Elliott Ng

Via Kara Swisher and TechCrunch I learned about the SearchMe private beta announce. I just signed up for it and look forward to seeing it. Kara broke the story but 5 days ago RichardbaxterSEO at YouMOZ noticed SearchMe crawler traffic and connected the dots that SearchMe earlier created the Wikiseek engine. Louis Grey also noted the crawler more than 30 days ago!

Caveat: I have not tried the product. But here’s my initial take:

1. SearchMe looks great for browsing and general navigational searches when you are just starting a search or research process.
2. For more targeted searches, I am not sure how SearchMe’s Visual Search–the iPod Touch like “cover flow”–can help me efficiently find what I’m looking for.
3. SearchMe should focus on discovery and browsing…sort of like StumbleUpon 2.0 powered by search.

1. SearchMe looks great for browsing and general navigational searches when you are just starting a search or research process.

One way they claim to help people is through CategorySuggest, the disambiguation of search results using categories. For example, if you type Labrador then they will show you visual icons for “pets” and “Canada” so you can better clarify your intent. This is great when you are typing in general terms and need the engine to help you express your intent more clearly.

But this is not unique. For example, Clusty.com (by Vivisimo ) has been doing automated clustering forever:

Ask (and Teoma before that) also does the same thing.

Yahoo! Search Assistant combines disambiguation with helping you refine your query:

Google does not separately do disambiguation but simply interweaves the results in their SERP page. But Google understands that Google doesn’t yet understand your intent. For example, they provide both the “Canada province” meaning and the “Dog” meaning in their first Wikipedia result:

2. For more targeted searches, I am not sure how SearchMe’s Visual Search–the iPod Touch like “cover flow”–can help me efficiently find what I’m looking for.

Caveat. I have not yet seen the product . But I simply do not believe that for more targeted searches, showing a thumbnail of the page is the best way for people to find what they are looking for.

The job of search engines is to understand intent, and to show the most relevant fragments of information from destination pages that help users select the best result for them. For example:

Live decided that for the query “digital cameras” the most important dimensions were photos, star rating, names of top products, and quick links to other guides and reviews.

I’m working at Kango.com, a new semantic search application focused on travel planning. We also have to decide what infomation to show to help people determine which website and product to research further:

Here we’ve decided that relevance to their search criteria (in this case “family friendly”), reviews and ratings, photo, price, and location (there is a map over to the right) are the factors that people want to know to decide if they want to do further research on a hotel. (Sorry about the debug “confidence interval” stuff on the right. That will be removed in the final product)

In these two examples, what if Live and Kango were constrained to only show the Visual Search “cover flow” experience that SearchMe is providing? To be fair, SearchMe provides a ListView feature that allows you to go back into a traditional search results page experience.

3. SearchMe should focus on discovery and browsing…sort of like StumbleUpon 2.0 powered by search.

In summary, SearchMe needs to decide what kinds of searches to focus on–general vs. specific, discovery/inspiration vs. specialized, and target verticals–where the Visual Search feature allows the user to get a rich, browse experience. In my opinion, SearchMe should become StumbleUpon 2.0, focused on discovery and browse. It should not compete with the host of vertical search engines (like ours at Kango) and even Google, Yahoo!, Live, and Ask, who are effectively blending the right page elements at the right time, to help people get more specific information.

Disclosure: I work at Kango, a vertical search application for travel planning utilizing semantic analysis. The thoughts on this blog are mine and don’t reflect that of my employer.

I attended SMX last week and greatly enjoyed Louis Monier’s keynote. Louis Monier is a founder of AltaVista, was an Ebay Fellow, joined Google in 2005, and then joined the stealthy “Google killer” Cuill in August 2007.

The Keynote was well received, “simply brilliant” according to Eric Enge at Stone Temple, and “interesting enough to make me close and shutdown my laptop and stop trying to do other stuff, and pay attention” according to Michael Grey at Grey Wolf. I enjoyed the simple, powerful presentation style and the vision of the “research assistant. If you did not attend and only want to read one post about it, read SearchEngineRoundtable’s post about the keynote.

