The new features might also reveal some secrets of Google Maps ranking factors. It’s definitely a space that’s well worth watching for local search marketing experts.

Some business directories and data aggregators contain huge percentages of bad listings including gone-out-of-business companies and duplicated listings. Pankaj Mathur of InfoUSA recently estimated that some sources such as Dunn & Bradstreet, Acxiom, Localeze and Google Maps might contain 17-to-18-million records, or up to 4 million more records than actual, live businesses! That’s as much as a 28.6% error rate!

I know from experience that some percentage of a business directory’s bloat can be caused by inability to figure out if a business has closed up for good or not (most data sources are not set up to verify each individual business, although InfoUSA remains an example of the gold standard, relying on a practice of phoning each and every business to verify it’s viability at least once per year, and also using a number of other signals of possible closure). These old, dead business listings clog up a great many online directories and can result in poor usability when consumers attempt to drive to their doorsteps for products or services.

However, another source of bloat is in the form of duplicate listings, and as a search engine marketer, it’s these duplicate listings that my clients sometimes have which concern me highly. Just as with duplicate webpages in regular SEO, duplicate listings in business directories and within Google Maps can potentially dilute down a company’s possible ranking score, resulting in lower overall rankings and poorer online performance.

Back in February, Google and other search engines jointly announced support of a canonical tag for webpages, enabling webmasters to specify which page URL should be treated as the main/authoritative one for search engine indexing in cases where multiple URL variations could occur.

If such a protocol makes sense for the general web search engines, why not a protocol to assist in reducing dupes in local search, too?

Google Maps help suggests that users can help them identify cases of duplicate listings, but the problem is that dupes may be constantly coming in through all of Google’s various data partners.

At the recent Kelsey Conference, local directory industry experts predict that Cost-Per-Call is increasingly going to become a dominant pricing model and this will mainly happen through individual directories displaying different tracking phone numbers for the same businesses – and all these different channels with separate phones feed into Google Maps, Yahoo! Local, Bing Maps, and other local search engines. So, potential issues from many various phone numbers could be increasing.

Tracking phone numbers are not the only potential problem – sometimes a company’s alternate phone numbers and fax numbers will get parsed off into separate business listings, causing further instances of duplication.

So, how could this “Canonical Phone Tag” effectively be accomplished? hCard Microformat already provides a framework for doing something like this. A canonical phone tag could be formed like this:

<abbr class=”tel” title=”000.867.5309″>000.123.4567</abbr>

In this example, webpage users could see the “000.123.4567″ tracking phone number when they view the webpage, and the machines could instead glean the authoritative, canonical phone number for the business, “000.867.5309″, and ignore the tracking number.

So, is this a solution in search of a problem?

How many of you have encountered instances where a tracking phone number has gotten indexed in a separate listing alongside a business’s primary phone number/listing? If so, this solution may be worthwhile to consider in keeping all of a business’s ranking weight combined in one listing as opposed to distributed across many.

Here’s Quova’s description of the innovations:

“Quova’s newly added patent describes a method for determining the geographic location of an Internet user based upon combining trace routes, user registration information, host names with textual patterns that reveal geolocation information and Internet Service Provider (ISP) service area information. These trace routes describe the pathways by which data moves through the Internet. Each node or ‘hop’ in the trace route is identified by an IP address. These interconnected nodes can be used to recreate the topology of the Internet. Each geolocation can then be assigned to these IP addresses in order to determine the location of each node, up to and including the end user’s IP address and the geolocation of that end user.”

I previously have written about Quova in my extensive article, “Geolocation: Core To The Local Space & Key to Click-Fraud Detection“. My earlier description of them reads practically as an endorsement – something I very rarely do at all. But, I think what I wrote is pretty accurate, overall. Quova is considered pretty much best-in-class of the companies providing geolocation data mapping, because of their greater variety of geo data sources, their more sophisticated mapping methods, and because they actually submit to a third-party audit for data accuracy.

As I’ve described previously, there are quite a number of providers out there which try to associate endusers’ IP addresses with geographic location coordinates, but I’ve always been a really huge skeptic of the overall error rates associated with this type of data. In the past there were estimates that geolocation error rates could be anywhere from 50% to 85%! For marketers attempting to precisely target ads and content to geographically cohesive groups, such high error rates were far too high to be acceptable from my P.O.V. five years ago.

