Gary Price posts details on Google's new movie shortcut. Stick a "movie:" prefix on a keyword search and it'll bring up a list of matching movies and movie reviews.
It appears to search in the movie title, description, and reviews, making it a good way to find movies that match a rough description or to find a movie if you can't remember the exact title.
Here are movie searches for "monty python" and for "hong kong crime".
The movie reviews appear to be spidered from the web, online news sources, and Usenet newsgroups. Google does something similar for the product reviews in Froogle.
Amazon.com can't be happy about this trend. Product and movie reviews at Google are an alternative to the product reviews at Amazon and the movie reviews at Amazon-owned IMDB.
Friday, February 25, 2005
TiVo, Flickr, WSJ, Wikinews
A few recent tidbits worth reading:
George Hotelling summarizes the rumors that Apple might be buying TiVo.
Om Malik summarizes the rumors that Yahoo might be buying Flickr.
Adam Penenberg slams the Wall Street Journal Online, saying that their subscription model dooms them to irrelevance.
Simon Waldman says Wikinews isn't very good. He blames the rapidly changing nature of news. "A news operation starts with a clean slate every day," he says, making it a different and harder problem than Wikipedia.
George Hotelling summarizes the rumors that Apple might be buying TiVo.
Om Malik summarizes the rumors that Yahoo might be buying Flickr.
Adam Penenberg slams the Wall Street Journal Online, saying that their subscription model dooms them to irrelevance.
Simon Waldman says Wikinews isn't very good. He blames the rapidly changing nature of news. "A news operation starts with a clean slate every day," he says, making it a different and harder problem than Wikipedia.
AOL Local Search
Michael Bazeley writes about AOL's entry into the crowded land of local search.
For part of AOL Local, Topix.net will provide local news, another excellent partnership in Topix's recent deal-making spree. Congrats, Rich!
See also my earlier post, "Local search is hard".
For part of AOL Local, Topix.net will provide local news, another excellent partnership in Topix's recent deal-making spree. Congrats, Rich!
See also my earlier post, "Local search is hard".
Saturday, February 19, 2005
Tivo Suggestions paper
I love TiVo. I can't imagine TV without it.
TiVo has a feature, TiVo Suggestions, that tries to recommend shows you might like. It's a great idea, helping viewers discover new shows they might not know about. Unfortunately, the quality is mediocre.
I've been curious why the recommendations are poor. And now I can find out. Kamal Ali and Wijnand van Stam from TiVo published a paper at KDD 2004, "Tivo: Making Show Recommendations Using a Distributed Collaborative Filtering Architecture" (MS Word doc file).
The paper says:
An early clue on the quality problems is here:
Getting into what kind of improvements could be made, let's start by summarizing their current system. Each TiVo is essentially a low-end PC connected to TiVo's server farm with a slow and only intermittently available pipe (modem over phone line).
At a high level, TiVo Suggestions currently works by sending all the ratings across to their server cluster from each TiVo, generate correlations between the shows, and send back the list of correlations to each TiVo. Because of the slow pipe and the amount of data to be processed, only 1/16 of the correlations are computed each day, so the correlations take up to 16 days to update. Each client TiVo computes new recommendations "at least once a day" using the correlation data.
The algorithm they are currently using, "item-item collaborative filtering", is designed for real-time, high performance applications. Since the recommendations don't need to be computed in real-time, just "at least once a day", it's worth considering whether the recommendations should be computed entirely on the server cluster. If they aren't updating the recommendations in real-time, a real-time algorithm probably is not the best choice.
They compute only 1/16 of the correlations each day, apparently because of computational resource constraints on their cluster, but this creates a bad user experience. Instead, if they really are hardware constrained, sampling or other data reduction techniques could improve performance without reducing quality.
They say they haven't experimented yet with biasing toward more recent data, another way of reducing data and improving the apparently quality of the recommendations.
They are using Pearson correlation as their similarity metric. The choice of similarity metric can have a broad impact on recommendations, biasing the recommendations toward more surprising or unusual recommendations or pushing them into the mainstream.
