Tag cloud

Blogumulus by Roy Tanck and Amanda Fazani

Saturday, October 22, 2011

Quarter 10 - Week 7

You'd expect that as the finish line draws nearer, you'd get re-energized to run faster. I do have another quarter left (oh, the choices one makes...), and I'm looking forward to a couple of really cool courses (New Enterprise Financing and Reinvention through Entrepreneurial and Intrapreneurial Leadership) *fingers crossed for no schedule clash*, but still it's almost a feeling of relief, and the 'slow down, whats the hurry' syndrome. I want to think that I'm going to miss college once I get out, but I'd be lying. I can't wait to be done with the busy Fridays and Saturdays, the sleepless and least-efficient weeknights. Yes, I know I'll miss the refreshing gyaan (in most cases) that's being poured on us, which we then go and spout to our colleagues as if we came up with it... but hey, three years is a LONG time in an MBA programme. Let's get it done.

Business Data Mining & Decision Models
This week continues with more lab sessions... so we do Artificial Neural Networks and Market Basket Analysis. The funda of the first is that the model slowly learns how to set weights to the various inputs such that the output can be predicted. It still uses the concept of training data (the set that is used to prepare the model) and the testing data (the set on which you try to confirm if the model does indeed work). We got to see that sometimes models need to be 'tricked'. You'd oversample one particular segment of input records, to balance their weightage towards getting the final answer.

Why? Let's take an example. Let's say you want to understand attributes of the people that are likely to use an online dating service in India. So ASSUME you run an analysis on your users (OMG, you have no privacy concerns, you have no ethics etc. etc...) and ASSUME you find out that a majority of the participants like to watch TV shows about war, wrestling and cricket. ASSUME you also find that the majority would still browse through the profiles of the opposite gender, EVEN if they were already hooked up. Now you're wondering if something's off, even the so called sensible women, epitomes of loyal partners in relationships *cough*, appear to be going nuts. What you don't get immediately is this. Anyone who uses chatrooms in India knows that the ratio of men to women in such sites is *very* high. It would have been much higher, had we not had some nutty Indian guys parading as women online, for whatever psycho reasons they have. So your results are actually dominated by male attributes. Therefore, you try to oversample the women in the group, to get a more fair representation amongst the genders. Then, the results you get are more likely to be attributes of the psychographics that visit online dating sites. Get it? No? Remind me not to join the faculty at the IIMB.

The next session was about identifying what options in a set go together. For e.g. when you visit a supermarket, is the purchase of bread and butter highly correlated, or is the purchase of milk and vermicelli highly correlated (yum, payasam!). Therefore, the Market Basket Analysis tries to see if you can identify any trends in co-purchases/activity in a particular field. Fun analysis, lots of weird looking graph, but some interesting thoughts from our lab guide on what could be the possible reasons for the correlations we seemed to find. For e.g. milk and cheese. A possible reason could be that they're both being refrigerated, hence stored together...). Incredibly fun stuff.

Strategic Thinking and Decision Making
We go even further down the emotions route, we add the concept of affect this week. There were a lot of interesting principles.. stuff like how group decision making appears to be as flawed as deciding by yourself, since people tend to go with consensus more than rational debate. Or how you're likely to think more about how a particular decision benefits you rather than the company. Or why people who're advertising about 'Feed a child' always show you an EXTREMELY impoverished kid, to bias a rational decision making process. It's not a bad thing, it's just that people appeal to your emotions, instead of your logic, in some scenarios.

Or even the fact how you don't understand the scale of certain problems. For e.g. if there are two jars, a really big one with a lower percentage of green to red gems, and the other is a much smaller jar with a higher percentage of red to green gems, you're more likely to pick from the big one when asked to get more than 10 green gems in your hand (because you seem to be overawed by the magnitude). Or even when you're gambling, you're given a chance to win really big with lower probability, or win very small amount with a much higher probability, and you pick the latter because you enjoy the concept of winning more than the value of what you're winning. So the point is that the way elements AFFECT your decision making process is not visible to you, but it should hopefully be visible to others. So try to get groups to think together, and question each other, rather than letting one person dominate the discussion and then sway the group in his/her favour.

Next week's a break for Diwali, so it should be quite chilled out. I so look forward to being waken up by noisy crackers at 5 in the morning. My alarm clock chime just doesn't convince me that its worth waking up early anymore, maybe this will change that perception...

No comments:

Post a Comment