The Software Abstractions Blog

Powerset is Not a Google-killer!

This was one of the many memorable statements from Steve Newcomb, co-founder and COO of Powerset, at a blogger/tech-media bash at the company's San Francisco headquarters tonight, where they publicly unveiled Powerset's capabilities, technology and people for the first time. Powerset, a heavily-anticipated Semantic search engine (see the earlier post for the divergence between traffic and buzz ), has been gathering a lot of press lately as a potential "Google-killer".

Of course, the quote above is taken slightly out-of-context (but it makes a great headline!). According to Newcomb, Powerset owes a debt of gratitude to Google, as do all new Search Engine companies: after all, Google brought tremendous value to Search. It was Google that put Search on the map, it was Google that convinced ordinary users that simply by typing in a query string, they should be able to see and access any information on the web within a page or two. So, explained Newcomb, no one at Powerset hates Google - although, of course, they will happily try to take market share away from Google by offering a better search.

Key Strengths

That they have a better search, at least for a large subset of queries, is something I'm convinced of,  after tonight's demo. I think Powerset has three key strengths going for it:

1. Innovative Technology
My preconceived notions of Semantic Language Search revolved around providing additional context by applying lexical and semantic algorithms to the query string. Powerset's search technology, it turns out, does far more - it creates a fundamentally richer representation of the content in its index, rather than the proximity-based strings of tokens used by conventional search engines. By creating an enhanced semantic understanding of both the query and the content, Powerset's search engine can provide much better matches and therefore relevance for search results.

This natural-language parsing technology was originally created at Xerox-PARC. Its algorithms can extract the meaning from a given page, although in the absence of any contextual information in the query, the search falls back to keyword-matching algorithms (similar to Google). The combination of natural-language processing of the query and semantic knowledge of the content, means that Powerset can return fundamentally different results for the following two queries - (a) What companies did Peoplesoft acquire, and (b) Who acquired Peoplesoft? - results that accurately reflect the different intents of these two questions. Conventional keyword-based search engines like Google and Yahoo! would be hard-pressed to provide different, accurate results for these two queries. Similarly, flipping a query, or replacing specific words with synonyms, does not affect the results found in Powerset.

The Powerset demo included many of these rich queries, where the relevance of the results is based on getting a semantic understanding of the queries. Some examples are given below:

• What politicians were killed by disease?
• Who mocked Blair?  [This produced results that included the words lampoon, caricature, satirical and impersonators - very cool!]
• What did Jesus say? [Again the results were quite meaningful; e.g. nothing about what was said to Jesus]

Of course, the catch here is that all this semantic mapping gets quite expensive when building the index. According to Newcomb, Powerset currently depends on Amazon EC2 as well as having their own Data Center.

2. Social Skills
No, I'm not referring to Emily Post. It was clear that Powerset fully understands the value of leveraging social networks and tapping into the Web 2.0 ecosystem. In addition to creating a destination site for Internet Search, Powerset also plans to focus on the underlying platform, eventually making it available for mashups, widgets and APIs.

As an example, Scott Prevost, Director of Products, showed us a demo of a mashup called The Entertainator. According to Prevost, this was an easy integration that was completed in an afternoon, to highlight the use of Structured Data to improve the results of natural language queries. I found this simple demo, based on data pulled from Freebase, to be very impressive! It correctly answered questions such as the following, in the latter two cases coming up with a single, accurate answer:

• What movies did Al Pacino star in?
• Who directed The Matrix?

and finally

• Who said "You can be my wingman"

This type of single-response search is very valuable for Mobile Search, Newcomb pointed out, where precision is critical for search to be effective.

3. An innovative Product Launch model/community
At today's event, Powerset unveiled a community product called PowerLabs - billed as a combination of Facebook, Digg and Google Labs. The idea is to encourage participation from a community of technologists, bloggers, tech journalists and search enthusiasts, utilizing the combined wisdom of the crowd to provide feedback, generate new ideas, suggest features and implement mashups. A higher level of participation enables community members to earn badges, higher levels of rank (a la World of Warcraft) and credit for features that actually get implemented. [There was no mention of cash prizes of any sort.] Participants can vote ideas up or down in Digg-like fashion. At a high level, this approach is similar to Dell's IdeaStorm. Used in this way, though, it's an innovative approach for controlling (and deflating) some of the hype surrounding the much-anticipated availability of Powerset's search engine, and allows Powerset to gauge market reaction and gain user acceptance prior to launch. Mark Johnson, Product Manager of Powerlabs, showed us a demo of some of the cool community features of this application.