They made a variety of decisions elsewhere that impact recommendation quality. They combine predictions using a weighted linear average. They have a variety of thresholds in their correlation computation (min-pair, min-single, etc.). They arbitrarily setting the confidence ranges for the content and collaborative filtering-based predictions. Experimentating with other choices here could find improvements.
It's a shame. TiVo is great, but more and more PVRs are coming on the market. Recommendations could be a key differentiator for TiVo.
If my TiVo knows me, knows what I like, and helps me find it, the experience on any other PVR would seem hollow, like I just lost a best friend. With the current quality of TiVo Suggestions, my TiVo feels more like a clueless stranger than a close friend.
[paper via Matt Haughey]
TiVo has a feature, TiVo Suggestions, that tries to recommend shows you might like. It's a great idea, helping viewers discover new shows they might not know about. Unfortunately, the quality is mediocre.
I've been curious why the recommendations are poor. And now I can find out. Kamal Ali and Wijnand van Stam from TiVo published a paper at KDD 2004, "Tivo: Making Show Recommendations Using a Distributed Collaborative Filtering Architecture" (MS Word doc file).
The paper says:
- TiVo uses an item-item (show to show) form of collaborative filtering which obviates the need to keep any persistent memory of each user's viewing preferences at the TiVo server .... It uses k-nearest neighbor with Pearson correlation to make show recommendations ... The collaborative filtering system is augmented on the client by a content-based Bayesian recommendation system to address the cold start problem for new users and shows.
An early clue on the quality problems is here:
- We have not performed formal empirical evaluations of its accuracy .... The number one item on our agenda for future work is a thorough empirical evaluation of the quality of suggestions.
Getting into what kind of improvements could be made, let's start by summarizing their current system. Each TiVo is essentially a low-end PC connected to TiVo's server farm with a slow and only intermittently available pipe (modem over phone line).
At a high level, TiVo Suggestions currently works by sending all the ratings across to their server cluster from each TiVo, generate correlations between the shows, and send back the list of correlations to each TiVo. Because of the slow pipe and the amount of data to be processed, only 1/16 of the correlations are computed each day, so the correlations take up to 16 days to update. Each client TiVo computes new recommendations "at least once a day" using the correlation data.
The algorithm they are currently using, "item-item collaborative filtering", is designed for real-time, high performance applications. Since the recommendations don't need to be computed in real-time, just "at least once a day", it's worth considering whether the recommendations should be computed entirely on the server cluster. If they aren't updating the recommendations in real-time, a real-time algorithm probably is not the best choice.
They compute only 1/16 of the correlations each day, apparently because of computational resource constraints on their cluster, but this creates a bad user experience. Instead, if they really are hardware constrained, sampling or other data reduction techniques could improve performance without reducing quality.
They say they haven't experimented yet with biasing toward more recent data, another way of reducing data and improving the apparently quality of the recommendations.
They are using Pearson correlation as their similarity metric. The choice of similarity metric can have a broad impact on recommendations, biasing the recommendations toward more surprising or unusual recommendations or pushing them into the mainstream.
They made a variety of decisions elsewhere that impact recommendation quality. They combine predictions using a weighted linear average. They have a variety of thresholds in their correlation computation (min-pair, min-single, etc.). They arbitrarily setting the confidence ranges for the content and collaborative filtering-based predictions. Experimentating with other choices here could find improvements.
It's a shame. TiVo is great, but more and more PVRs are coming on the market. Recommendations could be a key differentiator for TiVo.
If my TiVo knows me, knows what I like, and helps me find it, the experience on any other PVR would seem hollow, like I just lost a best friend. With the current quality of TiVo Suggestions, my TiVo feels more like a clueless stranger than a close friend.
[paper via Matt Haughey]
Friday, February 18, 2005
Matching content to audiences
Newspapers are increasingly talking about personalized news. Mary Lou Fulton (VP, Bakersfield Californian) recently said:
Content is king. Newspapers should exploit new distribution systems to get their content in front of the right audiences.
