Posts Tagged ‘eCommerce’

Facebook has announced a pretty nice graduate student fellowship program.  The first three reserach areas they mention are:

  • Internet Economics: auction theory and algorithmic game theory relevant to online advertising auctions.
  • Cloud Computing: storage, databases, and optimization for computing in a massively distributed environment.
  • Social Computing: models, algorithms and systems around social networks, social media, social search and collaborative environments.

(Hat tip to TechCrunch.)

Google has its own graduate Fellowship program (Nicolas Lambert got the 2009 Fellowship in “Market Algorithms”.)


Read Full Post »

Google gives its employees $1/day of free adwords advertising.  Beyond an employee perk, this gives Google’s employees the experience of being an Internet advertiser, i.e. experiencing the point of view of Google’s paying customers.  I have been using my $1/day account to advertise the divorce-consulting business of my sister in law (In Hebrew) and did indeed find this experience to be quite illuminating.

The first thing I learned is that the ad auction itself is just a small part of the whole thing.  Choosing the right text for the ads (all ten words of it), choosing the right keywords to target, etc, is much more prominent than setting the right bids.  “Tiny” issues come up everywhere, e.g. when I needed to choose keywords to target, it turns out that there are four ways to write divorce in Hebrew: גירושים, גירושין, גרושין, גרושים.  I’m not really sure whether these are all “kosher” spellings, but they are all searched for an do need to be taken into account.  Even more, the whole advertising campaign is peripheral, in principle, to the business itself, and frankly the auction logic is not the first concern of the advertiser.  This is obvious, of course, but is easy to forget for one whose work focuses on the auction logic.

Now for the auction itself: all together I spent several hours setting up the campaign, making up the ads, choosing keywords, looking at reports, and trying a bit of optimization.  The adwords user interface was very easy and convenient to start with, but it didn’t take long until I was attempting things that confused me (e.g. splitting my single campaign into two different ones), at which point I gave up, and stayed with what I achieved, which is quite fine actually.  I was especially impressed with Google’s automatic suggestions of keywords which were cleverer than what I came up with (I know that not really, just some learning algorithm, but they were eerily good.)

I was surprised and disappointed (as an advertiser, but frankly delighted as a Google employee) by the pretty high prices on the keywords that I targeted: my average cost per click is 88 cents, and this is for pretty low slots, on the average.  (Divorce is expensive, it seems, also on the Internet.)  This means that my $1 per day suffices for a single click per day, and no more.  I do get this single click almost every day, but have so far been unable to ever get two clicks in one day: neither optimizing by hand, nor letting Google’s automatic bidder do it.  I was professionally insulted by not being able to beat the automatic bidder, but have still not given up on getting an average of more than one click per day for my $1.  My click through rate is pretty high (compared to my expectations):  0.79%, so I usually get about 150 impressions every day of which one is clicked and results in a visit to the website.  Now these visitors are probably really good leads: not only have they searched for relevant keywords, but they also clicked on a pretty specific ad.  I suppose that if even 1% of them become clients (this is not so little: we are not talking about buying a sweatshirt; this is about handling divorce), then the advertising would be considered quite  profitable even had my sister in law paid for it.  Unfortunately, it is quite hard to gauge whether this is the case: getting and keeping the required statistics is easier to imagine theoretically than to do when you have to handle a small business.  In other words, I haven’t a clue what my valuation of a click is.   (The lack of knowledge of one own’s valuation has been discussed in AGT, but frankly I have not seen really convincing treatment of this issue.)

The set of reports about the performance of the campaign that adwords makes available  is quite impressive, and they are really nicely and simply done, but somehow I still don’t really know how how to optimize my campaign as to get the most and the best customers to the site.  I’m sure that more time on my part, as well as a more data-centric handling of my sister-in-law’s business would improve things, but the difficulty of getting and handling the right data is another lesson that I got from this exercise (again, I knew this theoretically, but now I feel it too).

Read Full Post »

Crowd sourcing has just been given a recent visibility boost with DARPA’s Red Balloon contest that was won by the MIT team.  At the same time, Amazon’s well-established (since 2005!) platform for crowd sourcing, the Mechanical Turk, is gaining attention as a platform for conducting behavioral experiments, especially in behavioral economics and game-theory.

Named after the 18th century human-powered chess-playing “machine” called “the Turk”, this platform allows “requesters” to submit “Human Intelligence Tasks” (HITs) to a multitude of human “workers” who solve them for money.  A sample of recent typical tasks include tagging pictures (for 3 cents), writing (and posting on the web) a short essay with a link (for 20 cents), or correcting spellings (for 2 cents, in Japanese).  This allows brief and cheap employment of hundreds and thousands of people to work on simple Internet-doable low-level knowledge tasks.  You may demand various “qualification exams” from these workers, and design such quals of your own.  Obviously workers are in it for the money, but apparently not just that.