Missing Pieces

Interestingly, it does not appear that Powerset is paying attention to two key areas of search improvement for the future: Personalization and Social Input (aka Wisdom of Crowds). For the latter, the Powerlabs community can provide feedback on the relevance of search results for particular queries, but that is a far cry from implicitly gathering user behavior information on a continuous basis to improve search results - examples of search engines that provide this type of functionality are: BayNote, Collarity and Loomia. Personalization, especially, would be useful input for Powerset searches, by implicitly providing user-specific context to minimal queries where there is little or no semantic information available.

Happy Employees

Most importantly, Powerset has a highly successful buzz going, not only in the blogosphere, but also within its own ranks. I spoke to several employees, both developers and managers, and each one was actively excited to be there and to be contributing to a solution they see as game-changing in Search - there was a palpable sense of energy and excitement in the place. This kind of excitement can provide a "positive override" that enables a startup to navigate the inevitable shoals and reefs which would sink a lesser company (although, if not carefully managed, it can also cause dashed hopes if expectations are too high).

Conclusion

One thing is for certain: a serious challenge from a well-funded startup offering meaningful semantic search, is bound to force Google to improve - this can only be good for search users in general.

Overall, Mr. Newcomb, I must respectfully disagree with you - Powerset is a potential Google-killer, or at the very least, I predict that Powerset will give Google a run for its money!

Further Reading

For additional perspectives on tonight's event, check out the following:

• Dan Farber's post on ZDNet Blogs:  Powerset: The natural language search mashup platform
• Pinhead's Progress:  Powerset and Powerlabs, roundup
• Kevin Burton's NEW FeedBlog:  Powerset - Out of Stealth
• Matt Labbe's whirn:  Powerset Opens Up
• Ashkan Karbasfrooshan on HipMojo:  Right Now, PoweRset Stands For PR

---

Dispatches: from the Searchnomics Conference and a Powerset evening!

Two major Search events, back to back!

On Wednesday, I attended the Searchnomics 2007 Conference at Santa Clara,CA. I expect to be blogging about those sessions for a little while, but two items in particular stand out in my mind: the closing keynote address by Marissa Mayer, VP of Search Products and User Experience at Google (which you can read about in detail in my Read/WriteWeb post ) and a highly entertaining session on Advanced Web Analytics by uber-evangelist Avinash Kaushik. [Another session on Web Analytics by independent consultant Eric Peterson, was alo very illuminating!]

Thursday's event was a blogger/tech-media bash at Powerset's headquarters in San Francisco. Powerset, a heavily-anticipated Semantic search engine, has been gathering a lot of press lately as a potential "Google-killer". Tonight was the first public demonstration of Powerset's capabilities, technology and people. A more detailed post follows soon ...

---

Looking for Rich, Famous and Popular - Search engines, that is!

As part of the research for an upcoming blog post, I've been looking for the top general-purpose search engines that combine innovation, buzz and high levels of traffic. As a starting point, I've taken an initial list and collected information about those engines, in terms of two metrics:

• Traffic - Alexa, Compete and Quantcast data, and
• Buzz - the number of del.icio.us results and number of blog hits in Google BlogSearch in the last month.

The whole thing is then compiled into two normalized ranks, based on the formulae given below. (The list excludes the top-five mainstream search engines: Google, Yahoo, Microsoft LiveSearch, Ask.com and AOL Search.) The results are interesting, to say the least!

[If you represent, or know of, an innovative search engine that has higher levels of traffic and/or buzz than those on this list, and would like it to be included in my upcoming article, please drop me an email at nitink (at) softwareabstractions.com with the subject: Search Innovation.]

