Research On Affect Recognition And Affect Elicitation Of Agents Based On Ontology

An agent is a kind of machine or software system which has ability to perceive the environment and to perform autonomously the tasks set by designers and users.Traditional agents can simulate the mental states of persons,such as beliefs,desires,and intentions,but not with the affective intelligence.So in the process of interacting with them,these agents give the impression that they are just a machine.The introduction of the technology of affective computing makes the agents have ability to recognize,analyze and synthesize affect similar to human.Affective agents are more believable,more accurate,more attractive,and more effective if they seem to have emotional states matching those of humans in the same situations.Thus affective intelligence is an important research direction of agent and multi-agent systems.Although agents have the ability to recognize and express affect,considerable work remains to be done.For example,for agents,how to achieve a consistent beliefs in affect data semantics of the user and affect knowledge of the agent? How to construct the affect recognition model based on physiological signal that reflects real brain function activities? How to elicit the emotions similar to peoples? That is,what time,under what conditions,what kind of emotions are elicited,and whether the emotion generated is correct(i.e.,similar to humans)?Based on the above problems,this dissertation discusses several key technologies and key issues involved in affective agents based on ontology,including the construction of architecture of the affective multi-agent systems,the construction of knowledge base,the construction of user affect ontology model,and the construction of the agent affect elicitation ontology model.The research on these problems will provide the technical support for the establishment of harmonious human-agents interaction.The main contributions and innovations are as follows:1.A top-level ontology model framework of affective agent is proposed.The agent is organized in a distributed structure in this framework,and the framework consists of a user affect ontology model and an agent affect elicitation ontology model.The user affect ontology model addresses the modeling of related emotional knowledge of the user and addresses the reasoning of the emotion states of the user,thus the agents have the ability to model the consistent semantics of both the context information and the emotion state of the user.Additionally,the agent affect elicitation ontology model addresses to model both the low-level reflexive emotion and the high-level cognitive evaluation emotion to generate the human-like emotions.Finally,the application of the top-level model in the depression caregiving agents is also discussed.2.The user affect ontology model is proposed,which can achieve both the user emotional knowledge representation and the user emotional state inference.We develop such a model which contains an ontology model called EmotionO+ and rules set based on EEG,which is obtained by random forest algorithm to predict emotional state.It presents not only an effective method to enable semantic representation of the EEG and fNIRS data,but also an emotion knowledge mining tool.Results using EEG data in the eNTERFACE 2006 dataset show an accuracy for our proposed model of 99.11% as compared to 97.8% for competing methods using the C4.5 algorithm.The experimental results demonstrate that the posited approach is potentially usable for sharing of physiological signal data related to emotional response between different healthcare multi-agent systems.3.The agent affect elicitation ontology model is proposed,which can achieve both the representation of the mental states of an agent and emotion elicitation based on the agent mental states.We provide a logical formalization of event-based emotional triggering for multi-agent systems psychologically grounded on the cognitive theory of emotions proposed by Ortony,Clore,and Collins using OBDI logic(observation-based belief,desire,and intention logic),but also addresses how symbolic model checking techniques can be applied to verify the properties of the emotional BDI(belief,desire,and intention)multi-agents model proposed herein by employing a modified version of the model checker called MCKBDI(model checker for knowledge,belief,desire,and intention).The results confirm that the modified MCKBDI can be applied to verify the emotion specifications of finite-state emotional multi-agent systems and that one can program agents capable of using and reasoning over emotions.This approach can be utilized to confirm that the computational model of emotion is formalized as intended.In summary,this dissertation explores the key technologies of the affective multi-agent systems and solves the key issues of knowledge sharing related to emotions between the users and the agents,affect recognition of the users based on the EEG signals,and affect elicitation of the agents based on OCC theory.Specifically,this dissertation constructs the ontology model of affective agent consisting of the user affect ontology model and the agent affect elicitation ontology model.Furthermore,this dissertation studies new methods of verifying whether an emotional logical model works as intended.Experimental verification comprehensively explores and clarifies the theoretical mechanism and feasibility of new technologies.This study has enriched the theoretical basis of the affective multi-agent systems and has made beneficial applications to help promote the further development and application of this method.