Uber India rape case: what can "Big Data" do?

The headline is not that unexpected; and I am sure there are ‘official’ taxi drivers who commit crimes including rape. So what is the difference?

The difference is that when a taxi cab company hires someone or rents out a cab to someone, they have a duty of care, they have to ensure that the driver has the credentials to drive, and has a clean enough record to be representing the company. The passenger can have some expectation that these checks have been put in place.

But when the transaction is one between two individuals, when the intermediary does not purport to be represented by any of the parties, then the intermediary has no responsibility in the transaction. At most they simply ban one of the individuals from using their services. And in this case, this seems to be what Uber has done; revoking the alleged rapist’s Uber account and giving his details to the police.

So the question is: whose role (if at all) is it to ensure that the service provider is indeed reliable, and how easy is it to do so?

Organisations like amazon have proposed a middle ground where service providers (sellers) are rated by past clients and a potential client can review these ratings before engaging in the transaction. Similarly, service providers can look up the records of the prospective buyers. Paypal even has an arbitration system. But these are for purely monetary transactions, not personal services.

In the case of Uber, this is much harder; they can require more details about people who register to use their services, including their presence/identities on social media. This would allow buyers to quickly check-out the service providers before accepting the deal. 

‘Big Data’ Analytics can be used to rate the users of the service, based on their performance, previous transactions, and their digital fingerprint (a bit like credit scoring). But, especially in less developed countries, not everyone has a digital fingerprint; the person picking you up in India might not have any internet presence. 

Still there is a business opportunity for anyone who would provide this rating service in countries where internet usage is pervasive. But to be rated properly, individuals would have to reveal their details, the less privacy the more accurate the rating. Hence there should be 2 scores, one the ‘risk rating’ or rather the opposite, reliability rating the other the degree of openness (availability/accuracy of data). 

While a service provider might have an incentive to do so; users might not. Thinking in terms of quadrants, the stars would be service providers in the high reliability high openness quadrant, and low reliability and low openness members would have to work extra hard to get customers and improve their scores.

‘Caveat Emptor’ is still a useful guide, but in cases such as rider-sharing, both the service provider and the user have a lot to lose; while a passenger can be robbed/raped, so can a driver. 

http://www.aljazeera.com/news/asia-pacific/2014/12/indian-woman-raped-uber-cab-driver-201412771324100241.html

http://qz.com/307870/this-is-how-your-uber-driver-is-recruited-and-trained-in-new-delhi/

Update:

http://economictimes.indiatimes.com/tech/software/delhi-government-bans-uber-says-it-is-misleading-customers/articleshow/45411180.cms

Simply banning Uber throwing the baby with the bath water

My Scottish SIngle Malt Whisky Sampler

I wanted to come up with a Scottish Single Malt sampler set, highlighting the wide range of tastes that single malts can offer. The idea is to have a quick way of understanding what type of whisky one likes, and then be able to zoom in on the personal favourite(s), without burning a hole in one’s pocket or one’s liver.

The dataset used is the well-known whisky dataset of 86 single malts rated along 12 taste dimensions: body, malty, nutty, smoky, tobacco, medicinal, spicy, honey, winey, sweetness, fruity, floral.

This is the sampler I came up with:

The 86 whiskies were split into 7 groups and a single malt has been recommended for each of these.

Compared to the ‘average’ single malt (overall), each group of whiskies displays distinct characteristics.

What do you do if the whiskies recommended above aren’t available, or are not exactly to your taste? Read on and find out.

Starting from the beginning:

Each whisky is represented by a node (dot) and the relationship between them is shown by a line; the darker/thicker the line, the stronger the relationship, the more similar the whiskies. There are clearly 2 groups above, coded red and turquoise, within the red and turquoise groups there are dark red lines, but in between them, just thin yellow lines. The red whiskies tend to be spicier/smokier and the turquoise ones sweeter.

Focusing on the spicier/smokier whiskies and only looking at the stronger relationships (filtering out weak relationships between whiskies that are not that similar):

2 groups emerge clearly, while tobacco is a common taste between them, Tullibardine, Fettercain and Craigallechie tend to be spicier, nuttier and maltier than GlenDeveron/MacDuff, Royal Brackla, Caol Ila, Pulteney, Lagavulin, Laphroaig and Clynelish who are more smoky, medicinal, as well as have more body.

Looking at the mass of whiskies on the bottom right, it is apparent that the relationships are also not that close; there are not that many strong relationships. There is room to refine the sampler further.

Focusing on the area highlighted in red:

Isle of Jura is most deeply connected to Talisker (normalized link strength of 0.99) and Oban (0.97).

Balblair is most deeply connected to Cragganmore and Oban (0.98 and 0.97 respectively). Cragganmore is most deeply connected to Balblair and Dufftown (0.98 and 0.98 respectively), Oban to Talisker and Ardberg (0.99 and 0.99).

Loch Lomond however is most closely connected to Glenmorangie (0.99), Isle of Jura (0.98) and Bruichladdich (0.98), Glenmorangie to Loch Lomond, Balblair and Inchgower (0.99, 0.98, and 0.98), Inchgower to Loch Lomond, Glenmorangie (0.98, 0.98).

Therefore, Ardberg, Talisker, Dufftown, Teaninish, Oban, Isle of Jura, Balblair and Cragganmore make up a group of smoky, spicy medicinal whiskies with body.

Loch Lomond, Glenmorangie, Inchgower, Springbank, Dalmore, Bowmore, Isle of Arran, Scapa, Bruichladdich, Bunnahabain, Highland Park, Glen Scotia make up the group of light balanced medicinal, honey whiskies.

Looking next at the turquoise, sweeter whiskies group:

The sweeter whisky group has 3 clear segments. Glenfiddich is on its own as a sweet, floral, fruity light whisky.

AnCnoc, Aultmore, Speyburn, Dalwhinnie, Glenallachie, Glenlossie, Glenkinchie, Miltonduff, Ben Nevis, Tobermory, Cardhu, BenRiach, Auchentoshan, Tamnavulin and Strathmill are honey, fruity, floral and spicy.

Contrary to the spicier whisky group, the large group of whiskies is much more deeply linked (a mass of dark connecting lines) and hence is not segmented further.

Hence the whiskies are grouped as such:

The sampler simply picks one representative from each segment. 

If the recommended whisky on the sampler is not available, just replace it with another member of the segment. For example, if Fettercain isn’t available, try Tullibardine or Craigallechie.

If you find that you’d like something ‘in between’ two whiskies, simply explore the range of whiskies that are linked to both groups. For example, if one is unable to choose between Talisker and Highland Park, then Loch Lomond or Glenmorangie who are connected to both segments would be good choices. Similarly if you’d like something ‘in between’ Macallan and Auchtentoshan, you can try BenRinnes, Balvenie or Dalwhinnie.

Happy discovery!