Ask HN: Why do people not like news personalization?
Like many I am a bit of an information addict with my main sources being rss, twitter & hackernews/reddit. It takes a lot of work to extract the decent information from all of these sources (probably about 10-15% of the information I receive is any good). The obvious solution to this problem seems to be applying a learning algorithm to create a personalized feed of news items. However I know there have been a few sites which have done this in the past with little success. Can anyone suggest any reasons why they have failed? Is it the implementations or that news personalization just doesn't work? Personally, I mostly dislike news personalization. For instance, I always log out of reddit unless I'm actually going to post something or add a comment, because of the personalization issue. My selection of techy-related aggregators like reddit and HN are enough personalization for me. The problem for me with personalization is that it's annoying to always see the same set of things. For instance, I'm mostly a Python programmer. But I've found that if, for instance, I'm subscribed to the Python reddit, I get too much Python-related content in my news feed. That's not really my focus -- I'm really a programmer first, and a Python programmer second. I'd rather see interesting programming things -- regardless of the language or focus -- than I would in seeing Python-related content. As anohter example, I've found the same is true with political stuff, which I've mostly tried to handle using RSS feeds. I can subscribe to political content that I'm interested in, but then I find that I just end up reading the same things and viewpoints over and over again. I'd rather read political content that's interesting and well written -- regardless of its political stance -- than I would read political content related to some sort of viewpoint or interest. That's the value of a "logged out" reddit/hacker news, or the front page of a newspaper; I'll see the most important things first, and then do my own seletion of whether or not I want to read the stuff contained therein. One interesting sidenote is that reddit basically dropped the personalization filter thing a long time ago. You may recall that they used to have a "recommended" articles tab, which was supposed to learn about what you were interested in based on your voting history, and then by clicking on the tab you'd see the top articles that it thought were relevant to you. AFAICT this feature no longer exists, and now they just have the sub-reddits system where you get a simple mix of content from subreddits you're subscribed to. I'm not sure exactly why they dropped that (it never worked very well), but that might be something to ponder a bit. It seems to me that what is really going on here is that the personalization is lousy. The point of personalization, one hopes, is to show you what you want to see. But you say the personalized version does not do that. That being the case, the personalization has completely failed. Which might answer the OP's question about why people don't like personalized feeds. On the other hand, it could be that you are upvoting not things you want to see, but things you agree with. Tragically, many people do. In that case, it's your own fault (or the fault of anyone who does such upvoting, and dislikes the results); the personalization algorithm is doing it's best, but it's getting bad data. And since upvoting based on agreement is so common, this also might answer the OP's question. Some guesses: One reason might be that the existing range of news portals is good enough for most people to get approximately personalized content. People can find one or more news outlets, aggregators, or blogs that more or less reflect their interests, whether it's HN, Reddit/Subreddits, Slashdot, CNN, Fox News, DailyKos, Lambda the Ultimate, or whatever, and they just read those, supplemented with links passed on from their friends via Twitter/Facebook/etc. So your market is the people who can't find any combination of those that works for them, which maybe isn't a ton. A different reason might be that Google News does personalization (via the "Recommended" box), and so already captures a decent part of the market for news personalization, at least in the newspaper-articles sense. A third might be that it's hard to do well. To really catch on, people need to rarely get articles they don't care about, and often have this feeling of, "yes! this is exactly the kind of news I want to read, and which I wouldn't have found otherwise". That's probably hard to do! The Google News recommendations don't really blow me away, for example, even though I've been using it long enough that it should have decent data by now. I liken this problem to browsing a library or bookstore vs searching its card catalog or asking for specific information. If I have a problem, I know what I want and it may not be something I've wanted before. If I want to relax and browse, I often want something different than I've wanted before. So I think it is a hard problem, as said here, serendipity is hard to program. However, I am absolutely sure we have not yet figured out the best way to browse. Search seems pretty good, but internet browsing can easily be too time consuming and aimless. No shared context. You cannot say "Did you read the article in the NY Times about XXXX" and expect that your friend, who you know reads the NY Time (online of course) may have done so. It's not possible to can serendipity. No algorithm can make a leap like a human mind, connecting two things that a machine would see as separate. This is why I've never used any personalization. I don't want to see only what a machine "thinks" I should see. True, the best articles are always the ones which take you in a completely new direction. However I have also found that those same interesting articles are often not very popular which leads me to think social news sites have their flaws. Perhaps the solution needs to not look at what is popular/similar to what you already like but rather new concepts (not that I have any idea how to achieve that)