Calculation Formulae

The Formulae used for calculating normalized ranks are as follows:

Chart

The Bar Chart of the normalized ranks is shown below - click on the image to view the bigger version :
(Note: The Actual Data Chart used is displayed at the end of this article)

Caveats

Pseudo-science: First of all, note that the diagram above is meant to be an indicator, nothing more (I imagine that the numbers are far from exact, although the overall trends are probably accurate)

Data issues: The grayed-out numbers in the data table below represent anomalies

• Google numbers are provided purely as a reference
• Powerset numbers are simply user visits to the general web site, since their search engine is not actually operational yet
• Actual blog hit numbers for Gigablast and Mahalo are outriders (very high); I suspect that there is some other reason for their use, such as Mahalo in its Hawaiian sense, so I've instead used corresponding data for the terms "Gigablast search" and "Mahalo search"

Observations

• In general, it appears that the older the site, the higher the traffic, regardless of buzz or innovation. This is not as clear from the picture above, but my larger data set also includes traffic numbers for the former portals - Excite, Lycos, AltaVista, and others, and their traffic numbers are high indeed (according to Alexa/QC/C)
• Sites such as Technorati, Topix.com and Del.icio.us, that combine Content+Features+Search have significantly higher numbers than the pure search engines - which is unsurprising
• Numbers from Quantcast and Compete seem to be fairly close in most cases, which is a good sign.
• Interestingly, traffic and buzz ranks are not always correlated, and sometimes diverge quite significantly [hint: see Powerset numbers]

Data Table

The data table used for the above graph is given below  - click on the image to view the bigger version . This is a snapshot of data obtained on 6/20/2007 from the various free web tools/services.

Data Table:

---

A conversation with Guy Kawasaki of Garage Technology Ventures

There is no need to introduce Guy Kawasaki - Evangelist, VC, Blogger and long-time Silicon Valley icon. His latest venture is called Truemors (here's the Software Abstractions review ) - a digg-like social network for headlines and short posts. Recently, I had the opportunity to get his thoughts via email on a variety of topics - our (electronic) conversation is reproduced below.

 

Q - Congratulations on your new startup, Truemors, which seems to be doing very well. What attracted you to this particular idea, out of the hundreds of potential candidate ideas out there?

GK - I was inspired by three events: Listening to James Hong explain how he created HotorNot; SpinVox enabling people to post blog entries via voicemail; and the success of Twitter. I love the concept of the democratization of information, so when I put all four together, I came up with Truemors.

Q - What is your business model for Truemors ?

GK - Hopefully we'll attract enough page views to sell advertising and sponsorships in significant amounts.

Q - How will Truemors differentiate itself from established social headline sites like Digg, Reddit et al?

GK - Digg, Reddit, et al are mechanisms for people to rate content. Truemors is a way to generate content. In an ideal world, people will start Digging our content and drive up our page views.

Q - What do you make of the mostly negative opinions of the blogosphere about Truemors?

GK - First, don't exaggerate. The feedback hasn't been "mostly" negative. It's been totally, unequivocally negative. :-) One of the consequences of the democratization of information is that you have to be able to take the Heat. So my reaction is, We shall see. Maybe they are right. Maybe they are wrong. But you never know unless you try.

Q - You've been an active blogger for a while now. What has your blogging experience been like?

GK - Very, very gratifying. I like working without having to get approval and waiting for publication. "I blog, therefore, I am," is my new mantra. I'm #15 or 16 right on Technorati now. My stated goal when I started blogging was to be in the top ten, so I've been fortunate.

Q - What type of key differentiators do you look for in your investments through Garage.com (now Garage Technology Ventures)?

GK - We're looking for two guys/gals on the West Coast who are trying to change the world with a software, IT, or clean-tech startup. We're a small fund so we don't participate in companies that need $25 million to break even or build a factory.

Q - What are the two or three most influential business books you've read?

GK - If You Want to Write by Brenda Ueland, Influence by Robert Cialdini, and Crossing the Chasm by Geoffrey Moore.

Q - Who are your business heroes?

GK - James Hong of HotorNot and Markus Frind of PlentyofFish. I love how they turned nothing into a big something.

