.org/user/nosebrain
Stephan Doerfel [email protected] https://www.kde.cs.uni-kassel.de/doerfel http://www.bibsonomy.org/user/sdo
Andreas Hotho [email protected] http://dmir.org/staff/hotho @hotho http://www.bibsonomy.org/user/aho
![Tag Relatedness in Image Folksonomies · folksonomies. According to the degree of user collaboration, folksonomies are classified in two main categories: broad and narrow [1]. In](https://img.pdfslide.us/doc/110x75/60508dfb391e7c39bf133dec/tag-relatedness-in-image-folksonomies-folksonomies-according-to-the-degree-of-user.jpg)
