Vol. 25 · No. 2-3

This paper proposes a model for explanations in a set theoretical framework using the notions of closure or fixpoint. In this approach, sets of rules associated with monotonic operators allow to define proof trees. The proof trees may be considered as a declarative view of the trace of a computation. We claim they are explanations of the results of a computation. This notion of explanation is applied to constraint logic programming, and it is used for declarative error diagnosis. It is also applied to constraint programming, and used for constraint retraction.

The number of news reports published online is too large for any person to read all of them. Not all of these reports are equally interesting. Automating the identification and evaluation of interest in news is therefore a valuable goal. This paper presents a framework that permits the identification of interesting news by means of violated expectations. Facts derived from news reports, expectations and related background knowledge can be used to (i) justify the decision to rate news as interesting, (ii) explain why the information in the report is unexpected and, (iii) explain the context in which the report appears. Explanations supported by this framework are general purpose explanations based on the data in the system. The explanations are natural language renditions of first order logic facts and rules.

CBR systems solve problems by assessing their similarity with already solved problems (cases). Explanation of a CBR system prediction usually consists of showing the user the set of cases that are most similar to the current problem. Examining those retrieved cases the user can then assess whether the prediction is sensible. Using the notion of symbolic similarity, our proposal is to show the user a symbolic description that makes explicit what the new problem has in common with the retrieved cases. Specifically, we use the notion of anti-unification (least general generalization) to build symbolic similarity descriptions. We present an explanation scheme using anti-unification for CBR systems applied to classification tasks. This scheme focuses on symbolically describing what is shared between the current problem and the retrieved cases that belong to different classes. Examining these descriptions of symbolic similarities the user can assess which aspects are determining that a problem is classified one way or another. The paper exemplifies this proposal with an implemented application of the symbolic similarity scheme to the domain of predicting the carcinogenic activity of chemical compounds.

In this paper, we describe precedent-based explanations for case-based classification systems. Previous work has shown that explanation cases that are more marginal than the query case, in the sense of lying between the query case and the decision boundary, are more convincing explanations. We show how to retrieve such explanation cases in a way that requires lower knowledge engineering overheads than previously. We evaluate our approaches empirically, finding that the explanations that our systems retrieve are often more convincing than those found by the previous approach. The paper ends with a thorough discussion of a range of factors that affect precedent-based explanations, many of which warrant further research.

In a military context, the process of planning operations involves the assessment of the situation, the generation of Courses of Action (COAs), and their evaluation according to significant points of view, in order to select the COA that represents the best possible compromise. To support this process, an advisor tool has been developed to assist a military Operation Centre staff in managing events and their related COAs, as well as prioritizing these COAs according to different evaluation criteria by means of a Multicriterion Decision Aid procedure. This paper describes an automated approach for explaining the ranking proposed by this decision support system.

In this paper we consider an interactive and explanation based dialogue approach to complex and `wicked' planning problems. Wicked problems are essentially imprecisely formulated problems having no clearly defined goals and constraints. The dialogue approach is aimed at reducing the problem complexity during interaction with the human expert. The involved software agents are mostly optimization procedures. The approach contains the following steps: (1) Selection of a specific concern in a proposed solution; (2) Calculation of a stakeholder defined ideal plan; and (3) Comparing the actually generated plan and the prototype based on a similarity measure. The comparison of the actual and the ideal plan looks at aspects of interest for the stakeholder such as resource consumptions or structural properties of the plan. The proposed approach is generic and was applied and customized to three classes of wicked problems: release planning, investment planning, and urban planning. All three applications are described and illustrated in the paper.