Q - It always amazes me that Apple users - the so-called "Mac faithful" - have displayed such fierce loyalty to Apple through all of the travails and even mis-steps of the company; now the wheel has turned full circle, and Apple is back on top.
What do you think is the "secret sauce" that inspires such passion and loyalty in Apple users?

GK - There's no secret sauce. It's purely engineering. When Apple makes stuff that people like, it does well. When it doesn't, it doesn't.

Q - Do you think the iPhone will    succeed wildly   or    fail miserably ?

GK - It's too early to tell. Basically everyone is conjecturing how good or bad it will be without having used it--unless you're Walt Mossberg and Apple is sucking up to you.

Certainly the first million or so will fly out of the stores, but so did the first 250,000 Macs. Then the hard work really begins: battery life, no keyboard, durability of the screen, ATT's sucky data network, size and weight, lack of Exchange server (I can't believe I'm saying this too!), price point (a mere 20-30x more than a Razr), cancellation fee of your current plan (unless your two-year anniversary happens to be June 29th).

Still, when all is said and done, it's very dumb to bet against Steve.

Q - What message would you give to would-be entrepreneurs out there?

GK - Asking bloggers about anything beyond tools for bloggers is dubious as a feedback mechanism. Whatever the blogosphere says, do the opposite. [Ed: Aw, gee, thanks for the compliment!   Not! ;-)]

Q - In one sentence, what's your core philosophy?

GK - "Empower entrepreneurs."

---

The Hidden Dangers of Special Casing Google Search Results

One of the key tests of the maturity of any software development organization, is how it deals with corner cases. Any software application, no matter how thoroughly thought-out and well-designed, tends to exhibit corner cases: relatively uncommon use cases or scenarios (often in a specific sequence) that lead to unexpected results.

A Question of Approach

One of the lessons that has deeply impressed me, over many years as a software architect and designer, is that the approach used to handle these corner cases determines the long-term stability and reliability of any software application.

The right approach to handling the unexpected results for these uncommon use cases (essentially the "long tail" of usage scenarios) is to look at the general case to see how each one fits in; then change the design appropriately to handle the outliers correctly. This algorithm can be expressed as follows:

1. Evaluate the unexpected result for a specific uncommon scenario
2. Generalize the problem - how does this use case fit into the big picture?
3. Change some part of the overall design to handle the issue
4. Rinse and repeat

Of course, changing the overall design is easier said than done. Sometimes, there are just too many specialized use cases to cover with a general design; the standard architectural approach to handle this problem is to create a general facility into which specific custom solutions can be integrated in a standard way.

The one thing you always try to stay away from, at all costs, is to tweak the code for that one special case, without evaluation of the general problem or consideration of the design. Yes, special-casing can be done and it will even work, for a short time; but do it often enough and pretty soon you have a mishmash of special cases interacting in unpredictable ways, the dreaded "spaghetti-code" syndrome.

I imagine that algorithmic solutions to the problem of Internet Search are similar; since there are many millions of possible search terms in the long tail, you want to approach the findability problem as generally as possible. Sure, there will be corner cases where the results are less than optimal; when that happens, the best approach is to solve for the more general case and modify the searching algorithms to try to improve the outcome even for the corner cases. The most risky option (albeit the one with the most instant gratification) is to special-case the results for that particular query - once you start down that path, it's a very slippery slope indeed. Apart from the obvious issues of lack of scalability of this approach and questionable ROI for the time invested, the real cost is long-term: with enough tweaking, the system rapidly becomes unstable until you can no longer be sure how the system will react in any general way.

Case Study

In this context, a recent article in the New York Times about Google's responses to unexpected query results (you can read it here via the 37 signals blog ) provide an interesting case study. It appears that Google uses both approaches: special-case tweaking, and algorithmic modifications - and frequently! As the article says:

... the search-quality team makes about a half-dozen major and minor changes a week to the vast nest of mathematical formulas that power the search engine.

After pointing out that the search engine has "many thousands of interlocking equations", the author gives us two specific examples, listed below.