For all geolocation data companies, error rates increase as the geotargeting level becomes more precisely granular. In other words, most of these companies likely have very high accuracy for country-level geolocation capability, lesser for regional targeting such as state or province level, less than that precision for city-level, and lowest accuracy for ZIP-code or postal-code level targeting.

I’ve also been highly skeptical in the past because there’s a low ability to actually test how accurate a given company’s data may be — most testing is based on sample sets which I’ve suspected may be highly skewed towards more accurate geolocations. Anecdotal stories of erroneous geotargeting have come up repeatedly throughout the industry, while there are few ways of calculating actual error, and geolocation data companies reiterate unverifiable claims of accuracy rates.

However, innovations such as found in this patent awarded to Quova help in improving the stature of the whole industry, and really help to reassure marketers and security administrators that the data is solid. I can critique testing methodologies for auditing the data, but I get less strident about it when I can see that the methods by which the data was arrived at are improving and are far less prone to error rates.

Since increasing amounts of advertising targeting are being based upon geotargeting, and since so much of our security screening of PPC ads’ click-through results are being based upon geolocation data as a major component of analysis, I think it behooves companies to use very high-quality data. Quova’s historically high price tags have caused some to look for cheap and easy alternatives, but people should be very careful about the methods used by those other companies. In many, many cases, the cheaper alternatives equate with far less sophisticated methodologies, which also equate with higher error rates. And, auditing undoubtedly also adds overhead to the pricetag, but using data that has zero third-party checks will open companies to greater liability.

Most of the cheap companies are using only ARIN network data which has a lower accuracy level than if it was further enhanced by data from major ISPs and network tracerouting.

Except for the simplest applications which do not involve security and fraud analysis, I think that Quova’s industry-leading patents in geolocation make it so that other geodata companies’ products cannot realistically compete in this arena.

Grauman’s Chinese Theatre in Street View, Los Angeles.

However, I see a much more immediately valuable use that could be put to text scanned from the images — quality improvement of address pinpointing functionality.

I’ve written before about a number of the causes of errors in maps, and one of the most frequent errors is in how addresses get incorrectly pinpointed down streets. If Google Street View images were scanned for text as Philipp describes, address numbers for businesses and residences could be much more accurately associated with locations.

Street View could be used to improve mapping quality in many areas.

Gur Kimchi, Principal Architect, Microsoft Virtual Earth

Gur provided a lot of demonstrations of existing and upcoming features from the Virtual Earth teams:

One of the interesting tidbits he mentioned was how Microsoft had acquired Caligari, a 3d modeling and animation company to expand defining realworld understanding within the virtual mapping apps. He provided a lot of examples of how Virtual Earth understands 3d projections from flat map data and can even plot streets beneath buildings in satellite/aerial photos. Unlike some top apps like Google Maps, Virtual Earth even corrects the Escher Effect in these pix.

According to him, the platform is capable of “indexing everything in the world”. He was referring to the real, physical, locative world, and mashing up other, virtual information with it.

He also stated that the key to monetization of it was to eventually create “deeply contextual advertising” which is highly relevant to the consumer — this is the key to ad effectiveness. “Almost mind-reading,” as he put it.

Regarding the mapping/display of sensitive security subjects, Gur stated that “…we never show critical infrastructure. The CIA tells us what to avoid (in America).” He also spends a lot of time meeting with government representatives from around the world in order to designate and avoid such sensitive areas.

The demonstrations were impressive and almost overwhelming. He showed a lot of content from Photosynth which we’ve seen before elsewhere, but it was still intriguing in the context of how it might be applied to search interfaces for mapping. Greg Sterling suggested that the audience might find it a struggle to understand how all the eye-candy could apply to search marketing, but I think it was still interesting to everyone to hear opinions from Gur, since he clearly is something of a visionary in the future of innovative mapping and user interfaces.

