随着年纪渐长，工作越来越忙，我们越来越难主动发现生活中美好的事物，随之也涌现出了很多自称了解每个用户的需求，能够帮你推荐喜欢的音乐、餐厅或杂志文章的应用软件。

    最近我和我大学的好朋友去了一趟华盛顿特区，这位朋友现在是一名厨师。说起来有点不好意思，这还是我第一次在八年级以后去华盛顿。我对这个城市一无所知，因此对我的帮助越多越好。我把笔记本电脑放在了家里，整整两天时间完全依赖移动设备，也就是我的iPhone和iPad（我们还第一次尝试了Airbnb）。

    在选餐厅的问题上，我依靠的是Ness。今年年方27岁的科里•里斯于2009年与人共同创办了Ness计算公司。这款应用有一个“相似度分数”，可以表示出你有多大的可能会喜欢某个推荐。里斯表示，Ness最终可能会成为一个个性化的搜索引擎，但是现在这个应用主要还是针对餐厅和咖啡厅。他不无自豪地说，用户们总是告诉他：“我觉得Ness很懂我。”新闻阅读器Zite的CEO、34岁的马克•约翰逊也说，Zite的用户们都表示：“Zite很懂我。”科技界中有不少精英人才都在搞推荐引擎，这一点也不值得奇怪。里斯说：“我认为，直接输入‘我应该和朋友在哪吃饭’或‘附近有什么很酷的商店’，这个概念已经开始在移动设备上成为现实了，就算是在户外也可以实现。”

    它的工作原理是什么呢？当你第一次打开Ness，它要让你按照五个档次，给当前位置附近的10家餐厅打分。我去华盛顿前，在纽约的曼哈顿完成了打分的过程，不过我发现这个过程是有缺陷的，因为它没有拉开菜系的档次。比如它把米其林三星餐厅老板丹尼尔•布鲁德的DBGB高档餐厅和汉堡王（Burger King）放在同一个屏幕里让人打分，同时这些餐厅里还包括了星巴克（Starbucks）。同时，在你给酒吧打分的时候，它列出的有些酒吧里也提供食物。比如说我喜欢一家叫Brother Jimmy’s的酒吧，是因为我喜欢它有往啤酒杯里扔乒乓球的游戏。他们的鸡翅还有可以，不过如果我给打它了四星，Ness会不会开始经常向我推荐其它弥漫着兄弟会作风的酒吧？不过自从我到了华盛顿之后，Ness的表现要好了一些。根据我在纽约打的分，它向我推荐了一些地中海风情的餐厅，一些中东风味，以及几家我的朋友慕名已久的高端美式餐厅。除了按照你可能喜欢的程度排名之外，Ness还按就餐价格列出了一张排名，好让你知道该进哪一家。同时它也会告诉你，某家餐厅是不是城里第一、第二、第三火爆的这种类型的餐厅。（比如它推荐的José Andrés' Zaytinya就是华盛顿最火爆的地中海风味餐厅。）我们最后选择了一家名叫Central Michel Richard的餐厅（Ness称我们喜欢它的可能性有82%），我们果然美美地吃了一顿。

    吃完午饭后，我们在通过Airbnb租来的公寓房间里连上了Wi-Fi，然后我花了一点时间在Zite上看杂志，Zite是一款像Flipboard一样的所谓“智能杂志”应用。虽然Flipboard在二者间的名气更大，但Zite似乎能更好地了解用户的阅读习惯，哪怕你不把它绑定你的社交媒体也是一样。我已经用Zite几个星期了，而且我发现，我“顶”或“踩”的报道越多，它向我的个人页面推荐的文章就越符合我的品味。

    As you grow older and busier, it becomes more difficult to make spontaneous discoveries. Or at least that's the theory behind a bevy of so-called predictive apps purporting to know each user well enough to hand them their next favorite song, restaurant, or magazine article.