- What if we thought about it a different way, and thought of our pool of content (local, wires, syndicated content, etc) separately from a single newspaper. We could slice and dice that content differently and come up with many versions of the publication for many audiences.
Content is king. Newspapers should exploit new distribution systems to get their content in front of the right audiences.
Thursday, February 17, 2005
Yahoo, the long tail, and personalization
Havi Hoffman has some interesting tidbits on personalization in his post on the Yahoo Search Blog.
Personalization exposes the long tail. It learns your interests, digs deep, and surfaces the gems from the internet's cacophony.
Fantastic to see Yahoo thinking about personalized search and recommendations in the long tail.
- The mainstream is not the only stream. There's room now for babbling brooks, crooked creeks, and tributaries where trends pick up momentum before they flow downstream. There's an audience for many voices.
We all need long-tail tools that support self-expression, personalized search, recommendations, and trust.
Personalization exposes the long tail. It learns your interests, digs deep, and surfaces the gems from the internet's cacophony.
Fantastic to see Yahoo thinking about personalized search and recommendations in the long tail.
Wednesday, February 16, 2005
Local search is hard
John Battelle posts on competing in the local search and yellow pages market:
John is optimistic that local merchants would come to Yahoo or Google's website to self-manage their information and advertising, but I doubt you could get anything like the coverage you need with a self-service model. Many of these merchants don't even have a web site. They aren't tech savvy.
The Yellow Pages has uses hundreds of sales people because it has to. They need to physically send salespeople out to talk to each merchant. Getting complete coverage requires a massive, expensive sales force. That's why local search is hard.
See also "Down on local search".
- I asked [Peter Negulescu (VP, SFGate)] why SFGate isn't an aggressive player in the local online advertising market ... He chided me for my ignorance regarding the 800-pound gorilla of local markets -- the Yellow Pages. "They have like 6-700 local sales people in every major region," he told me. "They visit every merchant in town." In other words, the Chronicle can never compete.
John is optimistic that local merchants would come to Yahoo or Google's website to self-manage their information and advertising, but I doubt you could get anything like the coverage you need with a self-service model. Many of these merchants don't even have a web site. They aren't tech savvy.
The Yellow Pages has uses hundreds of sales people because it has to. They need to physically send salespeople out to talk to each merchant. Getting complete coverage requires a massive, expensive sales force. That's why local search is hard.
See also "Down on local search".
VCs and the "me too zone"
Brad Feld (Director, Mobius Venture Capital) writes about trend-following in the VC industry:
- Every emerging market hits a point where there is a mad rush of early stage entrepreneurs and VCs piling in. In some cases, it drives rapid innovation; in most it creates near term over-saturation, lots of irrational financings, and plenty of carnage as the laws of Darwin play out over the next couple of years.
I'm afraid we just hit that point with RSS / Blogging. The meme has spread broadly -- which is great. Now we'll watch all gods children pile in to try to get something up and running in this "space" .... It's as if the whole world wakes up one day and starts working on the same types of things.
There is a huge adoption (and innovation) curve ahead of everyone who is doing stuff with RSS / blogging -- and there are plenty of good investments left to make and companies to create -- but the noise and clutter is about to get really loud.
Autolink in Google Toolbar
Gary Price writes about the latest version of Google Toolbar. It includes a feature called Autolink:
Useful, but this one makes me a little squeamish. Google is now modifying the web pages displayed. The modifications are useful, sure, but what does it do to Mapquest to have all addresses everywhere pointing to Google Maps? What does it do to Barnes & Noble if all ISBNs point off to Amazon?
Taking this a step further, what if your web browser rewrote all your web pages? Would you be happy if IE pointed all addresses it recognized off to Encarta Maps?
- With "AutoLink" enabled, web pages will be "enhanced" with additional links if Google thinks additional information might be helpful. For example, say your browsing a web page with numerous addresses on it. AutoLink will turn each of those addresses into direct links to the Google Maps database.
