Learning-Oriented Question Recommendation Using Bloom’s Learning Taxonomy and Variable Length Hidden Markov Models - Transactions on Large-Scale Data-and Knowledge-Centered Systems - page 37

Intuition: The SR model is based on the intuition that the unique question
identifiers are enough to eciently identify learning patterns within question
sequences and use them to produce accurate predictions.

The Random Recommender (RR) is a model that simply recommends
questions randomly, without relying on any knowledge or user history log
when producing recommendations. This model was considered only as base
comparison to show that the previous two recommendation models (LoR and
SR) are not random and that they actually outperform greatly the RR model.

The following section will show the performance of each of these recom-
mendation models on three different knowledge bases.

3 Evaluation and Results

3.1 Data

In order to thoroughly test the performance of the LoR model, three data sets
of questions, corresponding to three different domains, were collected: earth sci-
ences (from Wiki Answers 1 and MadSci 2 ), nutrition (provided by Sasha Wal-
leczek 3 ) and homeschooling (from Wiki Answers).
For reasons of robustness, corresponding to each of these data sets, a topic
taxonomy with the structure presented in Section 2.1 was manually constructed.
Table 1 gives an overview on the size of the data sets and the corresponding tax-
onomies. None of these taxonomies reflect a unique and complete image of the
actual domains. They are merely a snapshot of the domains from a particular
perspective. The topic-trees were constructed in a way to cover the question
datasets. In this particular case, the structure of the hierarchy does not influ-
ence the performance of the recommender models and, therefore, represents no
variable in the overall evaluation process. However, the performance of the rec-
ommender models does depend on the topic, knowledge and cognitive process
mappings.

Table 1. Overview of question data sets

Data set No. of questions No. of topics No. of questions with questions

Earth sciences 313 49 37

Nutrition 318 38 24

Homeschooling 191 42 39

1 www.wiki.answers.com
2 www.madsci.org
3 www.walleczek.at