Category: Uncategorized

Interviewing  blogger Jeff Jarvis about his recently published book, “What Would Google Do?“, Nick Summers asks:

Are there any areas in which Google itself doesn’t act very “Google-y?” Not disclosing its advertising revenue splits, for example.

Jarvis answers:

Right. There are areas where Google doesn’t act very Google-y, which are mainly about transparency. It can’t be transparent about its algorithms and how they operate, because then they will get gamed more. And those are special sauce. I wish Google were more open about its advertising arrangements and splits, so we had a better sense of the value of the market; I wish it were more open about the sources that it puts into Google News.

I’m glad to see Jarvis recognizing this lack of transparency, which I’ve occasionally railed about on his blog. Read the rest of the interview at Newsweek.

Some readers have noticed that I often take shots at Google on this blog, but seem to give Microsoft a pass. I assure you that I am not a Microsoft fan boy–in fact, Microsoft’s enterprise search subsidiary, FAST, competes with Endeca more than Google does. But today I’ll prove that I’m an equal-opportunity critic by talking about Microsoft Songsmith.

The idea is brilliant, at least in theory:

Just open up Songsmith, choose from one of thirty different musical styles, and press record. Sing whatever you like – a birthday song for Mom, a love song for that special someone (they’ll be impressed that you wrote a song for them!), or maybe just try playing with your favorite pop songs. As soon as you press “stop”, Songsmith will generate musical accompaniment to match your voice, and play back your song for you. It’s that simple.

What do the critics say? Here’s what Randall Stross writes in the New York Times

How satisfying are the musical results? Microsoft lets you hear for yourself in a promotional video titled “Everyone Has a Song Inside.” The video is getting more attention than the software because it’s awful, in unintentional ways. “Notes on ‘Camp’, ” the 1964 essay by Susan Sontag, identifies a category of art that isn’t campy, just “bad to the point of being laughable, but not bad to the point of being enjoyable.” The Songsmith video is exactly that.

But I have to wonder if the researchers and product developers at Microsoft imagined what would happen when they released Songsmith into the wild. People have been taking actual vocal tracks from pop songs and feeding them into Songsmith. The results are–well, I’ll let you judge for yourself from this rendition of “Sgt. Pepper’s Lonely Hearts Club Band.”

Those of you who know me offline or online hopefully know that I have a very critical eye, especially when it comes to technical innovations. But every now and then I run into colleagues whose research projects that scream for productization.

I’ve been talking to such a colleague and am eager to hook him up with the right angel or VC investors. He’s in the New York area, if geography matters to you, and the IR / NLP technology he has developed is of the sort that would generally appeal to readers here, but is in no way competitive with my employer’s.

If you are an investor looking for opportunities and are interested in learning more, please contact me directly. Please don’t contact me just because you are curious. I’m not trying to tease anyone, and I promise that, if something comes of this, everyone here will find about about it in due course.

Also, I realize I’m not giving much information away, which is not my style. I recognize that this stinginess may put off potential investors. I’m sorry, but it’s a conscious trade-off. I have a deep respect for my colleague’s privacy, and I also feel that the trust I am asking for acts as a useful filter: if you don’t trust my taste in ready-for-prime-time technology, then there’s no reason for either of us to waste each other’s time.

In any case, I’m not asking for money, just evidence that you are an investor (and no, not evidence in the Nigerian Letter sense). Also, I don’t have any direct stake in the outcome; I’m just trying to help a colleague. Although it’s the sort of venture in which I’d happily serve as an advisor.

You can contact me at dtunkelang at gmail dot com.

Wow, this is quite a week for news in the enterprise search industry! Yesterday, I woke up to hear that Autonomy is acquiring Interwoven; today, I hear from CMS Watch analyst Adriaan Bloem that FAST CEO  John Lervik is resigning, with their CTO Bjørn Olstad stepping in to fill his duties.