    I gave these tools a test run on a recent trip to D.C. with my best friend from college, who is now a chef. Embarrassingly, it was my first visit to D.C. since the eighth grade; I knew nothing about the city and needed all the help I could get. I left the laptop at home and went strictly mobile for two days, bringing only my iPhone and iPad. (We also tried Airbnb for the first time.)

    For restaurant ideas, I turned to Ness. Corey Reese, 27, co-founded Ness Computing in 2009. The app produces a "likeness score," a percentage that denotes how likely you are to like a particular recommendation. Reese says that Ness could eventually become a personalized search engine, but for now the venture is focusing on restaurants and cafes. He brags that users keep telling him, "It feels like Ness knows me." Mark Johnson, the 34-year-old CEO of newsreaderZite, also says that his app's users rave: "My Zite knows me." It should come as no surprise that more than a few smart people in tech are working on recommendation engines. "We think your entry point for 'Where should I eat with my friends' or 'What's the cool store nearby' is happening on mobile now," says Reese. "It's happening when you're already out and about."

    How does it work? When you first open Ness it asks you to rate, on a five-star scale, 10 restaurants near your current location. I did this in Manhattan before heading to D.C. and found the process flawed. Because it doesn't distinguish between levels of cuisine, it will ask you to rateDaniel Boulud's pricey DBGB in the same screen as it asks you to rate Burger King (BKW). It also includes Starbucks (SBUX). Similarly, it asks you to rate bars that happen to serve food. Sure, I like Brother Jimmy's -- for playing beer pong. Their wings are okay, but if I give it four stars, will Ness start offering me frat bars regularly?

    Once in D.C., Ness fared better. Based on my NYC ratings, it offered us a Mediterranean place, some Middle Eastern fare, and a few upscale American restaurants my friend already knew about. Ness includes, along with its percentage prediction, a price rating so you know what you're getting into. It'll also tell you if a place is the first, second, or third most popular restaurant of its type in the city. (José Andrés' Zaytinya, which it offered, was the most popular Mediterranean in Washington.) We chose Central Michel Richard (Ness promised an 82%) and enjoyed our meal.

    After lunch, connected to Wi-Fi in the apartment we had rented on Airbnb, I spent some time with Zite, a so-called "intelligent magazine" a la Flipboard. Though Flipboard has been the buzzier of the two, Zite seems to learn its user's reading habits better than Flipboard, even if you choose not to connect it to your social media. I had been using Zite for a few weeks and, indeed, found that the more stories and articles to which I gave thumbs up or down, the better it was getting with the stories it displayed on my personalized front page.

    Zite允许用户添加版面，所以会给你一种报纸的感觉。虽然我的品味比较笼统（比如运动、科技产品、文学、小说、营销等），但是Zite的版面也可以细分到让人难以想象的地步（包括针织、调酒、量子物理、意识流等统统都是选项）。但它也不是完美无缺的。比如它有一次推荐了一则来自俄勒岗州麦克明维尔市的地方报纸News-Register的报道，讲的是关于NCAA两支橄榄球队的故事，而我对它们一点兴趣也没有。另外在我添加的波士顿版面里，很多文章除了稍带提了一下“波士顿”几个字，完全和波士顿没有任何关系。不过它的瑕疵还是很少的，而且我用得越多，错误就越少。那天我差不多对首页推荐的四五篇文章都很感兴趣，我差不多花了一个小时的时间，在我的主版面上翻了19页，无论长文章还是短文章都看了，的确不错。

    Zite背后的主要“极客”约翰逊毕业于斯坦福大学（Stanford）哲学系，曾在SAP工作三年，然后跳槽到了微软(Microsoft's)，从事Bing搜索引擎的开发。另外，他也为Powerset和SideStep等搜索引擎创业公司工作过。后来他的脑中蹦出了完善一个推荐引擎的点子，这个创意让他欣喜若狂，而推荐引擎原本是Bing的目标。推荐引擎这种东西不可能每次都是完美的，而且你也不会希望它每次都是完美的。约翰逊表示，从开发者的角度看，“任何这类东西的问题都在于，意外发现某种东西的兴奋感是很容易创造的，我们喜欢把它归于偶然，但你真正想要的，是某种有指引性的发现，而不是完全的偶然。”我用Zite用得越多，偶然性就越小，当然这就达到了目的。但是随着偶然性的消失，发现某种全新事物的喜悦感也在消失。

