Campaign Selection

User Analysis (User Classification)

The recent user history as part of the current session provides important hints for the campaign in general, and for the design and selection of addressing within the advertising in particular. In order to exploit such hints, specific characteristics need to be identified, their technical derivability must be clarified, and their selection be combined into feature sets. Different contexts are both covered by topics, as well as being studied in regard to their effects on user mood or behavior. In addition, a concept for managing session information with cookies on the user side is to be developed.

The user profiles available from Behaviour-targeting provider United Internet Media AG (provider of the TGP Open Source System) which are aqcuired from the cross-session user history are also to be used for the AdMotional targeting.

As to record and describe the impact of web sites on a users’ mood, available models for the description of emotions or moods are to be tested (and adjusted or extended as necessary) for their applicability within AdMotional in cooperation with the University of Bonn. The aim is to develop a suitably exploitable classification scheme for moods and emotions. Furthermore, methods for mapping the information as present on the visited web sites into the model categories needs to be developed (Emotional Classification). This necessitates laboratory studies, developed and conducted by the University of Bonn (Department of Epileptology, Medical Faculty of the Rheinische Friedrich-Wilhelms University), using for example magnetic resonance imaging (MRI) techniques.

Incorporating additional information resources

Apart from context and user information, further targeting-relevant sources may be used. Among others, these include Geo-information (location), weather and e.g. current concert event services. Those sources are examined in regard to their suitability (towards the targeting process) and availability (technically and economically). Another challenge in this sub-project is the necessity to provide all information with sufficient run-time performance (i.e. speed). To allow for an optimal targeting on a wide range of mobile devices, not only GPS-based localisation but also Wi-Fi hotspots and mobile network cell information is being evaluated for their appropriateness within AdMotional.

Rule-based, adaptive targeting

The aim is to determine requirements for, and definition of, a rule language, capable of representing the dependencies among user-specific and landing-page related properties with logical expressions. Furthermore, a knowledge processing approach is to be identified, which allows for the deductive creation of a knowledge base, consisting of the above mentioned dependencies and their relationships with relevant user behaviour (i.e. Click/Action). The knowledge base eventually allows for the deduction procedure towards the selection of a particular campaign to be shown.

The central task of classical targeting is to select the most appropriate campaign and the corresponding advertising material – depending on campaign objectives and internal campaign optimisations, as well as simple criteria such as static context (called channels), time and/or place (IP-targeting and localization). Taking into account all the above criteria, the campaigns considered most appropriate including parameters for dynamic customisation are to be selected. This selection is based on rules. On one hand the different targeting rules contain logical conditions, and criteria-based values for the customisation parameters on the other hand.

First, relevant targeting criteria are to be selected for the rule mechanism. Certain restrictions will be necessary due to complexity, while other criteria might be added as results of the empirical examination, proving some criteria more relevant than others. The aim of this sub-project is the development and application of a methodology that allows for the evaluation – and, idealy, measurement – of the relevance of individual pieces of information for the targeting process, ranging from emotions and context, through session history to time and location.

To optimise the targeting mechanism, the rule mechanism is to be extended with a success-feedback channel. Therefore is is necessary to determine if the shown advertisement had the desired effect (i.e. Click or Action) on the user. If such effect was present, the responsible targeting rule with be reinforced. This way, the rule mechanism becomes capable of learning.