Monier offered no details at all about Cuill. But he claimed that search “still feels like its 1995” and  spread FUD issued healthy criticism about other company’s approaches toward search:

1. Human-powered search: This is basically Yahoo! Directory, Wikipedia, and Mahalo. Directories are great but are very small and not scalable. Message: good luck to ya Mahalo!
2. Personalization: Limited upside. People’s searches are highly diverse, as he illustrated by his own query history. Previous searches are not helpful to predict user intent on future searches. Message: don’t try this at home, Yahoo!
3. Social search: Theory is that one’s friends already have the answer. Monier challenges this idea: how will friends’ search history help refine search results for you? Message: Where are you going next, Eurekster?
4. Vertical search: Vertical engines can provide great results for a specialized niche. Example he provided was BlueNile vs. Ebay. But there are 10,000 vertical search engines and who wants to remember all of them? Do we need a search engine for vertical search engines? :)  Message:  Alt Search Engines, never knew (any of) you.
5. Natural language processing: How much good language is there on the Web? Parsing documents is extremely difficult to do when the language is bad. Furthermore, no one wants to type long queries in the search box, so there is limited semantic meaning to parse in the query itself.  Message:  Powerset, forget about it.
6. Semantic Web: The vision of the Semantic Web is for Webmasters to express semantic meaning of documents through standards.  Message: Twine, keep trying to change human nature (NB: actually I don’t really know what Twine is).

Here are my thoughts on these areas:

1. Human-powered search:  Disagree.  I think there is a model here that is very complementary with the large, horizontal search engines.  Models like Wikipedia and even Mahalo are great at create unique, original human generated content.  These systems are great at creating content worthy of linking to and getting indexed.  They are complementary to the generalized search engines.  However, they are not true search applications in Monier’s thinking, but part of an existing search ecosystem that is currently dominated by Google.  I think there is potential for 1000s of specialized Mahalos to exist alongside with Google and Yahoo!
2. Personalization:  Partially agree.  The more context there is around the search, the more personalization can pay off.  Within the context of a specific vertical, I believe there are ways to hold onto user context that can then be applied toward future search queries.  I agree that for generalized search engines, this is not low hanging fruit.
3. Social search:  Disagree. Monier seriously discounts the role of “discovery applications” like Delicious, StumbleUpon, Digg, and even Google Reader, Facebook and Twitter.  These applications may not be defined as “search” but are part of larger discovery process that can be complementary to traditional search.
4. Vertical search:  Disagree. I guess because I work at a startup that is taking a verticalized approach in travel, I take issue with Monier’s critique.  I saw lots of examples of Google’s universal search and Microsoft Live’s search delivering specialized, blended results based on the industry vertical of the search query.  The question is: Why would you go to a specialized store when you can go the supermarket?  But in fact the world of both ecommerce and brick and mortar commerce demonstrates that not everyone goes to Walmart for everything.  And as long as there is discovery, these vertical search engines are a click away.  The real question is: will Google index the content of these Vertical Search engines and allow them to deliver better results on specific verticals?  It seems that Yahoo! is taking a new Open Search approach to leverage the content, knowledge, ontology, and expertise of sites.  I am optimistic that a swarm of vertical search engines in partnership with Yahoo! and ultimately Google (if Yahoo! shows success) can improve search for users.
5. Natural language:  Partially agree.  I agree with Monier’s critique in general.  But I feel that natural language search can be used where the user content is more structured (like travel reviews) and can be fit into an ontology that is specific to one vertical.  I am less optimistic about applying natural language more broadly where there is much less real sense of the context.  Specific language, in a specific vertical, with a specific ontology, can enable us to take some baby-steps with natural language processing of documents to provide more ways to understand what users think and better match to searcher’s intent.  I totally agree that no one wants to type stuff into a search box.
6. Semantic Web:  Partially agree. Ahh, the dream of Web 3.0.  I agree with Monier’s assessment that there is a problem of incentives.  But that doesn’t mean that a semantic layer cannot be extracted from documents by a 3rd party with an incentive to create that layer.  That is what we are doing at Kango in the travel space.  But I haven’t figured out why I should share out all that hard work as a service back out to the cloud…

There are certainly search luminaries like Monier at the company and I’m sure they know what they are doing.  Now that we know what Cuill is not, I’m curious to see what they are.  But search and discovery is extremely diverse and Monier’s assessment may not be 100% correct for all situations.  I agree with Monier that Search is just at the beginning and look forward to making my small contribution to progress in this space!