The Long Tail, if you don’t already know, refers to a theory promoted by a book by Chris Anderson titled “The Long Tail”, which describes a sort of niche strategy of business, such as employed by Amazon.com or Netflix, that sell a large number of unique items in relatively small quantities. The idea is that while you can obviously sell large numbers of a few popular items (the “head”), the cumulative, smaller number of sales of all your many less-popular items (the “tail”) might easily add up to a far greater total amount.

“Head” items shown in red, “Tail” items shown in blue

Here at Netconcepts, we’ve been promoting the Long Tail concept in relation to natural search marketing for quite some time, since we’ve witnessed how its application can directly improve a business’s overall sales numbers. Indeed, businesses often get the most sales per item for their most popular products, but those products are also often the most competed on the internet, and sometimes the hardest to promote as a result. Even in the cases of top online retailers, we’ve seen that greater bulks of traffic and associated sales may often come from the bulk of less-popular Tail products.

Elberse’s findings are based only on a limited study of music and home-video sales, so it seems quite possible that her study might be true only for the companies and industries from which her data originated.

At the end of the day, theory is all well and good, but actual sales figures are far more important. Netconcepts has working formulas, methodologies and software (GravityStream) which have worked time and again to help improve internet retail sites’ traffic and associated sales. The Long Tail approach has definite applications for Search Engine Optimization (“SEO”), and understanding the concept and making use of it in a search marketing plan effectively will result in increased profits.

For more information on how the Long Tail can positively influence online business, read our white paper, “Chasing the Long Tail of Natural Search“.

(See also mentions of this study by Greg Sterling and Rafe Colburn.)

Behavioral ad targeting is nothing new on the internet, and I easily recall it being offered in one form or another as far back as about 1999. In fact, 24/7 Real Media currently offers behavioral targeting through their ad network as just one case in point. So what’s new with this incarnation is the way in which NebuAd collects data to base the targeting upon. NebuAd’s innovative twist on behavior targeting is based upon monitoring individuals’ internet browsing habits through their ISP, essentially seeing all the sites and pages that a user visits.

Other online behavioral targeting models are likely based upon ad networks cookieing users on all the sites where the ad network runs ads. This may leave a lot to be desired, since users are likely visiting a great many sites outside of the ad network, leaving them a little in the dark as to what the users may be up to currently — if a user only visited one or two ad network sites, there could be insufficient context to really make any sort of behavioral assumptions. NebuAd’s data collection method may provide them with the ability to more accurately target ads for far more site visitors.

NebuAd has already partnered with a few ISPs like CenturyTel to get access to their subscribers’ surfing data. NebuAd uses a sort of network appliance hardware to sniff out the sites/pages that users are requesting through their ISP’s network operating centers.

The few weak links I see in NebuAd’s approach are (1) the sensitivity associated with online privacy, (2) their dependence upon persuading sufficient numbers of major ISPs to cooperate with them, and (3) users coming in from ISPs which are not data partners of NebuAd, making them harder to profile.

On the privacy issue, NebuAd claims they do not generate a database that could be leaked or subpoenaed, since they encrypt user-identifiable data like IP addresses in a one-way hash. It all sounds good, but as consumers we just have to trust them that this is all handled well, and it only takes one server to be misconfigured and IP data starts getting logged.

The subpoena bit is just PR-spin, too, since a government agency or lawyer can still hit them with a subpoena to fish for anything that might accidentally be there.

Just a few years ago, DoubleClick bought Abacus Direct, intending to mashup online browsing with offline purchase habits in order to enable greater behavioral targeting. There was such intense public outcry over privacy concerns that DoubleClick felt compelled to halt the integration of the Abacus data. So, it nearly doesn’t matter if NebuAd does safeguard privacy well enough — if people are creeped out by having ads appear which seem to know what they’re up to, they might lash out at the advertiser, the ad network, and maybe even their ISPs.

On the second issue, will NebuAd get enough ISPs to partner up with them? They’re apparently already partnered with a number of ISPs, thought they won’t say who. This makes one suspect that perhaps those ISPs could be keeping the behavioral profiling a secret from their subscribers — something that would appear to run against the NebuAd pro-privacy stance.

NebuAd says that they require ISPs to allow users to opt-out of the profiling, but ISPs frequently bury subscribers under reams of privacy terms and conditions fineprint to the point where few subscribers can actually figure out what’s done with their data.