Example 1:

Recently, a search for “French Revolution” returned too many sites about the recent French presidential election campaign — in which candidates opined on various policy revolutions — rather than the ouster of King Louis XVI. A search-engine tweak gave more weight to pages with phrases like “French Revolution” rather than pages that simply had both words.

Example 2:

In 2005, Bill Brougher, a Google product manager, complained that typing the phrase “teak patio Palo Alto” didn’t return a local store called the Teak Patio.

So Mr. Singhal fired up one of Google’s prized and closely guarded internal programs, called Debug, which shows how its computers evaluate each query and each Web page. He discovered that Theteakpatio.com did not show up because Google’s formulas were not giving enough importance to links from other sites about Palo Alto.

It was also a clue to a bigger problem. (Emphasis added.) Finding local businesses is important to users, but Google often has to rely on only a handful of sites for clues about which businesses are best. Within two months of Mr. Brougher’s complaint, Mr. Singhal’s group had written a new mathematical formula to handle queries for hometown shops.

These examples are ideal for illustrating our earlier discussion. The first one clearly indicates special-casing: yes, the "French Revolution" search works now; the change fixed that particular query, but how did it improve the search engine algorithms as a whole? The second example, on the other hand, is a textbook implementation of the general approach: tackling and solving the more general problem represented by the misbehaving query.

It was a suprise to me, looking at Example 1, that Google allows this type of special-casing for use cases so specific; it may be only a temporary solution. It's common knowledge that Google focuses on hiring PhDs and "bright people" in general (even Marc Andreessen says so ); perhaps they have figured out a way to avoid or overcome the long-term issues that normally accompany a large number of these one-off fixes - I sure hope so.

Greg Linden posted an incisive article on this topic a few days ago, which is a great read.

Conclusion

For my part, I believe that the best approach to this thorny problem of fixing corner use cases for search, is to utilize the power of social input (aka the "wisdom of crowds") to automatically improve results over time. The Unified Architecture for Search proposed in my earlier post, included the idea of combining additional inputs - both semantic and social - with the standard PageRank algorithm to continuously improve search results over time.

References

Other folks have also commented on the NYT piece:

• Matt Cutts gives his analysis in Five Things you didn't know about Google's search
• Philipp Lenssen has a concise summary for the time-constrained
• Matthew Hurst of the Data Mining blog questions whether this is really Innovation
• Donna Bogatin looks at the impact on Search Marketers

A Unifying Framework for Web 2.0

What, exactly, is Web 2.0? How does it apply within the Enterprise and to the web at large? What technologies are involved and what are the benefits?

Ross Dawson, of Future Exploration Network tackles these questions in his recently launched Web 2.0 Framework, an excellent work which describes the many aspects of Web 2.0, both within the Enterprise and on the web at large. This unifying framework is remarkable because it brings together the definitions, characteristics, mechanisms, major players and outcomes of Web 2.0, in a simple and intuitive way.

Ross describes the objective as follows:

The intention of the Web 2.0 Framework is to provide a clear, concise view of the nature of Web 2.0, particularly for senior executives or other non-technical people who are trying to grasp the scope of Web 2.0, and the implications and opportunities for their organizations.

I see this as a critical area: in order to effectively sell the ideas of Enterprise 2.0 to the CxO, we have to be able to explain the mechanisms (the "how") in simple terms and focus on the expected outcomes in terms of helping the organization meet its goals ("the why"). Web 2.0 is made up of many different parts. Like the story of the Blind Men and the Elephant, when we focus on a variety of specific perspectives, we experience it in different ways; the framework does a great job of taking these different aspects and putting it all together.

 

 

One area ripe for discussion, where my thinking diverges somewhat from Ross's framework, is in the Emergent Outcomes section. I see three core outcomes from an implementation of Web 2.0 technologies:

• Personalization
• Collective Intelligence
• Semantic appreciation of content

In many ways, I see the other items - finding relevant content, enhanced usability, visibility of interesting content and even community - as inherent in, or following from, the three core outcomes described above.

In any case, this framework certainly moves things forward in the collective wisdom about Web 2.0; I look forward to continuing this discussion further.

---