FAST had some accounting issues that came out shortly after Microsoft acquired them last year, and it’s possible that Microsoft is simply clearing the deck of anyone associated with those issues. I’m just an innocent bystander, and my pedestrian concern is whether Bjørn’s new duties will prevent him from participating in a panel I’m organizing at SIGIR.

I am proud to report that the “Reconsidering Relevance” presentation has been enjoying  a warm reception:

• SlideShare: made top (most viewed) presentation on the day it was posted (January 9th) and was featured today (January 21st) by the SlideShare editorial team. Viewed over 1,000 times.
• Interviewed by Cloud of Data blogger Paul Miller about the presentation, and more generally about Endeca’s approach to enterprise search. Click  here to listen to the podcast.
• Presentation re-posted on Oracle’s Enterprise 2.0 and Content Management blog.

I am grateful for the attention this presentation has received, and I hope that the attention helps further the HCIR vision that the presentation advocates. I also promise that the YouTube video is forthcoming. Google is very apologetic about the delay, but they assure me that the upload is in process.

“A man’s got to know his limitations,” as Clint Eastwood famously said as Dirty Harry in Magnum Force. Well, as Panos Ipeirotis reports on his blog, Turkers (the people who are paid to perform tasks using Amazon’s Mechanical Turk service), actually do know their limitations.

Read the full post for details, but here is the punch line:

Turkers can self-report accurately the difficulty of labeling an example correctly! Since example difficulty and labeling quality are strongly interconnected, this also means that they are good at estimating their own quality!

It’s great to see Mechanical Turk used in the service of productive research, and not just as a way to shill on the cheap.

Just a heads up that Danah Boyd has published her PhD dissertaion entitled “Taken Out of Context: American Teen Sociality in Networked Publics“. Danah is a rock star in the social networking research community; you might have noitced that I cite her Master’s Thesis from time to time. I’m looking forward to reading her latest work, and to welcoming her to the Boston area, where she’ll be joining Microsoft Research New England.

Today was a harsh news day for Google, with TechCrunch posting a leaked thread of internal emails on why Google employees quit. The emails are intriguing if highly redundant; the schadenfreude comments are merely predictable.

But the more interesting piece of news is that Google is moving beyond its initial SearchWiki efforts to offer more meaningful personalization to users. Their new feature is called Google Preferred Sites. According to the unofficial Google Operating System Blog:

Preferred Sites is a new experimental feature for Google Search that lets you personalize the results by adding a list of sites you want to appear more often when you search. Based on your search history, Google suggests some frequently-visited sites, but you can add any other site.

As regular readers know, I have little love for SearchWiki. But Preferred Sites seems to be a real step, albeit a small one, towards allowing users to meaningfully–and transparently–personalize their search experience. I say “seems” because I haven’t had a chance to try it out. Perhaps someone with a lucky cookie who’s gotten to try it can comment on his or her experience.

Jeff Dalton recently updated articles he maintains on open-source NLP and machine learning tools and open-source search engines. Check it out, as well as the discussion about the pros and cons of open source enterprise search on LinkedIn.

Everyone has their price, but who knew it was so low? First, we see Burger King persuading people to trade 10 Facebook friends for a Whopper (suggested retal price: $3.69). Then some are suggesting that Twitter might create a business model offer companies a sort of pay-per-click (PPC) approach to friendship where they might pay $1 for each “friend” who follows a sponsored invitation.

But apparently Belkin may have read Ben Kunz’s “Modest Blogging Proposal” and not recognized it as satire. According to The Daily Background, a Belkin employee used Amazon’s Mechanical Turk service to pay people to write positive reviews of Belkin’s products–65 cents for each review. The scandal has received wide coverage through a post by John Biggs on CrunchGear.

I can’t say I’m shocked (shocked!) to find out that there’s payola going on here. And, by way of an “I told you so,” a big part of the problem is that reviews are anonymous, and anonymity doesn’t play well with information accountability.

But I am disappointed that people’s integrity is so cheap. Even Esau was able to swing a mess of pottage, which by my best guess would go for $5 in 2009 dollars.