Photos Spare Cycles MythBusters

Talk: Recent Innovations in Search and Other Ways of Finding Information

Peter Norvig, Google; Ken Norton, Yahoo!; Mark Fletcher, Bloglines/Ask Jeeves; Udi Manber, A9; Jakob Nielsen, NN Group

I went with bp and Neil to a BayCHI talk on "Recent Innovations in Search." I agree with bp's sentiment -- there were some interesting moments, but the talk was short on revelations or insights. I guess that is to be expected as the title of the talk is past focused ("Recent Innovations") rather than future focused ("Future Innovations"); it's hard to believe that the panelists would give away yet unrevealed technologies they were working on. I'm going to try and save as much effort as possible, given that bp posted his notes. In fact, as I am going to crib from his notes, or just omit what he already has, you should just go read them instead.


  • Peter Norvig, director of search quality, Google.
  • Mark Fletcher, vice president & general manager, Bloglines (founder) at Ask Jeeves.
  • Udi Manber, CEO,
  • Ken Norton, senior director of product management, Yahoo! search (formerly director of product strategy at Inktomi).
  • Jakob Nielsen, Nielsen Norman Group

Format: 5 minute overview by each panelist, followed by questions from the moderator, then questions from the audience

Peter Norvig's 5 minutes: * overview of recent Google technologies: question and answer, search for handhelds, maps for handheld search, maps (dragging is the differentiating feature) with satellite imagery, suggest (Ajax), desktop search

Ken Norton's 5 minutes: * Intro: noting the people sitting on the ground in front of him, he said that he was an introvert, and he was glad that he could look downwards and still feel like he was making eye contact * 2/2004: Yahoo launched its own search engine (previously used Google) * My Yahoo! search beta: enable users to search/share/save content. Similar to having bookmarks for searching (I wonder how different this will be from A9). * video search (can add content using media rss) * desktop search also searches Y!IM, Y! mail, Y! address book * Y!Q: key observation is that search is contextual, want to provide search tools that can be placed in context, at the "point of inspiration." Example he showed was a "search related info" link at the bottom of an article that popped up a Yahoo search in place. * Y! mobile * creatve commons search * Flickr acquisition: bring Flickr DNA into Yahoo, community-buildig, tagging * Coming: personal/contextual/multimedia/desktop/local/travel/social * * Overall, I got the impression that Yahoo's strategy is very much within the remix culture, i.e. trying to provide tools that enable users to use the content (vs. just finding it)

Mark Fletcher's 5 minutes * Mark's Intro: "Battle of the Introverts." "I'm not going to look above the second row at all." * Blog content is the fastest growing content segment on the Web. 1.6M new articles/day. * demo of bloglines with his own personal account. For those of you somehow interested, Mark's folders are: * Friends * VCs * General Bloggers * Press/Journalists * Tech * Aggregators * Entertainment * News Aggregators * Test * I was transcribing all I could about his bloglines account, because I noticed some interesting things when he expanded the "Aggregators" folder. In that folder he had "Competitors/Clients," "Competitors/Services", keyword search feed for "Bloglines", and I believe two different feeds dealing with "Rojo." These folders were interesting because 1) they demonstrated that Bloglines has a PubSub feature that neither bp or I knew about, and 2) Bloglines seems really focused on Rojo

Udi Manber's 5 minutes * Showed off their photo Yellow pages and how you oould walk or "run" down the street. The important insight was that this was search without keywords. You could say "Its this place on Castro Street" and then try to find it by "walking" up and down the street. * Started talking about the SUV + camera technology used. Have a street map that updates to show where images have been collected from so far (GPS). Video camera dumps data direct to laptop. * Anecdote regarding a security "exception." Amazon policy requires computers to go to screensaver lock after 15 minutes -- pretty hard to unlock while you're driving. The compromise was that they plug a mouse into the laptop and place it on the floor of the SUV, where the vibrations are enough to keep the laptop "active." Problem was that the mouse would flip over and start clicking, so they created a "neutered" mouse that has no buttons. * Showed a video of one of their drivers who had stopped to adjust the camera, only to find out that he stopped in front of the State Department, which aroused the interest of the security detail there. There is a spliced video showing more and more security folks coming out to investigate, culminating with one of the officers taking a photo. Early on, the driver had been asked if the video camera had any tape in it. The driver answered honestly, "No" (it dumps direct to hard drive). The officer didn't quite get it, which is why they have the video of this all occurring. Udi's punchline: just as in the search paradigm, the point is to ask the right question.

