=Paper=
{{Paper
|id=Vol-512/paper-12
|storemode=property
|title=Using Domain Models for Context-Rich User Logging
|pdfUrl=https://ceur-ws.org/Vol-512/paper12.pdf
|volume=Vol-512
|dblpUrl=https://dblp.org/rec/conf/sigir/DignumKKSFR09
}}
==Using Domain Models for Context-Rich User Logging==
https://ceur-ws.org/Vol-512/paper12.pdf
Using Domain Models for Context-Rich User Logging ∗
Stephen Dignum Yunhyong Kim Udo Kruschwitz
School of Computer Science School of Computing School of Computer Science
and Electronic Engineering Robert Gordon University and Electronic Engineering
University of Essex Aberdeen, United Kingdom University of Essex
Colchester, United Kingdom y.kim1@rgu.ac.uk Colchester, United Kingdom
sandig@essex.ac.uk udo@essex.ac.uk
Dawei Song Maria Fasli Anne De Roeck
School of Computing School of Computer Science Department of Mathematics
Robert Gordon University and Electronic Engineering and Computing
Aberdeen, United Kingdom University of Essex Open University
d.song@rgu.ac.uk Colchester, United Kingdom Milton Keynes, United
mfasli@essex.ac.uk Kingdom
a.deroeck@open.ac.uk
ABSTRACT consequently, a considerable amount of time is spent by users
This paper describes the prototype interactive search sys- trying to learn the domain characteristics even before they
tem being developed within the AutoAdapt project1 . The are able to identify the adequate questions to be submitted
AutoAdapt project seeks to enhance the user experience in to a search system. In the AutoAdapt project, we hope to
searching for information and navigating within selected do- analyse and accelerate this learning process by implement-
main collections by providing structured representations of ing a system that presents and logs several domain model
domain knowledge to be directly explored, logged, adapted representations in response to each stage of a user’s logged
and updated to reflect user needs. We propose that this search activity. By encouraging and logging the direct inter-
structure is a valuable stepping-stone in context-rich log- action of users with domain model representations, collective
ging of user activities within the information seeking en- domain user behaviour can be understood within context.
vironment. Here we describe the primary components that The analysed user needs can be incorporated back into the
have been implemented and the user interactions that it will system to adapt domain knowledge representations that are
support. presented to the users, creating a continuous feedback loop.
Provision of domain model knowledge has been shown to
Categories and Subject Descriptors aid user search for the information they need [3]. A domain
H.3.3 [Information Search and Retrieval]: Query For-
model is effectively a structure that characterises the do-
mulation; H.5.2 [User Interfaces]: Natural Language; I.2.7
main dataset from the domain user perspective, e.g. a graph
[Natural Language Processing]: Text Processing
where nodes represent domain concept terms and edges be-
tween nodes their relationship, possibly weighted to express
General Terms how specific the term is or how closely related the terms are
Domain Model, Graph Traversal, User Logging within the collection.
1. INTRODUCTION One of the difficulties in using traditional logging of user
activity, such as submitted query terms, URL clicks, and
Searches within document collections like intranets differ
page viewing time, to adapt search systems is the lack of
from those within the general World Wide Web [6]. The
sufficient context for identifying the user actions that are
terminology, structure, and services provided within an in-
truly relevant to the user’s information need. We implement
tranet are selected to meet organisational requirements, and,
methods of explicitly visualising domain models to accom-
∗Copyright is held by the author/owner(s). pany each search step, in addition to a list of links to search
SIGIR’09, July 19-23, 2009, Boston, USA. results, and a set of query term suggestions. By concur-
1 rently logging user interaction with the these components
http://AutoAdaptProject.org
we have a mechanism to enable subsequent weblog analysis.
For example, different document selections following the ex-
ploration of the same path may indicate relevance between
documents, different paths leading to the same document
may indicate relationships between paths, a comparison of
path before and after a document selection should yield some
understanding of the nature of the document selection.
