Professor David Crystal is a linguistics expert and has developed ‘semantic ad targeting’ technology that he claims can be used to more accurately place ads in the correct context.
We talked to Prof. Crystal and ad pepper’s board director Sacha Carton about the technology…
Tell me about iSense…
Sacha Carton: Our technology has been provided under the brand name iSense, our range of semantic ad targeting solutions. The technology allows for precise targeting of display ads, banner ads and videos in our network of publisher sites.
This is our key offering in the UK and Europe. The semantic engine that is driving this targeting was developed by Professor Crystal at Crystal Reference Systems Limited, which ad pepper acquired in 2006.
How was this developed?
Prof. David Crystal: In the mid 1990s, when I first started working on online ads, the technology was very primitive, and ads were often placed in very inappropriate contexts, such as ads for knives next to new articles about stabbings.
The first solution was to develop contextual targeting, where ads were matched to the content on web pages, but this is still a limited approach.
In order to understand what a page is about, you need to target the entire semantic content of the web page. We developed semantic targeting technology that analyses everything on the page. The key here is to understand which words you can use to target ads, and the relationships between these words.
To understand which words and phrases were relevant to each target category, we have had to manually trawl through the entire dictionary, assigning words and phrases into 3,000 categories.
This work took myself and a team of 40 lexicographers almost ten years to achieve, since we patented the idea in 1998. Ad pepper’s clients have a selection of 3,000 target categories, each containing 250 words and phrases.
How does it work in practice?
SC: It took us sometime to integrate it into our ad servers, but this has now been completed, and we have been selling the product for six months now. Advertisers tell us their campaign objectives, and we target the relevant categories.
For example, we have provided ad targeting for Setanta, who were interested in precise, club-level targeting for some of their ads. Our technology allows us to precisely target a banner or other ad on a Man Utd related website, if we are promoting a live Man Utd game on Setanta.
DC: We do this by targeting and understanding the words and phrases that are used in websites about Man Utd. This could be names of players, the ground, manager and so on, so it is something which we need to keep up to date.
The core of the technology is ‘sense disambiguation’. Words can have different meaning in different contexts. The average number of senses per word in the dictionary is 2.4, so we need to understand which meanings are relevant in each context.
How accurate is the targeting?
DC: Our first analyses a few years ago found that the targeting was 90% accurate, but we think it is now more like 98%. This compares with less than 50% for other methods of ad targeting.
The only kinds of pages we can have problems targeting are those that are image heavy, with little text to analyse. If there is little text on the page it is difficult to classify.
Does the technology only work in English?
SC: No, we have rolled this out in Germany, Holland and Denmark, and plan to add more languages as we go on, with French, Spanish and Swedish to come.
DC: Once we had the words in English, we had to go through the same work for other languages, which goes beyond merely translating the words into understanding their meanings in various contexts.
How do you avoid ads appearing next to offensive or inappropriate content, as happened last year with Facebook?
SC: The acquisition of Crystal Semantics allowed us to develop Site Screen, which allows us to define sites with questionable content where we will not run any ads. We can read the content of the site to determine the nature of the site and avoid adult content, violent content or anything else that the advertiser wants to avoid.
We offer this service to advertisers on our own network, but also we also offer this service as an ASP to other clients, including the NSPCC.
Is it suitable for branding campaigns?
SC: As well as wanting to reach their target audiences, brand advertisers want to control the environment where their ads are displayed. Currently, most can only target ads at tag level, which relies on how well the websites and content has been tagged.
We allow advertisers to micro-segment content on web pages, which allows for more accurate and relevant targeting, something we feel addresses a core need for advertisers.
There has been a lot of controversy about Phorm recently – where do you feel it has gone wrong?
DC: They have put their money on a contextual based solution, which we feel is an unsophisticated method of targeting advertising, as it doesn’t solve ambiguities.
While our understanding of where to target ads is based on an understanding of the web page where ads are targeted, they don’t drill down to page level, so theirs is an adolescent approach to targeting.
What kind of take-up have you been getting since you launched iSemse?
SC: We have seen a huge uptake from publishers, and we now have more than 500 portals and other content sites in our network.
Winning an IAB innovation award recently has helped us to get agencies interested in our products, and we are now dealing with some of the top agencies with blue chip clients.
Do you have any direct competitors?
SC: Our main competition in online ad targeting is contextual advertising firms. There is no shortcut for the work we have done to develop our database and technology so it would take someone to invest a lot of time, resources and expertise to compete on semantic targeting.
As advertisers seek more control over how and where their ads are placed, more will choose our product.
Online Advertising Networks Buyer’s Guide 2008
Ten digital trends to watch out for in 2008