Advertisers often assume that targeting and personalization are useful - not only for themselves - but also for the buyer. When people browse through a magazine or catalog they see much that is not necessarily relevant; the reason why is obvious: the magazine aims its ads to a more or less broad audience, which means that any particular person is likely to see ads in which he has little interest. Online advertising, in contrast, is not subject to the same constraint: advertisers are able to target a small section of the population, and even personalize ads to a particular user. A person buying books about a particular topic will see recommendations for books on similar topics, and will see ads for related products. So, instead of getting many irrelevant ads, the person only sees ads that are closely related to his own shopping habits.
There is a growing interest in personalizing online experience, an indication of which is the following mission statement from The Journal of Personalization Research, a 10 year-old online publication dedicated to the study of personalization. The journal aims at disseminating "novel original research on interactive computer systems that can be adapted or adapt themselves to their current users." According to a recent New York Times column, the reason why personalized marketing is "being hailed as the latest breakthrough" is that it "tries to show consumers the right ad at the right time" (Helft and Vega 2010). From the standpoint of the advertiser, all this is indeed wonderful. Yet, it is worth asking whether, in addition to the potential benefits of such effort, there are reasons for concern. In this article, I discuss one such reason.
One general consequence of these practices is that they reduce the information-content to which the buyer is exposed. He sees only things which his browsing habits justify, and hence things that are likely to be similar to the ones he (or people with similar habits) saw in the past. In other words, there is less variation in the information content the buyer is exposed to, he ends up getting the same information again and again. He is exposed to a narrow world, the world relevant only to him, as it were. The aforementioned New York Times column by Helft and Vega illustrates the point well. It begins as follows:
The shoes that Julie Matlin recently saw on Zappos.com were kind of cute, or so she thought. But Ms. Matlin wasn't ready to buy and left the site. Then the shoes started to follow her everywhere she went online. An ad for those very shoes showed up on the blog TechCrunch. It popped up again on several other blogs and on Twitpic. It was as if Zappos had unleashed a persistent salesman who wouldn't take no for an answer. "For days or weeks, every site I went to seemed to be showing me ads for those shoes," said Ms. Matlin, a mother of two from Montreal. "It is a pretty clever marketing tool. But it's a little creepy, especially if you don't know what's going on." People have grown accustomed to being tracked online and shown ads for categories of products they have shown interest in, be it tennis or bank loans.
Ms. Matlin saw the same pair of shoes wherever she went. However, if advertising is information, as Ogilvy (1985) noted, then Ms. Matlin might have benefited from the exposure to different kinds of ads (with different contents). Such exposure might have taught her things about the world and about herself of which she was unaware. It may be said, then, that one especially important unintended consequence of targeting and personalization is that by exposing the user to variations of the same, the chances of discovery by serendipity decrease considerably. The chance encounter of an author unheard of is unlikely in a world customized to fit the patterns of past behavior.
A similar problem has been noted by Sunstein (2007) in the context of political behavior, and much earlier, by communication scholars (e.g., Beniger 1987, Rucinski 1992). Sunstein (2007) noted that whenever people voiced their ideas at the town square everyone was necessarily exposed to arguments of different sides on a controversy. However, in a world in which people are exposed to ideas mainly through reading specialist blogs or newspapers online, they only read things with which they already agree. The result is, sadly, that people are not able to communicate effectively anymore. Nor do they learn to understand and respect different ideas, which in turn means that they do not refine their ideas nor discard them because they learned something new from someone having a different view. The world of ideas is becoming compartmentalized, with every compartment tightly sealed to imports from the others.
It interesting that recommendation systems have the same problem, one which has attracted the attention of some scholars. Recommendation systems have focused almost exclusively on providing accurate recommendations, to the exclusion of novelty. Nonetheless, a good recommendation system should achieve not only accuracy in its recommendations, but also novelty and serendipity (Resnick and Varian 1997, Adomavicius and Tuzhilin 2005, McNee, Riedl, and Konstan 2006). Fleder, Hosanagar, and Buja (2010) put the problem well: "Personalizing websites means that we may no longer see the same newspaper articles, television shows, or books as our peers... [R]ecommender systems will create fragmentation, causing users to have less and less in common with one another."
In what follows, I briefly discuss the role of serendipity both in science and in shopping, and in the next part of this two-part article I review some of the existing approaches to increasing the chances of discovery by serendipity (the review will focus on recommendation systems, in part because the literature on this topic has began to address the fact that these systems often recommend "more of the same" and hence lack diversity).
The first task is to show that serendipity plays an important role in discovery generally. Andel (1994) defined serendipity as "the art of making an unsought finding" and Barber and Fox (1958) noted that "by its very nature, scientific research is a voyage into the unknown by routes that are in some measure unpredictable and unplannable. Chance or luck is therefore as inevitable in scientific research as are logic and what Pasteur called 'the prepared mind.'"
The importance of serendipity in scientific discovery is well-known, and has been repeatedly noted (see, for example, Roberts 1989, Merton 1993, Merton and Barber 2004). Indeed the Wikipedia entry on Serendipity lists dozens of cases in which serendipity has been an important element in the scientific discovery of, for example, the chemical structure of cyclic compounds (Kekule), penicillin (Fleming), the neural control of blood vessels (Bernard), X-Rays (Roentgen), and so on. The discovery of the double-helix model for the structure of the DNA by James D. Watson and Francis Crick may be offered as an additional example. The recognition only dawned on the scientists after having seen, by chance, an X-ray diffraction image taken by Rosalind Franklin and Raymond Gosling (Watson 1968).
Serendipity also plays an important role in the context of shopping, though this role is not as well-understood (Rowley 2002). Leafing through the pages of magazine may lead the reader to see an ad about a long-forgotten item, the memory of which the ad may help revive; and browsing the shelves in a bookstore may lead to the chance finding of an interesting book. A brief personal story illustrates the point. In a book sale, late in 2007, after having looked at the books in the Computer Science section, I caught myself looking at the books in the Education section, which happened to be near the Computer Science one. Even though I rarely look at the Education section, I did look at it then, which led to the discovery of Teacher in America a book by the cultural historian Jacques Barzun. I bought and read the book in the following weeks, and was very much impressed by the ideas expressed in it. Since that serendipitous encounter I became an avid reader (and rereader) of Barzun's books.
Now, if the things we are exposed to are exclusively determined by our past interests - which is what recommendation systems in general do - we are constrained to buying the same or similar things. Discovery by serendipity is severely limited. We miss the things that might interest us. Books in a bookstore are broadly categorized by topics, but in any one category the books are organized alphabetically, which, in regard to the contents of the books, is the same as random. Two books placed side by side in a shelf may have very little in common, apart from belonging to the same broad category and from having been written by authors with names beginning with the same letter. Buying books online is a completely different experience, and chance encounters are less likely to happen.
In terms that are closer to the main issue this article is addressing, whenever users are recommended an item (book, movie, etc.) by some automated system or, alternatively, shown an ad in the search results page of a search engine, they would be best served if they were shown not only things which are an obvious extrapolation from their past habits but also things that are new or surprising. By providing more of the later the chances of discovery by serendipity increase. I turn now to a review of some of the approaches that have been advanced to increase these chances.