We present here a working system including a graphical do-
main model presentation, a document list and term sugges- main model presented to the user. As we intend to mod-
tions designed to capture the described information. ify the domain model over time based on responses to the
model presented, it is essential that a complete copy of the
presented model segment is retained in the database. Of
2. RELATED WORK particular interest is the term positioned at the centre of
It is frequently pointed out that users are reluctant to leave
the graph and the co-ordinates of the other terms. Using
any explicit feedback when they search a document collec-
this information and the term clicks we can determine how
tion. However, implicit feedback, e.g. the analysis of log
the model was traversed, allowing us to identify which terms
records, has been shown to be good at approximating ex-
were also visible and ignored. Suggested terms (derived by
plicit feedback. For example, users often reformulate their
the model) are also recorded along with any selection (to
query and such patterns can help in learning an improved
expand, or replace initially submitted query terms).
ranking function [2]. The same methods have shown to im-
prove an adaptive domain model on a local Web site created
The logging structure allows us to record a number of user
using formal concept analysis lattice structures [4].
decisions without the need for explicit feedback. For exam-
ple, the selection of a term in a domain model can provide
It has already been evidenced that users want support in
a ranking of terms, i.e., above those shown but not selected.
selecting search words for query formulation but also it has
Also, suggested terms derived from a particular traversal can
been recognised that they want to stay in control with re-
be ordered. In addition, we can compare sessions that have
spect to making the final decision to submit a query [8].
resulted in the same URL being selected in order to capture
Furthermore, it has been noted that users like to be pro-
related terms or similar portions of the domain model. It is
vided with system-guided query suggestions even if sugges-
also possible to compare portions of different domain models
tions are not relevant to the current query [7]. Users have
to discover missed relationships or terms.
shown signs of being more inclined, in a search environment
that supports navigation, to submit new queries, or resub-
mit modified queries, than to navigate away from the result 5. FUTURE WORK
set [5]. Finally, increased activity in developing interactive As the next step, we propose to test the infrastructure in
features in search systems used across existing popular Web this document across several domain collections and model
search engines suggests that interactive systems are being creation/adaptation algorithms to extensively evaluate the
recognised as a promising next step in assisting information effectiveness of the system in capturing the context of user
search. The work proposed in this paper is very much in interaction.
line with what Belkin calls the challenge of all challenges in
IR at the moment, to move beyond the limited, inherently 6. ACKNOWLEDGEMENTS
non-interactive models of IR to truly interactive systems [1]. AutoAdapt is funded by EPSRC grants EP/F035357/1 and
EP/F035705/1. The JIT visualisation toolkit2 was used for
the domain model visualisation.
3. USER INTERFACE
In figure 1 we can see a screenshot of our demonstration sys-
tem. There are four basic components, a) a simple entry box 7. REFERENCES
for query terms, b) a list of URLs with associated snippets, [1] N. J. Belkin. Some(what) grand challenges for
c) a graph displaying a segment of a domain model, and d) information retrieval. SIGIR Forum, 42(1):47–54, 2008.
a list of suggested terms for query refinement. [2] T. Joachims and F. Radlinski. Search engines that
learn from implicit feedback. IEEE Computer,
The user enters a set of query terms, this results in a num- 40(8):34–40, 2007.
ber of documents being returned. Using the query terms, [3] U. Kruschwitz and H. Al-Bakour. Users Want More
additional terms are automatically extracted, e.g., from the Sophisticated Search Assistants - Results of a
domain model and the highest ranked documents. These Task-Based Evaluation. JASIST, 56(13):1377–1393,
terms are then represented as nodes in a graph as a segment November 2005.
of the domain model. The user can then traverse the graph [4] D. Lungley and U. Kruschwitz. Automatically
by clicking on the nodes (the effect is to make that term the maintained domain knowledge: Initial findings. In
centre of the graph). On term selection the list of suggested Proceedings of ECIR, pages 739–743, 2009.
terms is updated to show terms closely related to the se- [5] M. Mat-Hassan and M. Levene. Associating Search and
lected term. The user can then add the term to the existing Navigation Behavior Through Log Analysis. JASIST,
query or use it as a new query. 56(9):913–934, 2005.
[6] M. White. Making Search Work: Implementing Web,
The modular nature of the software allows a standard user Intranet and Enterprise Search. Facet Publishing, 2007.
interface and logging structure (described in the next sec- [7] R. W. White, M. Bilenko, and S. Cucerzan. Studying
tion) irrespective of the domain model creation and adap- the Use of Popular Destinations to Enhance Web
tation algorithms employed. We can, for example, examine Search Interaction. In Proceedings of SIGIR’07, pages
different interaction styles and evaluate other domain model 159–166, Amsterdam, 2007.
visualisation tools. [8] R. W. White and I. Ruthven. A Study of Interface
Support Mechanisms for Interactive Information
4. LOGGING INFORMATION Retrieval. JASIST, 57(7):933–948, 2006.
In addition to logging user query terms with presented and 2
http://blog.thejit.org/javascript-information-
selected URLs it was decided to log the segment of the do- visualization-toolkit-jit
Figure 1: Screenshot of AutoAdapt Demo System.