Useful, but this one makes me a little squeamish. Google is now modifying the web pages displayed. The modifications are useful, sure, but what does it do to Mapquest to have all addresses everywhere pointing to Google Maps? What does it do to Barnes & Noble if all ISBNs point off to Amazon?
Taking this a step further, what if your web browser rewrote all your web pages? Would you be happy if IE pointed all addresses it recognized off to Encarta Maps?
Monday, February 14, 2005
Tuning search to individuals
Interesting tidbit on Yahoo's plans for personalized search from a HBS Working Knowledge article:
- Yahoo is able to collect data, such as click patterns, from its users and use the information to individually "tune" searches to users' personal tendencies. Such product improvements, [Yahoo Director Bradley] Horowitz said, will change the nature of search from a one-size-fits-all experience to a more individually oriented one.
Findory Favorites
Findory's newest feature, Favorites, makes it easy for readers to easily see their favorite news and weblog sources.
For example, on the BBC page, clicking the "Add Favorite" button at the upper right will put a link to BBC articles in your "My Favorites" list. You can also mark weblogs such as Boing Boing or Gizmodo as favorites.
Reading news on Findory is easy. Read a few articles. Get personalized recommendations. Click on a few sources. Add a few favorites. Read more articles. Get more personalized recommendations. It's a virtuous cycle. The more you use Findory, the better it gets!
Contrast that with a using a normal feed reader. First, you hunt down a few RSS feeds. Manually cut-and-paste URLs into the feed reader. Hunt down more feeds. Manually add them. Skim all your feeds desperately looking for interesting articles. Realize that many of the feeds are boring and useless. Remove those feeds. Laboriously hunt down more feeds. Realize you now have too many feeds and can't read them all. Remove some feeds. Ugh. You spend more time configuring and customizing than reading news.
Customization and personalization are at opposing ends of a spectrum. Most feed readers are all the way at the extreme of customization, requiring readers to do work -- lots of work -- to be able to read news.
Findory is a different. It's personalized and learns what you need. It requires no effort. It's easy. It all just works.
For example, on the BBC page, clicking the "Add Favorite" button at the upper right will put a link to BBC articles in your "My Favorites" list. You can also mark weblogs such as Boing Boing or Gizmodo as favorites.
Reading news on Findory is easy. Read a few articles. Get personalized recommendations. Click on a few sources. Add a few favorites. Read more articles. Get more personalized recommendations. It's a virtuous cycle. The more you use Findory, the better it gets!
Contrast that with a using a normal feed reader. First, you hunt down a few RSS feeds. Manually cut-and-paste URLs into the feed reader. Hunt down more feeds. Manually add them. Skim all your feeds desperately looking for interesting articles. Realize that many of the feeds are boring and useless. Remove those feeds. Laboriously hunt down more feeds. Realize you now have too many feeds and can't read them all. Remove some feeds. Ugh. You spend more time configuring and customizing than reading news.
Customization and personalization are at opposing ends of a spectrum. Most feed readers are all the way at the extreme of customization, requiring readers to do work -- lots of work -- to be able to read news.
Findory is a different. It's personalized and learns what you need. It requires no effort. It's easy. It all just works.
Saturday, February 12, 2005
Yahoo building weblog search?
Steve Rubel and Nathan Weinberg speculate that Yahoo is building a weblog search similar to Technorati and Feedster after seeing evidence of a new Yahoo RSS crawler.
See also my earlier post, "Will Technorati die?"
See also my earlier post, "Will Technorati die?"
Implicit vs. explicit sharing
Ross Mayfield (CEO, Socialtext) slams personalization in favor of social software:
Personalization relies on the strength of the community. It is the shared data between users that allows the system to find relationships and make recommendations.
But there is one major difference. With personalization, sharing is implicit and anonymous. With social software, sharing is explicit and public.
There are two problems with explicit sharing:
- The basic problem with Personalization is that tailoring information to you limits social discovery. Users contribute value to the database only for them and the service provider, not for each other ... Nowhere in this mode is sharing, conversations, remixing and socializing information.
