8

process it in an arbitrarily malicious way, easily cheating an unaware citizen.

Our aim here is thus to implement the toolbox in a way preserving transparency, by following the principle of Minimum Trusted Computing Base. In particular, one of the recommendations we want to follow is to give citizens (users in our context) the ability to check the results of all applications lying on the toolbox. To this aim we will develop a certified trusted tool for both data and tool results verifiability.

1.1.3 Approach

Our aim is to design a secure, transparent and knowledge-based platform, combining both extensibility and scalability. The platform will be extensible meaning that users-citizens, governors and companies-are allowed to propose their opinions or endorse existing ones. Moreover, the platform will be scalable, permitting special users-governors and companies-to design new applications lying on the trusted core, the toolbox.

In the following sections we will describe how we intend to give knowledge learning and creation a predominant role (Sect. 1.1.3.1), how we will secure the system from malicious attacks (Sect. 1.1.3.2) and the mechanisms we will apply to oblige applications to supply the users with correct results (Sect. 1.1.3.3). Finally we will give implementation insights (Sect. 1.1.3.4).

Our contributions will be applied to a concrete framework, called EUGAGER.

1.1.3.1 Knowledge

In order to make knowledge the predominant force entailing decisions, we will design a multidimensional collaborative recommender system, which offers to the user multiple personalized possibilities to filter EUGAGER's content. It will be necessary to create user profiles, which contain the user's ratings for items on multiple dimensions such as interest, agreement and importance. This allows statistical analysis of opinions and personalized filtering of new and existing content so that the user can be informed of new documents or applications, which concern him in different ways. The variable types of rating must be analyzed and classified either manually (if possible in advance) or by statistical analysis of the user behaviour later (like modern search engines adjust from time to time to the user behaviour).

Even if peoples' opinions are often diverse there are types of ratings, which are more objective than others. E.g. most people would agree that a text containing swearwords is malicious in a political discussion. An opposite example would be the rating of a song depending on the users liking which is of course very subjective. So we can analyze the rating dimensions in this regard and similar ways to explore dependencies of various natures between them. Those interpretations can motivate new kind of similarity measurements between users depending on different weightings of the rating dimensions.

There are different points of view to explore and research how the dimension weights may be interpreted and set. As proposed before the user should have the possibility to set them manually before every search and we will also provide predefined templates. Another way is to handle them class-dependent, so that all contents (or the class it belongs to) will have a specific profile of dimension weights to boost its important rating dimensions and therefore the quality of the resulting similarity measurement. It is also possible that an object has more than one profile, depending on the searcher's perspective.

Maintaining user profiles - which is necessary for a collaborative filtering system - is always a privacy concern and must be done securely. Of course the user has to agree actively that a profile is created. In general the user may enter EUGAGER with and without identity verification which implies that at least two profiles are being used so that there is no way to find a connection between the anonymous and the official behaviour ofthe user.

The challenge posed by the proposed EUGAGER toolbox and applications goes beyond the state of the art. Thus research is necessary.

In order for the system to be useful in the e-government decision process, users' opinions must be used to learn and create new knowledge. In the remaining part of this section, we will describe how we intend to perform these tasks.

Knowledge Creation. In order to create (synthesize) knowledge, we intend to build EUGAGER as a Darwinian system, meaning that we