    下一项是音乐。我当时还不想付费体验最近广受好评的Spotify，但是这款应用可以让你免费在平板电脑上使用它的音频设置。（这款应用是由30岁的丹尼尔•艾克的公司开发的）。一开始Spotify的用法和Pandora很像，输入一首歌曲或一个歌手的名字，Spotify就会播放类似格调的音乐。一开始我输入的是说唱歌手卡迪小子的名字，然后我又回到Zite上阅读文章。在阅读的过程中，Spotify播放了很多我已经听过、而且也很喜欢的歌曲。还有几个新歌手的歌曲我也很喜欢，但是也有不少是我不喜欢的。和Pandora一样，用户只能快进几首曲子。Spotify里当然也有广告，而且其中大部分广告都是一个叫“战前女神”的乡村乐队的幕后花絮，而我对这支乐队一点兴趣也没有。Spotify和Pandora唯一显著的区别就是，Spotify播放的大量音乐都来自我输入的那个歌手，相比之下其他歌手的歌曲则非常少。比如在我输入卡迪小子、布鲁斯•斯普林斯汀等歌手时都出现了这种情况。这并不是坏事，因为或许你很喜欢你选择的这个歌手，但是的确少了些发现的乐趣。

    当天晚上吃晚饭的时候，Ness向我们推荐了一家叫Birch and Barley的餐厅，我的朋友早就说过要到这家餐厅吃饭了，结果这家餐厅也没有让我们失望。我俩和另一个卖酒的朋友一起吃了晚饭。餐厅为我们端来了一道道的意大利面、鱼和腌肉，每次还有三种不同的啤酒。第二天早上，虽然昨晚的美食还没完全消化掉，我们又坚决地为中午的大餐做起了准备。这次Ness为我们推荐一家叫Estadio的西班牙餐厅，说我们有81%的可能会喜欢。不过，那家餐厅的菜让人很失望，我的厨师朋友也觉得非常一般，因为我们认为Ness这一次推荐失败了。

    当然，一款美食应用不可能每次都100%命中目标，一款音乐应用也难免会播放一些你不喜欢的音乐。就连Zite也是一样，虽然大多数时候都是成功的，但也不可能报道我喜欢的所有话题。它不可能取代那几家我作为忠实读者每天都要访问的新闻网站。使用Spotify的时候，即便有时候它给我推荐一首很好听的歌，而这首歌来自一支对我来说比较陌生的乐队，结果往往是我并不喜欢这个乐队，我喜欢的这首歌对于这个乐队来说也是超常发挥的作品。它的音乐品味永远赶不上我的一个音乐家朋友，也赶不上我的某个音乐评论人朋友。对Ness来说也是一样：如果我们放弃了Ness，只去我的厨师朋友推荐的餐厅，那么我们吃的每一顿饭应该都是非常美妙的。

    Zite allows you to add sections, giving it a newspaper feel, and although my own are rather generic (sports, gadgets, literature, fiction, marketing), they can get more hyper-specific than you'd ever think of wanting (knitting, mixology, quantum physics, and consciousness are all options). It is not without its flaws: The app once gave me a sports story from Oregon'sMcMinnville News-Register about two NCAA football teams I care nothing about, and in the Boston section I've added, stories frequently appear that have nothing to do with Boston apart from a mention. But its errors are rare and become more so the more I use it. When I fired it up in the apartment that day, I was genuinely interested in four of the five articles that showed up on page one. In total I spent nearly an hour swiping through the 19 pages of my main section, reading articles both short and long. Not bad.

