This post stems from today’s lecture on the Market for Lemons and my initial research for my paper. I’m interested in the effectiveness of modern recommendation engines, especially in the digital music market. Our discussion today came to the conclusion in that a simple market where buyers and sellers have imperfect or asymmetric information the expectations of the buyers can become self-fulfilling prophecies. We’ve also seen how cascades can develop that will perpetually attract buyers to a product that is not optimal if the buyers do not interpret their signals correctly.

It seems you could tie these two concepts together to explain why a poor recommendation engine fails, or, to look at it from a slightly more optimistic perspective, to determine the characteristics that are necessary to ensure that a recommendation engine is effective. We have discussed how recommendation engines are a form of digital information cascade. Users are drawn towards highly recommended songs which in turn offers many more opportunities for more positive recommendations; likewise, a song with only a small amount of recommendations might never get the number of clicks necessary to make it very popular, even if it is intrinsically very good. If a recommendation fails to promote naturally good songs (or does so at too slow a rate), then, given today’s lecture, it seems that the site or store would be doomed.

I can see how users would quickly become disillusioned with the recommendations and come to the conclusion that the recommendations are worthless. In this case the user’s value of the site’s recommendation engine drops, and depending on the user’s reliance on recommendations, the user might deem the cost of navigating the site to be more than its value, and either switch to a site where the recommendations are more trustworthy, or only stick to the songs that they know to be good. This would then lead to even worse recommendations since now users aren’t bothering to use the recommendation engine at all or not even using the site.

It might be a bit of a stretch to relate these two ideas, but it came to mind during today’s discussion. Recommendation engines are a huge factor in the success of an online business. Chris Anderson, author of The Long Tail, talks extensively (and often repetitively) about the need for accurate recommendation engines, and this article reinforces his sentiments. Millions of dollars are spent to make recommendation engines as useful as possible, and I think that we can tie together some of the concepts of this class to see what really goes on as a network of users determines which engines sink or swim.