=Paper=
{{Paper
|id=Vol-2491/abstract30
|storemode=property
|title=None
|pdfUrl=https://ceur-ws.org/Vol-2491/abstract30.pdf
|volume=Vol-2491
|dblpUrl=https://dblp.org/rec/conf/bnaic/VandeveldeDC19
}}
==None==
https://ceur-ws.org/Vol-2491/abstract30.pdf
An Interactive Knowledge Base Application for
Group Assignment
Simon Vandevelde, Kylian Van Dessel, and Herman Crauwels
KU Leuven, De Nayer Campus, Dept. of Computer Science
{s.vandevelde, kylian.vandessel, herman.crauwels}@kuleuven.be
The aim of this work is to research how well the IDP system[1], developed
at KU Leuven, functions in an interactive context. The IDP system is an im-
plementation of the knowledge base paradigm (KBP), meaning that it stores its
information in a knowledge base, to which it can apply inference methods to
solve problems. An advantage of such a system is the reuse of the knowledge
base. Using the same information, different inference methods can be used to
achieve different goals. Recently, research has been done into the interactivity of
the IDP system[2]. To further investigate the system’s applicability in an inter-
active context, we developed an application for group assignment using IDP as
backend. The main challenge was ensuring the scalability of this implementation.
The research is divided in two parts. In the first part, we focus on how
to implement group assignment using the IDP system. In the second part, we
develop an application around this implementation in the IDP system. At KU
Leuven Campus De Nayer, each year around 150 students sign up. These students
are assigned to groups based on their background and personal preferences,
allowing them to feel more comfortable in their new environment. This process
is currently done manually, and is not time efficient. Our implementation in the
IDP system is subdivided in a coarse initial group assignment and a refined
incremental assignment.
The initial group assignment forms groups based on residence and former
high school, stimulating students to carpool to campus. Additionally, because
students usually prefer to stay with their former classmates, most preferences
will already be fulfilled. After starting with a naive implementation, we added
improvements to increase its time-efficiency. Some of these improvements are:
avoiding symmetry (group layouts that appear different, but are logically the
same), calculating the distances between students on the fly instead of inputting
a distance matrix, and choosing the optimal precision for the distance.
The incremental assignment recalculates an existing assignment to fulfill spe-
cific student preferences. In this way, the assignment can be updated incremen-
tally each time new preferences are received. For each preference, the user decides
on a priority, which is then used as a weight for the soft constraints in the opti-
mization problem. For instance, the priority of two students wanting the same
group could be lower than the priority of 2 students not wanting to be together.
Copyright c 2019 for this paper by its authors. Use permitted under Creative Com-
mons License Attribution 4.0 International (CC BY 4.0).
2 Simon Vandevelde, Kylian Van Dessel, and Herman Crauwels
The IDP implementation minimizes the number of violated preferences, weighted
by their priorities.
Fig. 2. Duration of incremental assign-
Fig. 1. Duration of initial assignment. ment.
For the second part, we need to be able to interface with the IDP system.
The application uses PyQt5 to construct a Graphical User Interface (GUI) and
the Pyidp3 API[3] to communicate with the IDP system. This way, the IDP
system controls the intelligence of our application, whilst the GUI provides a
user-friendly system. It is important to work interactively, so it is possible to
change the system parameters for making a group layout. For example, the
desired group size and the number of groups can be altered. For the incremental
assignment, it is possible to add preferences, with a priority, to calculate multiple
possible solutions and to show the differences between the current and a newly
calculated layout.
As shown in fig 1, the initial assignment does not scale well. The implementa-
tion relies heavily on mathematical calculation-intensive tasks, by which it loses
the advantage of the IDP system. However, this long calculation time can be
overcome by limiting the amount of time we allow the IDP system to run. This
way the result is not ideal, but it is satisfactory seeing as it is only a starting
point to use in the incremental assignment. The incremental assignment on the
other hand succeeds in recalculating a group layout with up to 100 preferences
within 10 seconds, as shown in figure 2. We prove that in a realistic setting, i.e.
considering a realistic amount of preferences, our system provides satisfactory
results. The system will be used by the planner at the campus next year.
References
1. Bogaerts, B., Bruynooghe, M., Janssens, G., Denecker, M.: Predi-
cate logic as a modelling language: The IDP system. arXiv.org (2018),
http://search.proquest.com/docview/2071739773/
2. Van Hertum, P., Dasseville, I., Janssens, G., Denecker, M.: The KB paradigm and its
application to interactive configuration. Theory and Practice of Logic Programming
17(1), 91–117 (2017), http://search.proquest.com/docview/1853690223/
3. Vennekens, J.: Lowering the learning curve for declarative programming: A python
API for the IDP system. vol. 10137, pp. 86–102. Springer Verlag (2017)