E-government involves data sharing between different partners such as citizens and government agencies. Thus, the use of personal data in such cooperative environment must be done in legal ways and for legal purposes. In this context, issues related to data protection, such as privacy, have to be considered. This paper adopts a multi-agent based approach to manage privacy concerns in e-government systems. The proposed model provides a mechanism for e-government systems to evaluate trust degree reached by digital government processes. For this purpose, concepts of responsibility proposed in multi-agent systems and access rights used in security models, are integrated in this work. The research provides an evaluative framework for trust degree related to e-government process.
Privacy refers to the ability of individuals to control the collection, retention, and
distribution of information about themselves [
The rest of this paper is structured as follows. In the second section we present different visions for privacy protection in the literature. In section 3 we describe our proposed model including the fundamental concepts used and their formal representations. Section 4 is devoted to experimentations and simulation results. In section 5, we make a comparative study of our work with other proposed model in the literature. Finally, section 6 summarizes the contribution of this work, and provides conclusions and future work.
Many approaches have been used to manage privacy concerns. We specially note
those based on users preferences such as the P3P (Platform for Privacy Preferences)
[
There are also a number of agent based privacy schemas in the literature. We note
Hippocratic Multi-Agent Systems (HiMAS) [
In this section, we introduce our multi-agent based model that we call ABC (AgentBased Control). First, we describe the concepts used. Then, we present the ABC mechanism, and finally we describe our proposed techniques.
The ABC model enables to manage privacy concerns in e-government systems. This model offers a mechanism based on the use of a set of privacy rules and a set of information on the parties' rights, roles, responsibilities and restrictions to make a statistical assessment of trust. Consequently, ABC model enables the subsequent authorizations to transfer private data.
In the ABC model (see Fig. 1), we define two kinds of agents: Admin agents and AP agents (Authorization Provider agents). Admin agents represent the staff working in government agencies. AP agents are charged to provide authorizations to Admin agents in order to communicate with each other or to access objects in the system. Each Admin agent in our model plays a set of roles (e.g. the tax controller, the mayor, etc). A role includes a set of responsibilities (e.g. mayor roles: sign documents, validate, etc) that are restricted by access-rights (e.g. read only, write, read-write, etc). These access-rights are used to protect the resources and they are managed by a set of privacy rules. Access rights are also justified by a specific context. In fact, what is appropriate in one context can be a violation of privacy in another context.
We define responsibilities as the restricted behavior of a given agent. In other words, responsibility is the behavior that the system expects from an agent based on restriction rules (RR). We note that restriction rules are used to manage agent behavior at an internal level (e.g. making temporary results in a standard format before continuing execution, requesting specific authorizations before performing some actions, etc.). To more explain the concept of responsibility, let’s take this example: we suppose that A1 is an agent responsible for sending e- mails to agent A2 (we suppose that e-mails are confidential). When we observe A1’s behavior, we find that A1 sends e-mails to agent A2 and sends a copy to the agent A3 at the same time. Thus, A1 in this case did not assume his responsibility as some restriction rules are not respected.
Citizens must have control over their personal information handled by government
systems. The Fair Information Practices (FIPs) [
In the ABC model, we define the following privacy rules that are in compliance with the FIPs and considered as the core needed to test and apply ABC model: - R1: each agent must assume his responsibilities.
- R2: each agent has access only to objects needed for doing the set of his responsibilities.
- R3: agent cannot use the context to access linked data outside the set of his responsibilities.
For the privacy rules specifications, we are based on the notation of the rule-based
systems [
In the ABC model, we define a distributed architecture in which sets of the agent in different groups are interacting with each other. Each group (container) represents the set of Admin agents running in the same e-government agency. To have access to objects in the system or to communicate with other agents, each Admin agent must be authorized from AP agent that exists in his group. To keep control of the system, AP agents use a Rule Base (RB) and a Knowledge Base (KB). RB includes the set of privacy rules and KB includes the set of knowledge in the system: agents, their roles, their responsibilities, their RR, their access-rights and their resources. In ABC model, we suppose that in case of failure, AP agents can switch roles dynamically with Admin agents. AP agents delegate the control of the system to the most trusted Admin agent. This delegation decision is based on the computation of Admin agent’s honesty degree that will be explained in the next section.
In this section, we define techniques used for privacy protection in ABC model: the
computation of the trust and the honesty degrees. Formally, our ABC model
correspond to the following set: {A, Re, Au, PR, T, R} such that:
A: the set of agents in the system
Re: the set of agent responsibilities
Au: the set of authorizations
PR: the set of privacy rules
T: the trust degree reached in an e-government process. T is defined as:
where k represents the total number of agents in the system, h represents the honesty
of agent j estimated by agent i (see Fig. 2). h is defined as follows:
(1)
(2)
After each transaction, an Admin agent i can give feedback to Admin agent j
according to the service quality of j. Thus, a feedback score S is calculated as follows:
S= P-N, where P is the number of positive feedbacks left by agents and N is the
number of negative feedbacks from agents. The S value is disclosed to all agents in
the system. This reputation model has been presented in [
We represent risk degree associated to the use of personal information (R), by the following:
R = 1-T . (3)
Using agent technology, we can make « behavioral simulation » that enable us to
create a virtual image of the reality. This is considered as a powerful predictive
analysis tool that enable decision makers to test their idea via scenario in an artificial
environment before implementing their decision in the real world. In this section, we
present our Multi-Agent Based Simulation (MABS) of agents’ behavior during the
company formation process in Tunisia. We chose this scenario because it is complex
(involves numerous administrations and many interactions) and requires the collection
of many sensitive data at every step. We represent governmental agencies involved in
this process by a set of agents (twelve agents) interacting with each other on behalf of
the user (the citizen) and we simulate their behavior during the whole process using
the ABC model. Our MABS is implemented using JADE platform [
At every simulation step, we evaluate Admin agent’s honesty and the trust value.
The first experimental result is the evaluation of trust degree reached during the running of company formation process. The trust (versus the risk) degree obtained during nine steps of our simulation is plotted in Fig. 3.
As observed, we note that T is increasing notably during the simulation and R is decreasing (as there is a dual relation between T and R).
According to our model, honest agents (having h=1) are placed in a trusty zone and dishonest agents (having h<0,5) are placed in non-trusty zone. As shown in Fig. 4 (a), we note that in the first step of our MABS only one agent is honest. However, when we observe agent’s behaviors during next steps, we note that the number honest agents increase (Fig. 4 (b) shows an example of this increase during the step 4 of our MABS).
According to this interpretation, we find that agents placed in the non-trusty zone want to behave like honest agents, they want to move to the trusty zone. This enabled us to interpret the increase of the trust value during the simulation.
Regarding to some standard evaluation criteria, such as the use of access control
mechanisms, the use of user preferences, the trust evaluation metrics [
Acknowledgments. This work has been supported by a common grant from Tunisian and French government relative to the project STIC DGRSRT/INRIA entitled: multiagent coordination models for e-government systems.