Many data mining applications deal with the very nature of data (including meta-data) about objects. This is, of course, true also for educational data mining. If we browse through EDM’09 proceedings, we can easily find that most of the articles are indeed of this type, where the main subjects are, in most cases, students. Here are just a few examples for the data researched:
- Student’s skill knowledge
- Student’s choice to go off-task
- Symptoms of low performance [of students]
- Students’ drop out
- Students’ knowledge and learning
- Students’ pace
- Students’ consistency
- Students’ mental models
In all of these examples, not only the discussed measure was of students’, but the data itself was constructed upon data of students. Of course, it is not possible to predict students’ behavior without researching students’ behavior, but are these measures enough? This question is very interesting as some learning configurations and learning environments enable us to learn more about students not only from their own behavior, but also from the social network underlying their collaboration or mutual actions. Two different works of that type are worth mentioning.
MIT Gaydar. Although not being published (yet?) in a scientific journal, a students’ project in MIT has suggested a very interesting, not to say revolutionary, result: Information about Facebook users and their friends in this social network might reveal sexual orientation, and in particular might point out gay men, even when they have not indicated this fact in their profile. It might be criticized that many methodological details should be improved, however the idea and its implementation are definitely intriguing (and, of course, raise a lot of ethical questions).
Discovering missing links in Wikipedia. Wikipedia, and wiki-based applications in general, have been a fertile ground for dozens and hundreds of studies from all kinds of point of views, including from the educational angle. This particular research (Adafre & Rijke, 2005) uses similarity between Wikipedia pages – i.e., finding clusters of pages by their content – for discovering missing links in a certain page (according to the links in its cluster members). Putting it in other words: Data about a page’s “close friends” reveal some important hidden information about the page itself.
Attempts have already been done in the direction of understanding students’ collaboration – e.g., in (Talavera & Gaudioso, 2004; Kay, Maisonneuve, Yacef & Zaiane, 2006) – however, it seems that mining social networks is somehow different. If we borrow similar ideas to those presented in the Gaydar, Wikipedia examples above, we might think of a few research directions using data mining methodologies for studying social networks in the learning/teaching context:
- Predicting students’ success/failure by analyzing their online collaborators’ grades in a wiki-based learning environment;
- Developing a homework recommendation system based on what your Twitter-followers twittered;
- Updating a student model according to the student’s friends’ models.
These are, of course, only a few provocative(?) imaginary examples. It is clear that as we enrich our sources of information, research and its applications would only benefit. However, we should consider that not only direct data about the students may reveal important information about them, but also that indirect data may lead to some very direct conclusions. As the old saying states: “Tell me who your friends are, and I’ll tell you who you are.”