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

---

Comments

[ashkan karbasfrooshan]

not surprisingly, a fantastic overview.

good stuff.

ash

[Bert Armijo]

Nice overview. I'll have to go have a look.

As a new crop of search engines emerges, I'm curious to see if anyone provides some of the things on my wish list, like sorting results. They all return a carefully ordered list of results, but I'd like to be able to sort the top N by domain, age of index, etc. In other words, don't treat search as an end product, but as an intermediary result that I can then manipulate.

[Andy]

You should take a look at www.linguisticagents.com. It’s a start-up company that has developed a natural language understanding technology that will be used in many applications in addition to search. Let me know what you think -

[Lisa H]

InverSearch is a natural language search engine. Users send inquiries to businesses and receive relevant results in the form of responses. It’s as easy as texting or sending an email.

Business leads are received without advertising or bidding on keywords. Businesses get to view sales leads before choosing to respond, and pay only if they do, making the process immune to click fraud.

[saas application development]

very nice informative post... i got lot of knowledgeable stuff here.. thanks...