By contrast, consider how social software enables people to create their own networks. Groups form, information is shared and implicit and explicit relationships are fostered.
Personalization relies on the strength of the community. It is the shared data between users that allows the system to find relationships and make recommendations.
But there is one major difference. With personalization, sharing is implicit and anonymous. With social software, sharing is explicit and public.
There are two problems with explicit sharing:
- It requires work. A lot of work. People don't like to do work.
- The quality of recommendations is low. Why? Chris Anderson said it best: "No matter who you are, someone you don't know has found the coolest stuff." Limiting discovery to what your friends have found cripples the system.
- The appeal of personalization is sheer convenience. Today social software fails, with a few exceptions, to deliver the same level of convenience at scale, but give it time.
Discovery in the information stream
Rich Skrenta (CEO, Topix.net) has a long post that touches on the incremental web, streaming information based on keywords vs. topic, human vs. automated aggregators, and discovering content in the long tail. An excerpt:
Even if you monitor just a few tens of sources, you are facing a daily stream of hundreds or thousands of articles. It's a painful, overwhelming task to manually skim it hunting for relevant content. There is precious little discovery in the current model.
- There are 4-8 million active blogs now. At this size, you can still "know" the top bloggers, and find new posts worth reading by clicking around. But when the blogosphere grows 100X or 1000X, the current discovery model will break down. You'll need algorithmic techniques like Topix.net or a Findory to channel the most relevant material from the constant flood of new content.
Even if you monitor just a few tens of sources, you are facing a daily stream of hundreds or thousands of articles. It's a painful, overwhelming task to manually skim it hunting for relevant content. There is precious little discovery in the current model.
Friday, February 11, 2005
First, make sure it works
Todd Bishop at the Seattle PI uncovers some amusing flaws in MSN Search's advertising.
One particular MSN Search ad suggests that you should search for "Pizza in Redmond". Curiously, Todd finds, MSN Search has the worst results by far on that query of any of the major search engines.
MSN is rumored to be spending $150M for this advertising campaign. I'm not sure they're getting their money's worth.
One particular MSN Search ad suggests that you should search for "Pizza in Redmond". Curiously, Todd finds, MSN Search has the worst results by far on that query of any of the major search engines.
MSN is rumored to be spending $150M for this advertising campaign. I'm not sure they're getting their money's worth.
Jeeves loves Mozilla
Ask Jeeves EVP Tuoc Luong says that they are considering making Ask Jeeves Desktop Search open source, partnering with the Mozilla Foundation, and releasing a branded version of the Firefox browser.
I like the idea of making Ask Jeeves Desktop Search open source. With the flood of desktop search applications, releasing the source for Ask's product is a way to spur adoption and continued innovation.
As for a branded version of Firefox, I'm skeptical about how many people would download it, but any users it had probably would tend to come to Ask Jeeves frequently.
Google recently embraced the Mozilla Foundation as well, hosting the default Firefox start page and paying the salaries for two of the lead developers.
I like the idea of making Ask Jeeves Desktop Search open source. With the flood of desktop search applications, releasing the source for Ask's product is a way to spur adoption and continued innovation.
As for a branded version of Firefox, I'm skeptical about how many people would download it, but any users it had probably would tend to come to Ask Jeeves frequently.
Google recently embraced the Mozilla Foundation as well, hosting the default Firefox start page and paying the salaries for two of the lead developers.
Thursday, February 10, 2005
Google to help Wikimedia
Google has offered to help Wikimedia -- the non-profit that maintains the free, community-maintained, and very cool Wikipedia encyclopedia -- with hosting and bandwidth.
Apparently, the offer has no strings attached, no requirement for advertising.
This is a generous move by Google. But it is also true that Google has a lot to gain from supporting Wikipedia and other Wikimedia projects.
By using Answers.com for definitions, Google already indirectly includes some Wikipedia content. Google may be planning to include more encyclopedia content, perhaps to counter the Encarta integration into MSN Search, perhaps just to improve the usefulness and relevance of search results.
