Well, as you can see, my blogging frequency is really low... But that is about to change! This has been a very educational 5 months since the last post (more about that somewhere else).
Today's topic has to do with a concept from business which is relevant to educational data mining (EDM):
Churn and
Customer Relations Management.
The article pulled from today is:
Lejeune, M. A. P. M. (2001). Measuring the impact of data mining on churn management. Internet Research, 11(5), 375–387.
Churn
Lejeune does a nice job laying out the basics of the churn issue and why companies care about managing attrition. Churn is defined as "Churn or customer attrition is defined as `the annual turnover of the market base' (Strouse, 1999)" (p.377). The electronic commerce is assumed to be the driving reason churn has become so great in the last few years. Lejeune mentions that having competition only "one click away" requires companies to have a multi-faceted marketing and management strategy to acquire and retain customers.
In the education world, there are several parallel conditions of churn: student attrition in higher education, class churn at the start of a semester, and use of educational tools, to name a few. Alternative schools, classes, and tools are everywhere (even in K12) to students and teachers. However, the view of churn in education may not be entirely negative, either. For instance, an educator interested in the growth of their student would certainly be happy when their student is ready for a greater challenge than can be currently provided.
In the world of the Instructional Architect, churn happens more than we would like, but again, that isn't a bad thing. Preliminary data has shown that about 10% of our registrants actually return within six months of exposure. We hope that this represents "early adopters" and that more will come soon. However, as the IA is built for research and supporting teachers, if they find something different that works for them, then we are happy with that. The problem has been knowing where they go and why—a nice research topic for someone if they are interested…
Customer Relations Management (CRM)
Part of the reason for data mining the IA is to figure out what user segments exist so we can better address their needs. Business has shown that the cost of acquiring a customer is about equal to the cost of a winback (unless I missed something in Lejeune's paper). So, maintaining a good relationship with current customers becomes paramount. This relationship is generally best handled on a 1:1 basis—the holy grail of marketing.
A perfect marketing and retention strategy has become easier through data mining all the many bits of information that users provide—many times without realizing they are helping (a nod to privacy concerns). With the ability to gather customer (a.k.a. user) information and almost immediately apply that to their current user experience, CRM has come a long way.
Another reference for CRM and Data Mining is:
Cooley, R. (2003). Mining customer relationship management (CRM) data. In N. Ye (Ed.), The handbook of data mining (pp. 597–616). Mahwah, NJ: Lawrence Erlbaum Associates.
DM vs. Statistics
Lejeune makes an argument that DM &ne Statistical analysis, and supports the idea that data mining has something more to offer than the standard statistics and reporting. I would have to agree wholeheartedly. His two reasons stem from the need for timely analysis and not historical facts; and that statistics traditionally find the more obvious variations, which may not be the most meaningful.
In education the
t-test, ANOVA, and regression have held sway for many years. However, each of those tests begins to be insufficiently powerful when dealing with the amount and kind of data available in DM.
While most would differentiate two bodies of DM by the machine learning concept of supervision (even myself), Lejeune offers the following DM objectives:
- Descriptive
- increasing understanding of data and their content (usually requiring unsupervised methods).
- Predictive or Prescriptive
- for forecasting and devising, at orienting the decision process (usually requires supervised methods).
Privacy
This is a great concern in today's educational environment, and no less for the IA. As Lejeune recommends, we have already included the use of personal information in our privacy policy.
Data sources
The three typical sources for e-business DM are: (a) clickstreams, (b) cookies, and (c) customer registrations. The same is true for EDM. We are also seeking to tie these with surveys and interviews in the near future—perhaps we can find out where our less frequent visitors are going, if anywhere…
Wrap-up
Lejeune goes on to say,
"Descriptive data mining methods appear useful to understand differences and particularities of the various categories of clients. It allows customer segmentation, formation of homogeneous clusters or categories of customers characterized by a small variance within groups, and a high variance between groups. Based on these clusters, data mining methods have th ability to detect common features of the individuals belonging to the same cluster" (p. 382).
Currently in the IA we know we have users, but we don't know much of how they are using the tool—especially after the workshops. So, this characterization is very important to us at this time. Later, there will be many more questions to be answered, but we need a stake in the ground of our user groups, today.
He then describes some customer taxonomies which have been developed from data mining exercises. Then he describes some sensitivity measures to see how responsive DM can be in churn management.
What I want to pull from this is that DM is helpful in education, just as it is in business. We may have different domains and considerations when it comes to use, but DM is still very applicable.