Many users are browsing the internet from work or school, and I suspect that corporate and educational networking departments will be unlikely to participate by providing NebuAd with their users’ data. So, there could be a substantial number of users that NebuAd will be unable to profile and target.

Still, all things considered, NebuAd may have built a very compelling business proposition — they could be positioning themselves to eventually be acquired — they’d make a very attractive advertising component if coupled with various other targeting technologies like geotargeting, demographic targeting, keyword targeting, contextual advertising, etc.

(click to enlarge)

It’s interesting and hypnotic to sit and watch where in the world the most recent article update has occurred. It’s sort of educational to witness the sequence of article edits, too, because I think it gives one a real feel for how fast articles can get edited, reverted, re-edited, etc. Just while watching it all unfold, I’ve witnessed little minor tiffs as someone in South America edits and article and then someone in Canada shortly after goes and reverts the edit or adjusts it as well.

The mashup is created by László Kozma, and it’s inspired by David Troy’s FlickrVision which allows you to see the most recent photos uploaded by endusers on a worldwide map.

One thing that I’d suggest that László might add to this: display the username of the person who’s making the edit in the little pinpoint cloud, hyperlinked to their user page!

Spock is one of a new breed of people search engines which pulls data in from a variety of online sources including MySpace, LinkedIn, My Yahoo!, Wikipedia, company websites, blogs, and other sources to compose these composite profiles which include photos, descriptions, links to people related to the person in question, and tag lists of common keywords. Check out this search I did for “Danny Sullivan”:

(click to enlarge)

And, here’s the profile Spock generated for the search engine marketing “Danny Sullivan”:

(click to enlarge)

This automatic generation of profiles from other data sources, similar to a meta search engine, is not all that new, of course – ZabaSearch has been touted for doing similar stuff to compose info on people out of various public records, sort of like a poor man’s background search. And, ZoomInfo has worked to build a directory of searchable business profiles of individuals. IceRocket also used to have a metasearch engine that pulled in data from a handful of various singles/personals sites.

What makes Spock a bit different is how they’re actively composing these profiles from sources that really haven’t been associated with one another previously, and making them publicly available, for “free” (eventually paid for by ad revenue, of course). While the general public likely hasn’t been aware of it, the CIA or NSA has actually also been working on a similar sort of search engine system which automatically composes secret dossiers of information on individuals from a multitude of sources including credit card information, criminal databases, as well as many of the same online sources used by these web services like ZabaSearch, ZoomInfo, and Spock.

Spock has already sparked a small amount of concern just as all people search engines have, due to how they remove barriers to easily finding information about individuals – something which always makes people very nervous. A recent Time article on Spock, “Online Snooping Gets Creepy“, states that it’s controversial because while individuals may opt into providing info on many of the sites that Spock gleans information from, those individuals have not opted into participation in Spock, nor have they agreed to having each datum they put in one site associated with all the other data they may have in other websites. An even bigger concern, both with Spock and with the secretive dossiers generated by the American government on individuals, is that these sorts of automatically-generated profiles inevitably will have some amount of mistakes associated with them, and those mistakes can negatively impact people’s lives.

Nearly any of us who have the misfortune of having extremely common surnames like “Smith” or “Jones” and common first names have had the experience of having our identities confused with many other people’s, and the age of information has only accelerated the effect. At my old company, I had my email address assigned to other “Chris Smith”s not once but twice, for instance, and I used to receive confidential email notes intended for the “Chris Smith” who was in our HR office. I know another gentleman who is a retired, conservative, law-abiding citizen who’s very common last name has put him on a watch list, so he looks forward to special treatment any time he goes to an airport.

What if Spock frankensteins-together a profile that contains incorrect, embarrassing information? A prospective employer might not hire you, people might choose not to date you, and people could assume the worst about you based off incorrect info.

White pages directories have long had to manage consumers’ privacy concerns, but at least that industry has had some time to build up some ways of managing those issues. And, giving individuals the power to choose to be unlisted if they desired removes some of the sensitivity. In white pages directories, one can typically choose to be unlisted or “ex-directory”. In Spock, you can apparently register to update/correct your info, but not remove it. This is another case of technology outpacing our understanding of how to deal with it.

