Andrei Broder from Yahoo! Research gave two seminars at Politecnico di Milano on introduction to internet monetization (in Como) and on targeted advertising (in Milano) as a branch of computational advertising.
Computational advertising is about finding the best match between a given user in a given context and a suitable advertisement.
The context can be a web search result page, or a content page provided by a portal. The ads should match these contents.
Ads aim at showing the product, provide information, induce direct action (e.g., direct marketing such as expiring coupons, which induce some urgency in the reader) but also build a general and long-lasting image for a brand.
The core motivation of targeting the advertising is that sending the appropriate advertising to more interested users is the best option for everybody: the advertiser (who gets to the users he is really looking for), the advertising company (who gets more clicks on the ads, i.e., a higher click per view rate, and therefore earns more money), and the user himself (who can finally get advertising interesting for him).
Advertisers use targeting in several ways (geotargeting, demographics, adv. channel, and so on).
The problem addressed by this discipline can be summarized in the problem introduced by information overloading. When there is a large availability of information, there is a scarcity of interest of users. In modern time one must put a huge effort for getting users’ attention.
Technically speaking, the goal of targeted advertising is to raise the Accuracy-Reach curve (i.e., precision-recall curve).
Targeted advertising can be achieved through two main approaches:
- rule based: the advertiser provides a set of rules for targeting the right segments of users (e.g., based on age, geographical info, sex, and so on).
- model based: the advertisement platform defines a user model that lets it address better the kind of users that might be interested in an ad.
Modern targeting is model-based and it relies on the concept of persona, which is basically a relevant behaviour and profile of user. A persona is a facet of personalities. A single user may cover different personas.
The other crucial aspect is based on interests or topics that people like. These can change in time. To play good targeting you need to draw temporal pictures and consider combination of topics that are relevant at the same time.
Demographic targeting is the classical approach.
One important technique is re-targeting, which is a particular case of behavioral targeting: you do something on a web site and later on, when doing searches or other online activities, you get advertisements from that web site.
That can be simply achieved through cookies added to your browser when visiting the original site. The question is how much is this acceptable? Do users feel their privacy has been violated?
One basic solution is obviously to delete cookies in the browser. Statistics show that people tend to be more careful in time and delete cookies more often.
|Seminar by Andrei Broder at Politecnico di Milano on Targeted Internet Advertising Issues.|
|Marco Brambilla introducing Andrei Broder at the seminar in Como.|
|Andrei Broder during the seminar on Internet Monetization in Como.|