contentDetection allows to target Ads based on semantic content of a web page

Manufacturer:	mindUp Web + Intelligence GmbH
Contact:	Herr Joachim Feist Tel.: 07531 2825810   Contacting via e-mail

mindUp contentDetection® is a technique to allow contextual advertising on web pages. By crawling the pages on which banners reside and detecting the topic of the page it allows to generate banners with high relevance to the surfer. This leads to high click rates and conversion making such banners to a success for the publisher as well as the advertiser. Because affiliates do not need to optimise banners by themselves, integration is easy and together with the high revenue potential leads to a high percentage of webmasters using this type of banners compared to banners of other advertisers.

contentDetection is a service that finds out the best fitting general topic and relevant keywords and phrases for a webpage. For example for a web page about the apple iPod it would find out, that the page best fits to the “Audio” category and "MP3-Player" subcategory and also yield a set of keywords like "Apple iPod Shuffle", "Apple iPod Nano", "MP3-Player" and so on.

Another possible usage is to steer banner pools based on the category information returned by contentDetection or to fill search boxes with a keyword fitting to the page.A third usage is to use the category / keyword information to steer text ads.

contentDetection setup

The system builds on text understanding algorithms by mindUp. The technique is trained on the target product data and the target category system. This makes sure that only topics that exist on the advertising platform are being returned and only keywords that will return products / text ads will be used. This also allows the system to be mainly language independent.

Asynchronous crawling

When the system is called with a new URL this URL is passed on asynchronously to one of the crawlers. The page is read and the different parts of the page (title, header information, content, style) are analysed by the text understanding algorithms. The results are passed back and stored for future calls for the page analysed. This mode is used for huge amounts of new URLs.

Synchronous crawling

In some business processes an immediate result ist wanted, e.g. in the process of generating an advertisement for a specific URL. By using a certain parameter setting, the system crawls the page and returns the result immediately.

Follow-Up Crawlings

contentDetection results always have a validity date. The caller should store the result up to this point in time. Shortly before validity ends the system visits the pages in the system again and will return the updated information when being called next time. contentDetection automatically detects wheter the page has changed its topic or not hereby adjusting its recrawling period for this page.

contentDetection Parameters

The system allows for steering parameters by the caller: He can force the system to return only results from a certain category range or exclude certain categories. He can force the system to use certain keywords no matter what has been found on the page. The number of keywords returned can be set from 1 to 9. The result always contains one level one category, the number of level 2 categories can be set from 1 to 5. contentDetection can display its results as XML or Javascript.

contentDetection Features

• Language Independent: The system learns parts of its rule system from product data. This learning process is mainly language independent. The system therefore can be made available for most languages including Chinese.
• Sophisticated Rule Learning Algorithms: mindUp is expert for Forecasting, Data Mining and Text Mining. The algorithms used have been developed and refined in various applications over many years.
• Retraining: By a monthly retraining of the system also new topics and keywords can be detected.
• Hints by Caller: It is possible to give the system hints such that the classification of one page is limited to certain categories allowing the end user to influence the results.
• Optimised Recrawling: Pages that were crawled initially are visited from time to time to update the content information. The system detects which pages change their topics often and which don’t and hereby changes its recrawling time period.
• Tuning: The learning process is guided by a tuning database that measures the quality of the whole system and allows for a constant quality increase.
• API: The system is coupled to the product API
• Multi-Platform Solution: The solution is built in Java and can run on various platforms. Usually it is hosted by mindUp but can also run on severs of the client.

• Advertiser
• Huge shopping platforms
• Affiliate networks
• Advertising Agencys