• Home
  • Articles
  • Why context suggestions and personalized search don't work

Why context suggestions and personalized search don't work Why context suggestions and personalized search don't work

We're all pretty much used to serious media services, file storages, social networks and media libraries offering context suggestions when we use them. Either by email or in a personal account menu, gently offered as an advice or advertised on the same page — suggested alternatives are everywhere now. 'Users who purchase this might also like X' or 'The books you may also like Y' or 'Try listening to this 1989 music album of a singer from the same decade as the song you opened'. These are standard recommendations based on the information incorporated into the database of the platform you use. This probably started with Google's personalized search, as the idea of studying what the user wants and needs makes it easier to provide those very things.

But let's face it, those suggestions look silly and unnecessary for the most part. That is the bane of any automated algorithm — lack of actual personality. Those recommendations are based on generalized data and equations created to be as broad as possible. Basically, services such as Google, Spotify or Netflix look at your personal data and study the history of your activity. Then they combine it all (some better than others) and try to make you stay a little longer by offering something else beside the thing you actually needed. This might prove useful, but underwater rocks crush all hopes for that.

The very first issue to appear with context suggestion is that they are very repetitive. Most of the Spotify or Pinterest suggestions I check every morning seem to be the same as yesterday and the day before. That's because my browsing history and interests on a particular site are limited, and the suggested entries are limited proportionally. That was the issue I had with Google search: it limits your access to information for no good reason. Luckily, media services simply suggest something, and you can actually search for anything. But when you see a Spotify suggestion to listen to Van Halen for the tenth time, it gets frustrating.


Second, automated context suggestions do not really take into account your personal taste and preferences. Of course, the more advanced mechanisms such as Google's, which accounts for gender, race and age, can provide a good guess aboutwhat a person would like. But most of the time suggestions based on group tastes and target audience appeal are misses rather than hits. Not everyone who's 15 to 21 will like young adult novels. And by god, not every middle-aged woman likes '50 Shades of Gray'. Personal preferences based on things that cannot be included into a database, like the way a particular band sounds or the style in which a specific author writes, are what really defines our interest, and the best contextual recommendations can do is try to provide as many possible alternatives as they can find. However, abundance and generalization actually balances out the repetitiveness mentioned above. At least you can try to find something new and fresh to like.

The third problem I have is the way those services and sites gather information on me. They are simply too thorough. Including the data about a search I did recently does not really help understand what I like. If I needed a download manager once, it does not mean I would need similar programs any time soon, yet some of the sites keep offering software I'm not interested in because I searched for it once. Some have it worse than others, online book stores probably being the worst in this regard.


The fourth problem is of a little different nature. So, what if I want to use Spotify or Pinterest, of even Google most of the time, as a working space where I can do research? This immediately messes up any contextual help I might have in the long run. Because research means variety, and variety destroys specialization, which is the core of any contextual recommendation. As a research tool, such services are limited by personal context, and as something personal, they are tooquick to grab onto any bit of browsing history you have, thus, harming both aspects of their use.

All of this looks rather sad. Contextual suggestions and personalized search seem to be unable to satisfy any real demand. But that's not true. Of course, Netflix has crude algorithms that can offer you a terrible movie, Facebook tries to make you add someone you don't like or don't know as a friend just because you were in the same school, and Spotify thinks that one reggae song is reason enough to mark you as a fan of the genre. But despite all the flaws, those suggestions might provide something good. Finding long-lost childhood friends based on you personal info in a social network, discovering new bands you might like or simply finding a new interest while stumbling upon a piece of research material. All of this works, because there was a reason to add suggestions in the first place. But they surely and definitely need improvement.


So, how do we improve personalized search and suggestions? By enabling the option to turn the contextual element on and off. Google has been asked numerous times to make personalized search optional. Optional disabling of contextual suggestions gets rid of half of the problems at once. Broad research is no longer confusing, and your personal recommendations will not be affected later if you block the data collection on current browsing.

Another thing to do would be allowing users to personalize the algorithms, adding specific filters in addition to the data collected. For example, the option of listing the categories of software you don't like or have no interest in or stating the genres and artists you don't want to see in suggestions can be added. Basically, personalization of your personalization can make the system itself a lot more detailed and useful.

Finally, in order to get rid of the limitations brought by personalized context suggestions, adding a broader level of suggestions can help. For example, crime novel readers can be offered a genre they might also like and not just specific books, action movie fans might be interested in disaster movies, etc. Offering an idea instead of a product is probably a perspective way to keep a visitor interested.

All this aside, the main reason why context suggestions and personalized search don't really work is because they are still young. All the tweaks and upgrades will surely happen in the future. So, for now, we just have to suffer the imperfections of crude algorithms and smile on sites offering us to buy an axe and a hockey mask along with a DVD box set of Disney cartoons.