I’m personally not terribly concerned with services like Spock, since I’m accustomed to them, and since I have consciously exposed a lot of personal information on the internet. But, I’m sympathetic to people who may’ve wanted to keep their informal MySpace profiles separate from their professional information, and who may now be surprised to find them all smashed together on services like this.

Otherwise, Spock is a really cool idea, and I like a lot of the clever things they’ve done in assembling information and making it searchable. There appear to be a number of useful applications possible in Spock, aside from the fluffier celebrity profile stuff.

Oh, one other aside, too: Spock appears to’ve purposefully built themselves with very search-engine-friendly URLs, likely with the intention of optimizing to get lots of natural search traffic. For instance, my “Danny Sullivan” search query was formed like this: http://www.spock.com/q/Danny-Sullivan. I bet this explains their hyper-focus on celebrity profiles, since searches for those celebs’ names would likely drive a lot of traffic/revenue to them, once their pages are indexed! Spock is already enjoying a PageRank of 5, and Google is showing 666 pages indexed so far.

(VentureBeat’s writeup on Spock is worth reading, and I see that Greg Sterling just mentioned Spock on Search Engine Land a few minutes ago, too.)

 This new method is called “Supervised Multiclass Labeling (SML)”.  In an article published in the IEEE (Institute of Electrical and Electronics Engineers) Computer Society journal, TPAMI (Transactions on Pattern Analysis and Machine Intelligence) called “Supervised Learning of Semantic Classes for Image Annotation and Retrieval“, researchers Gustavo Carneiro, Antoni B. Chan, Pedro J. Moreno, and Nuno Vasconcelos describe a system wherein one could train it by supplying a sort of seed set of photos which have been labelled by humans with keywords of things seen in the photos, and then the seed set is used by the system on a database of photo images which are unlabelled. The system then will calculate the probability that various objects or “classesâ€? it has been trained to recognize are present – and labels the images accordingly. After labeling, images could then be retrieved via keyword searches using the newly developed meta data.

The accuracy of this new method has apparently been superior to that of other previously published content-based image labeling systems developed by information retrieval specialists. The SML system can also split up images based on their identifiable regions of content – a process which has historically been quite difficult for software systems to accomplish. For example, this method could separate a landscape photo into mountain, sky and lake regions and then identify those things based on the training data.

One of the engineers who contributed on the development of the method and the associated published paper, Pedro Moreno, is a researcher at Google who sometimes contributes on the Google Research Blog. Google just happens to have large quantities of images to use for such research, of course.

John Battelle recently mentioned the dream of being able to truly search by an image’s content, and it would appear that the concept is really close to fruition through this SML method.

Now, one must ask oneself, how might a search engine like Google first develop a good sample set of human-labeled images in order to “train” this sort of new algorithmic labeling program? But, oh, wait — Google has already had an Image Labeling program in beta release which invites users to come in and submit keywords to be associated with images through a sort of game. And, Vanessa Fox told us in the SES Images & Search panel discussion in Chicago last year, that some of their users really enjoy participating in the Image Labeler program. So, it’s not at all hard to connect the dots — if the automated Supervised Multiclass Labeling system were to be hooked up with the rich, trustworthy data developed through the Image Labeler program, Google would almost overnight have the ability to perform true image search based on images’ graphic content.

If that Image Labeler user-tagging method were associated with this new algorithmic method, the random users could become live “trainers” for the software. As time progresses, the software could become steadily more accurate and more efficient at adding appropriate words to be associated metadata for images.

Using this method, Google might only need a relatively small seed set of images to be tagged by humans in order to train their software to identify millions of other images. The end result would be a fantastically better Image Search, using more accurate data for associating users’ keyword search requests with images appropriate for the keyword. This sort of advantage would put them ahead of nearly all the other image search services out there.

Of course, there are other true image search services, but none of them have the user following of Google, I would bet.

A press release from UCSD on the research paper includes a nice video of UCSD professor Nuno Vasconcelos talking about the SML method if you’re interested.

(Nota bene: I’ll be speaking on increasing website traffic through optimization of images and for optimizing for Image Search at SES Conference next week. I’ve also previously blogged in the subject of image optimization.)