A computational model that aims to endow intelligent software agents with the capability to understand emotions is the focus of this paper.

A general framework for machine understanding is tailored to the problem of emotion understanding, resulting in an architecture with four functional components. Each component is described in detail in the first half of the paper, highlighting the proposed contributions. The core of the proposal is a metamodel that consists of an episodic memory and a semantic memory. The general knowledge an agent has about emotions is stored in the semantic memory, based on the Ortony-Clore-Collins model of emotions [1]. In addition, graphs in the semantic memory that are learned from the agent’s experience are encoded in the episodic memory. A concrete experience encodes an action performed by one agent (the owner of the memory or any other agent) toward another agent, the emotion that the action elicited in this agent, and its intensity. Agents are categorized into different groups according to their goals, making it possible to apply what has been learned about a group of agents to a different agent in the group. Three different categorization methods were implemented and tested.

The second half of the paper describes how the emotion-understanding framework was evaluated. The testbed was a multiagent simulation in which agents tried to apply what they know and have learned about emotions to elicit target emotions in other agents. The evaluation was aimed at determining the contribution of different memory components to the process. Five memory configurations were designed by selectively enabling or disabling some components. Their performance was compared using three metrics taken from the information retrieval field (recall, precision, and F-score).

The proposed framework and the evaluation performed are really interesting, and the paper succeeds in explaining them clearly and with a level of detail that would allow researchers interested in the design of intelligent agents to put it into practice. The simplifications assumed and the limitations of the work also are stated.