• Co-Supervisor :
  • Patricia Everaere (CRIStAL)
  • Funding : ANR
  • Start year :
  • 2021

One of the main ways to obtain information and knowledge about our environment is to interact with other agents or sources of information. These agents or sources will send us information, which we can then take into account to form our own opinion. To take this information into account, we can use tools from the theory of belief change.

But before this integration step, there is a preliminary step that is indispensable: deciding whether this information is reliable enough to be taken into account. And, more generally, it would be interesting to evaluate its level of reliability (for example, to know how to resolve potential conflicts with other information that we already possess).

The problem can be summarized as defining a measure of the trust we have in each source providing us with information.

We want to base this evaluation on objective information. And a way to conduct this evaluation is to compare the different information received from the different sources, in order to identify the most discordant sources. And to study how this discordance informs us about the reliability of each source.

This is an important issue for many applications where one receives information from sources of unknown reliability. This is the case for all of us when we start interactions with a stranger. After a few interactions, where we compare his opinions with our own and with those we have obtained from other people, we conduct an evaluation of the confidence we can have in the claims of this person. But this problem also arises for an information agency obtaining information from different sources whose reliability must be specified. One can also imagine improvements in distributed database queries (such as web queries for example), where with an estimation of the degree of reliability of each source, one would manage to resolve conflicts, and provide better answers.

The goal of this thesis is to study how to define measures of confidence of information sources based on an evaluation of the information historically provided by these sources and by confronting them.

Conflicts between the information sources will be used to evaluate the reliability of the different sources. These conflicts can be measured using tools derived from the inconsistency measures for propositional logic. These measures allow us to evaluate the importance of the conflict, but also to locate it, whether it is at the level of the sources, but also at the level of the information itself.

Several points will be studied. The first will be to study the literature on these confidence measures (in particular in other disciplines such as databases and the web). The second point will be to propose definitions of trust measures. The third point will be to study the properties of these measures. In particular, we will define the properties that we would like to obtain from these measures. We will try to obtain a characterization of reasonable measures by means of representation theorems. We will also try to show the relevance of these measures, i.e. how these measures of reliability capture the underlying reality.