Information Technology Reference
In-Depth Information
LoR reflects better the learning process depicted in the questions sequences
depicted during our experiment.
4 Summary and Future Work
In this paper, a new recommendation technique, called learning-oriented rec-
ommender (LoR), is introduced with the goal to improve the user's learning
experience while interacting with a QA system.
The evaluation of the learning-oriented recommender is not as easy task for
at least two reasons. First, in order to train and learn the recommender model, a
substantial history of user learning activity is needed, which is not influenced in
any way by other recommenders or other external factors. Secondly, even if such
question sequences that reflect the users learning process were to be collected,
there is no clear, well established metric to evaluate the performance of the
recommender from a learning perspective.
However, a first step was made towards a better understanding of the
learning-oriented recommender's capabilities. From each of the above mentioned
three datasets of questions (i.e. earth sciences, nutrition and homeschooling), five
sets of 20 questions were randomly selected and users were asked to order each
set according to their learning preferences in the sequence that they, personally,
would ask them or want to learn about.
Five evaluation measures were used to compare the performance of the LoR
with a simple VLMC (SR) over the question sequences and a random recom-
mender (RR). Results show that while the SR outperforms the LoR with respect
to prediction power, the LoR achieved a much higher coverage and learning util-
ity. The RR was used as a base reference. The obtained results confirm our
initial intuition: question sequences are first influenced by the underlying topics
and their order, and then, within each topic, by a particular order of learning
objectives.
However, further evaluation is required to show that the LoR has great
potential in offering an improved user learning experience. To show this in more
detail, we intend to conduct an online user study. Additionally, we also plan to
analyze the influence of the knowledge-based structure on the recommendation
performance.
Additionally, it would be desirable to investigate the potential of an auto-
matic topic-tree generation, and, more importantly the automatic assignment of
questions to topics and to learning objectives.
With the information overload, new aspects of existing disciplines are iden-
tified or entirely unknown, unexplored fields of study are discovered. In the first
case, a restructuring or extension of the current curriculum is required. The
second case demands the settlement of the first building blocks. Learning pat-
terns represent relevant knowledge about these domains. By using the learning
patterns derived from our recommender model, we could establish new fields of
study and (semi-)automatically generate curricula for these domains. Further
research in this direction is expected to answer the question whether and how
to exploit the learning-oriented recommender model for this purpose.
Search WWH ::
Custom Search