Artilium

Thursday, August 20, 2009

Establishing Relevance from Mobile Internet

I am sure you have observed systems which learn your preferences and favourites from your activities – e.g. menus that adapt to most commonly used options, web sites which suggest items you may wish to buy based on your browser history and on-line purchase decisions or even just the request to “press red now”. This is all about establishing relevance from user interactions in order to deliver more relevant and tailored services. There has been a lot of effort recently applied to learning about you from your web browsing history. There are a number of useful parameters that can be obtained from browser activities and given the growth in mobile browser activities these techniques can be equally applied to mobile subscribers. The outputs from this processing include age and gender demographics, long and short-term interests and commercial intent.

When a user has visited one site where the demographics are known and then clicks through to another site, the demographics of the new site can be predicted by summing the portions of the demographic data from each of the proceeding sites that brought users to the current site.  With a reasonable sample of reference points or “known truths” and by tracking click activities via cookies, reasonable estimate of a users’ gender and age can be established. Quantcast is one company at the forefront of extracting demographics from browser activities (i).

Short-term interests can be established based on current browser activities and by correlating this with longer-term interests. Long-term interests can be established by looking at the content of URLs and establishing relationships. The normal place to start analysing URLs is via the metatags, any highlighted data (e.g. headings, capitalisation, bold etc.) and by extracting any search terms. Of course the number of times a page is visited and time spent on a page will influence the analysis. Hyperlinks often indicate related content and by measuring the content similarity of multiple linked pages the URL content data can be enhanced even if the user does not follow that link. Measuring the content similarity of multiple related pages (e.g. from hyperlinks, browser sessions or from a number of sessions) can be very processor intensive but it often reveals superior user relevance data to that obtained from just the metadata. Latent Semantic Analysis and Cosine Similarity are two potential techniques for measuring the relationship between sets of text and for extracting any correlated meaning. Phorm is one of the leading companies active in the extraction of meaning from browser activity (ii).

Commercial intent is a very useful parameter to establish in order to offer relevant services and goods where revenue can potentially be generated. There is a higher commercial intent when a user intentionally searches for and/or browses through commercial content where items are for sale. Where the subscribers’ interests obtained from previous browser or purchase history indicates specific interests, the estimate of commercial intent will increase when the current browsing aligns with one of these interests.  Microsoft has already released simple tools to establish on-line commercial intent (iii).

In summary, from mobile browser history it is possible to establish a user’s age, gender, interests and their commercial intent.  There are a growing number of companies with an active interest in this area and a number of software tools are becoming available. Using these tools, and with the access to rich data sets and the processing capabilities of the ARTA platform, the relevance of services offered can be highly tailored to each user.

(i) http://www.quantcast.com/, 18 August 2009
(ii) http://webwise.phorm.com/, 18 August 2009
(iii) http://adlab.microsoft.com/Online-Commercial-Intention/, 18 August 2009