Jakob Nielsen's 5 minutes * words/query string: * 1994: 1.3 * 1997: 1.9 * 2004: 2.2 * People are adapting to search technology, getting more confident * Success rate for search: * All searches: 42% * Those who scrolled the first page: 48% * Low experience users: 32% * High experience: 50% * External search engine: 56% * Internal search engine on Web site: 33% * Using a site's own search engine is a miserable failure, yet it also constitutes a large portion of search queries * Intranet search success is even worse * 40% of search population is not search literate * He showed a video of a particular user in the 40% group trying to do a search on headaches. There are multiple, painful search errors, from text going into the wrong text box, submitting with empty search boxes, using the address bar (www.head, highlighting a word and then clicking "submit" (i.e. expecting the highlighted text to be submitted), etc... His point was to break people's expectations of what a large percentage of the user base looks like, but I'm not sure that the audience quite internalized this point (too caught up in humor). * Out of the various usability problems classified across 25 major Web sites, 11% of them related to search

Is search a miserable failure * ... * Udi: search is hard to do, many sites have only 1 programmer. Wants to make search something that you can get off the shelf. * Norvig: anecdote from personal experience. in 1999, trying to search for info on a thinkpad on IBM's site, felt like the user in Jakob's video. Went to Google instead and found it immediately -- Google, an external search engine, did better than IBM on its own content.

How has search evolved? * .. * Norton: hard to let users know about new tools when their context is a search results page. * Fletcher: Ask Jeeves search query length is getting shorter (moving away from natural language). The long tail phenomena applies to search queries -- many unique search queries never seen before dominating. * Udi: search query length is proportional to the size of the text input box you give users, i.e. the size of the text input box is a usability consideration. * Nielsen: average search box is 18 characters wide, but soon recommendation will be 30 characters.

(Didn't hear question well) Avoid becoming portal? * Norton: search box is the UI. understand user's intent * Udi: problem in that some queries are easy (e.g. "population of India") vs. others are hard even though they seem appropriate to the user (e.g. "what is the best DVD player" -- "best" is a bad search term but users expect it to work). Also, "How many powerful tools do we want to give users?" Fine balance between power and usability. * Norvig: the problem is that we don't allow the user to have a dialog, don't even tell the user the language the search box understands. Wants to get more towards the librarian model, where you can have a dialog to refine your search, giving the user advice along the way. You use more than 2.2 words when talking to a librarian.

Tagging: is it a new paradigm? * Udi: Not different. CS Joke: "Anything you can do I can do meta." Metadata is just data. * Norton: what's interesting is that it's not different. Link text functions like tags. The more metadata the better. * Jakob: have to figure out passive techniques for collecting metadata (e.g. what is the user eyeballing). Tagging is too much work for most. * Norton: people will tag if they derive benefit (e.g. flickr) * Udi: MapR example -- looked at Flickr tags and tried to overlay ones with geographical tags onto a map. Problem with too many photos appearing in Maine -- the "ME" tag was interpreted as Maine.

CHI theme: social context important for evaluating data * Norton: 360 is about trying to build a framework for people to manage relationships and share info (vs. services like LinkedIn which try to connect you with other people). Need framework to capture social interactions (e-mail, IM) * Fletcher: need to do more with blogs and blogrolls as sources of this data * Norvig: people looking for info they trust, and they trust their friends, but they also want to discover new things beyond their friends -- need to merge. * Jakob: need to swing pendulum back towards groupware * Udi: not just social context, but context in general. For example, the same person searching twice may be looking for entirely different things. A9 remembers previous search results to try and provide some of this context.

What don't you use search for (example of Norvig asking Greenspun for camera advice) * Norvig: stuff that's not available or too complicated. Also, search not good for synthesizing data

Current success is from leveraging the info on the web. How do we continue to leverage that kind of information? Is there a significant investment in allowing the publishing community to expand, making the browser a better publishing platform?

  • Fletcher: blogging, but could do better (e.g. how many people have written an essay in a browser)
  • Udi: 10 years ago we were thinking that HTML was so easy that anyone could write it. Now we have much better tools.
  • Norvig: don't want to make it too easy, want the information to be worth the effort of writing it
  • Udi: tool to send a fax to a random person at Google
  • Jakob: search engines are leeches, sucking value out of the Web. tradeoff between content producers and search engines. customers are loyal. will use Amazon to search for and buy their book.

Spoken language interfaces * Norvig: 2 aspects: 1) do you use a keyboard (did a test of this, not a great success), 2) can you make a dialog of the process * Norton: has to be seamless. Can't have spoken language search that then gives you a textual search result. * Jakob: would have to be combination of verbal + visual

Post a comment


related entries.

what is this?

This page contains a single entry from kwc blog posted on April 12, 2005 10:24 PM.

The previous post was Budget stat visualizations.

The next post is Search Keyword Fight I.

Current entries can be found on the main page.