    Johnson, the brainy geek behind Zite, is a Stanford Philosophy major who spent three years at SAP (SAP) and then worked on Microsoft's (MSFT) Bing. He also worked for search startups like Powerset and SideStep. Johnson is enraptured with the idea of perfecting a recommendation engine, which, after all, is what Bing is supposed to be. Such a thing, really, could never be perfect every time, nor would you want it to be: From a developer standpoint, he says, "the problem with any of this stuff is that serendipity is pretty easy to create -- it's what we like to refer to as randomness -- and really what you want is some kind of guided serendipity, not total randomness." The more I use Zite, the less I see randomness, which of course is the point, but as randomness vanishes, so does the magic of discovering something totally new to me.

    Next up: music. I wasn't about to pay for the much-hyped Spotify just yet, but the app (founded by 30-year-old Daniel Ek) will let you use its Radio setting for free on a tablet, which essentially works just like Pandora (P). Enter a song or artist and Spotify plays you tunes it deems similar. I began by typing in Kid Cudi and returned to Zite. While paging through articles, I heard a fair amount of stuff I already knew and liked, plus some new artists I liked, but also more than a few misses. As with Pandora, you may only skip a certain number of tracks. There are ads of course, and many are, well, bad -- the majority relentlessly pitched me "behind-the-scenes" footage of Lady Antebellum, a country group I had zero interest in trying. The only obvious, stark difference from Pandora I could identify was Spotify playing me an awful lot of songs by the artist I entered, as opposed to songs by similar artists. This happened with Kid Cudi, Bruce Springsteen, and others. That's not a negative, assuming you like the artist you select, but it does mean less discovery.

    That night, for dinner, Ness gave us Birch and Barley, which my friend had on his list anyway. It didn't disappoint. We ate with a third friend, who owns a wine shop, and the chef stuffed us with course after course of pasta, fish, and cured meat, plus three different beer samples each time. But in the morning, resolute to have another big lunch despite our food hangover, Ness suggested a Spanish restaurant, Estadio, promising an 81% likeness. The food disappointed, and my chef friend was unimpressed and had to conclude the app had failed us this time.

    Of course, a dining app isn't going to be on-target 100% of the time, nor can a music app avoid playing you some songs that you dislike. Even Zite, which I found most successful at what it does, cannot cover everything for me. It won't replace the handful of news sites to which I'm loyal and visit every day. On Spotify, even when I hear a great song by a band that is new to me, it often turns out I don't like the band, and the song I loved was an anomaly. The app will never have as much cred with me as one of my musician friends or authoritative, music-critic peers. The same goes for Ness: Had we ditched it and gone only to restaurants my chef pal already knew about, there wouldn't have been a single less-than-fantastic meal.

    虽然这些应用每一款都算不错，但是它不能取代一个熟悉你、也熟悉你爱好的人，因为还不算是突破性的进展。无论是对于这三家创业公司，还是对于很多正在开发推荐引擎的公司来说，更紧迫的问题是，这些公司必须扩展自己的能力，才能实现里斯为Ness设定的目标，也就是“在人们寻找下一个他们可能喜欢的事物时，充当他们可信的信息来源”。这意味着Ness必须要涵盖更多的东西，而不仅仅是餐厅。

    里斯表示，利用Ness现有的技术，Ness还可以用来推荐书籍、电影、旅行目的地或是夜生活场所。约翰逊认为对于Zite来说也是一样。他同时指出，开发一个全方位的推荐引擎将面临一个难以避免的挑战，也就是要完善一个“社交图谱里的Google”。或许最终的产品将成为谷歌的一个强力竞争对手，它可能通过两种方式提供人们要寻找的一东西，一是严格根据人们的搜索历史，二是根据人们的个人交往情况进行推荐。但是要注意，这两者都基于一点，也就是所谓的大数据。

    但是我们真的想要这种东西吗？脸谱（Facebook现在想做的就是统一整个网络，Facebook原本是做社交起家，然后成了你疯狂贴照片的地方，现在又包括了即时通讯、地理位置服务、社交游戏和一个商业市场。现在我们还不知道，这种“大一统”是否会让用户买账，抑或会令有些用户感到心烦，甚至导致流失用户

    许多贪心的网络用户喜欢使用不同的应用来实现不同的功能。比如我喜欢用Kickstarter向一些很酷的社区项目提供资金，用Twitter发布新闻，用Facebook进行个人分享，用Instagram发布照片。我相信这些应用在各自的领域都是完美的，我也不想要一个“一站式的服务”。更重要的是，许多社交应用，比如Kickstarter和Tumblr等之所以吸引人，并不是应为他们尝试着去懂你，而是它们的用户喜欢把自己的兴趣投射到这个平台上。

    亚马逊(Amazon)和Netflix在预测技术上还停留在“1.0”时代（约翰逊把这两家公司称为“老经典”），但它们仍然非常成功，而且可能很难打败。亚马逊的推荐引擎依赖于一个基本公式，它向你推荐的产品基于你的浏览史、购物史，并且与其他顾客购买的产品进行关联。这个模式是成功的。Netflix采用的也是简单有效的法子，随着你选择电影的时间越来越长，Netflix会变得越来越聪明。如果说这几个科技巨头在预测技术领域都做得不错，那么像Spotify或Ness这样的公司，如果指望单纯靠增加几个细分领域就能获得成功，恐怕很难。

    目前来说，我还是依靠我自己的人肉推荐引擎吧。因为我的朋友和家人比任何一款应用软件都更加了解我。至于我是否了解我自己，这个问题就留给谷歌（Google）、Facebook、苹果（Apple），以及大量关于你和我的数据吧。（财富中文网）

    译者：朴成奎

    That each app does a decent job, but cannot replace the usefulness of a live person familiar with you and your likes, is no breakthrough epiphany. The more pressing question may be which startup -- from these three or from the myriad others already out there -- will end up expanding its repertoire to achieve what Reese says is his mission with Ness: "become that trusted source for people to find out the next thing they'll like." That sounds like it would encompass a lot more than restaurants.

    Reese says Ness could just as easily use its technology to recommend books, movies, travel destinations, or nightlife activities. Then again, Johnson believes the same of Zite. He also posits that coming up with an all-in-one recommendation engine will be inextricably linked to the challenge of perfecting a "Google for the social graph." Perhaps the final product will be a Google (GOOG) rival that offers what you're seeking in two forms: one based strictly on your own search history, the other inspired by your personal connections. Take note: It will all rely on big data.

    But would we want such a thing? Consolidating the Web is exactly what Facebook (FB) is trying to do; what began as a place for checking people out eventually became your photo dumping grounds and now includes instant messaging, location services, social games, and a commercial marketplace. It remains to be seen whether all of this is annoying enough to cost them users or if people will just give in.

    Many avid Internet users prefer to have their various functions in separate silos. I likeKickstarter for funding cool community projects; Twitter for breaking news; Facebook for more personal sharing; and Instagram for photos. I trust each of these entities for the activity it's perfected. I wouldn't want a one-stop shop. Moreover, many of these outlets, like multifaceted Kickstarter as well as, say, Tumblr, are appealing precisely because they do not attempt to know you; instead, their users tend to project their own interests onto the platform.

    Both Amazon (AMZN) and Netflix (NFLX) (Johnson calls these "the old classics"), which are akin to the "1.0" of predictive technology, still work pretty well and may be hard to beat. Amazon's recommendation engine relies on a basic formula (despite the highfalutin term they've given it, "item-to-item collaborative filtering") that suggests products to you based on your viewing history, your purchase history, and which related products other customers bought. And it works. The same goes for Netflix, which, as you spend more time choosing movies, becomes quite smart indeed. If these giants of the tell-me-what-to-try space are doing just fine, it may be tough for a Spotify or Ness to simply add more verticals and hit the gas pedal.

    For now, I'll rely on my own human recommendation engine, thanks. My friends and family know me better than any one app ever could. Whether I know myself, well, that's probably a question for Google, Facebook, and Apple (AAPL), and their vast piles of data on me -- and you.