Like Google's support of Firefox, Google's generosity is not without self-interest. Nevertheless, it's encouraging and gratifying to see Google helping out these tech non-profits.
[via Slashdot]
Apparently, the offer has no strings attached, no requirement for advertising.
This is a generous move by Google. But it is also true that Google has a lot to gain from supporting Wikipedia and other Wikimedia projects.
By using Answers.com for definitions, Google already indirectly includes some Wikipedia content. Google may be planning to include more encyclopedia content, perhaps to counter the Encarta integration into MSN Search, perhaps just to improve the usefulness and relevance of search results.
Like Google's support of Firefox, Google's generosity is not without self-interest. Nevertheless, it's encouraging and gratifying to see Google helping out these tech non-profits.
[via Slashdot]
Social computing at Microsoft Research
Bernard Moon has a fun article on his visit to the Social Computing Group at Microsoft Research.
As [MSR GM] Lili Cheng said, they are "like a startup group within Microsoft that has to pitch to other groups for them to learn about and consider our ideas." If you put up a hurdle that high, few are going to be able to jump it.
[via Robert Scoble]
- "We started with this little project ... [where] we looked at our emails, scanned them, and built [clusters] based on who you communicate with." This simple program -- which users don't have to input anything for or do anything with -- dynamically changes over time. The system automatically adds people to users' lists and groups them based on who users email and copy.
The most interesting application to come out this idea is an email program the group dubbed Inner Circle, which lists alphabetically a user's top 40 correspondents ... When an Inner Circle user clicks on a person in that list, the program displays all the conversations the user has had with that person as well as any relevant documents or links that were sent.
Another interesting application ... is Wallop, a group social network and blogging environment ... It includes a level of interactivity that other [blogging] sites can't currently match: You can post comments on just about anything (such as pictures or text), and you can mark an item -- not just the overall profile of a hot guy or girl -- on a person's page as a "favorite."
As [MSR GM] Lili Cheng said, they are "like a startup group within Microsoft that has to pitch to other groups for them to learn about and consider our ideas." If you put up a hurdle that high, few are going to be able to jump it.
[via Robert Scoble]
All of the world's information
David Kirkpatrick at Fortune interviewed Marissa Mayer (PM, Google). I particularly liked what she had to say about expanding Google's content:
- One of our themes now is "more of the information you want." That implies huge index sizes. Now it's eight billion pages, and we'll get really aggressive, growing into more forms of content, like print from books, and video.
Eric [Schmidt] had a great quote the other day. He said, "When we said all of the world's information, we meant all." So that means ... there's very little content we're not interested in.
Wednesday, February 09, 2005
CNet's new feed reader
Steve Rubel reports that CNet is about to launch its own feed reader, Newsburst. Welcome to the club, CNet.
There's already integrated feed readers in Firefox, My Yahoo, and My MSN. Ask Jeeves now has Bloglines. CNet will have Newsburst. Oh, wait, there's more, hundreds of other RSS readers, all fairly similar. There's even several newspapers trying to offer their own feed readers.
The products are largely indistinguishable. Launching undifferentiated products into a crowded space is futile. With giants like My MSN, My Yahoo, and Firefox already major players, the world doesn't need yet another feed reader.
What the world needs is innovation. Solve the problems ([1] [2]) with current feed readers. Come up with new ways to read news. Help people find what they need.
There's already integrated feed readers in Firefox, My Yahoo, and My MSN. Ask Jeeves now has Bloglines. CNet will have Newsburst. Oh, wait, there's more, hundreds of other RSS readers, all fairly similar. There's even several newspapers trying to offer their own feed readers.
The products are largely indistinguishable. Launching undifferentiated products into a crowded space is futile. With giants like My MSN, My Yahoo, and Firefox already major players, the world doesn't need yet another feed reader.
What the world needs is innovation. Solve the problems ([1] [2]) with current feed readers. Come up with new ways to read news. Help people find what they need.
Subscribe to:
Posts (Atom)