Recently, the Mechanical Turk is being used to conduct behavioral experiments.  Gabriele Paolacci is methodologically repeating experiments of Kahneman and Tversky and reporting on them in his blog. Panos Ipeirotis reports on his blog studies of some aspects of the Mechanical Turk as well as results of various behavioral game-theory experiments on it.  I’ve seen others report on such experiments too.  Markus Jacobsson from PARC gives general tips for conducting such human experiments using the Mechanical Turk.

Turk-based behavioral experimentation has the immense appeal of being cheap, fast, and easy to administer.  There are obviously pitfalls such as getting a good grasp on the population, but so does any experimental setup.   Such a platform may be especially appropriate for Internet-related behavioral experiments such as figuring out bidding behavior in online auctions, or how to best frame a commercial situation on a web-page.  Could this be a tool for the yet not-quite-existent experimental AGT?

Read Full Post »

Google Ad Exchange

Google just announced its ad exchange:

The DoubleClick Ad Exchange is a real-time marketplace to buy and sell display advertising space. By establishing an open marketplace where prices are set in a real-time auction, the Ad Exchange enables display ads and ad space to be allocated much more efficiently and easily across the web. It’s just like a stock exchange, which enables stocks to be traded in an open way.

Google's Ad-exchange

There are some existing competitors, but Google’s entry may be game-changing.  Lots of research problems beg themselves.

Read Full Post »

An intriguing new paper, A computational view of Market Efficiency by Jasmina Hasanhodzic, Andrew W. Lo, and Emanuele Viola has just been uploaded to the arXiv.


We propose to study market efficiency from a computational viewpoint. Borrowing from theoretical computer science, we define a market to be efficient with respect to resources S (e.g., time, memory) if no strategy using resources S can make a profit. As a first step, we consider memory-m strategies whose action at time t depends only on the m previous observations at times t-m,...,t-1. We introduce and study a simple model of market evolution, where strategies impact the market by their decision to buy or sell. We show that the effect of optimal strategies using memory m can lead to “market conditions” that were not present initially, such as (1) market bubbles and (2) the possibility for a strategy using memory m' > m to make a bigger profit than was initially possible. We suggest ours as a framework to rationalize the technological arms race of quantitative trading firms.

Read Full Post »

A “textbook system” based on social choice theory would have a centralized mechanism interacting with multiple software agents, each of them representing  a user.  The centralized mechanism would be designed to optimize some global goal (such as revenue or social welfare) and each software agent would elicit the preferences of its user and then optimize according to user preferences.

Among other irritating findings, behavioral economics also casts doubts on this pretty picture, questioning the very notion that users have preferences; that is preferences that are independent of the elicitation method.  In the world of computation, we have a common example of this “framing” difficulty: the default.  Users rarely change it, but we can’t say that they actually prefer the default to the other alternative since if we change the default then they stick with the new one.  Judicious choice of defaults can obviously be used for the purposes of the centralized mechanism (default browser = Internet explorer); but what should we do if we really just want to make the user happy?  What does this even mean?

The following gripping talk by Dan Ariely demonstrates such issues.

Read Full Post »

The current “killer application” for algorithmic game theory is certainly on-line advertising. The traditional advertising industry is  undergoing great turmoil due to this on-line advertising as well as other technological changes which may not only give viewers more relevant and less annoying ads, but also may start addressing the main problem from the point of view of advertisers, captured by the famous quote  “Half the money I spend on advertising is wasted; the trouble is I don’t know which half”.

This highly entertaining video is from the point of view of traditional advertising (thanks to Eyal Manor for the link).

Read Full Post »

From Marginal Revolution, a story on how the eBay has reduced the looting of antiquities…

Before reading question: Can you guess how this works?

Easier question: why should you expect eBay to actually increase the looting of antiquities?

Read Full Post »

Internet advertising is perhaps the number one current application (revenue-wise) for algorithmic game theory, and is paying the salaries of many AGT researchers in Microsoft, Yahoo, Google, and elsewhere.    Beyond the basic “generalized second price auction” used for ad-auctions, there are many other issues and questions involved.  

About a week ago, Google annouced a new beta program for interest-based targeting (a form of behavioral targeting).  The basic idea is very simple: allow advertisers to target ads to segments of people according to their previous Internet behavior and not just the current web-page that they are on.   E.g. if you visit many sports web-sites, the it makes sense to show you sports-related ads even on a general news page rather than, say, detergent ads.  In general, this should result in less annoying (or even actually useful) ads for the viewer, higher effectiveness for the advertiser, and higher revenue for the web-site  “publisher” (and of course, for my own current employer, Google).

Technically, this is quite easy to do using browser cookies, and is already heavily used by most advertising middlemen on the web.   The catch is of course in the privacy issues.  It seems that the main innovation here was in how Google managed to handle the privacy issues, as explained on it’s public policy blog.  The main technical ingredient here is the ads preferences manager that gives viewers full control of their data.

Read Full Post »

%d bloggers like this: