Opinion Analysis in Interactions

3Analyzing User Opinions in Human–Agent Interactions

The forum and call center data used to develop the models described in the previous chapter were obtained from both oral and written human–human interactions. These interactions are symmetrical, in that both speakers are human and may thus be assumed to have similar capacities for understanding and social intelligence.

In this chapter, we shall consider human–agent interactions, in which there is a significant asymmetry between the two participants, both in terms of their capacities (at the present time, the understanding and social intelligence of agents is significantly inferior to that of humans), and in the way in which they are considered by the system. For the human user, the aim is to understand and recognize behaviors; for the agent, the aim is to generate behaviors that are relevant to the interaction and can be understood by the human participant.

We shall present our research concerning the development of modeling and detection models of users’ socio-emotional behaviors in the context of asymmetric human–agent interactions. As in the previous chapter, we shall focus on the verbal content of speech using sentiment analysis. Figure 3.1 illustrates this approach, showing opinion and sentiment phenomena expressed verbally by the user in bold underlined text.

Our first research question (Q1: relevance of theoretical models for the construction of computational models as a function of the application), defined in the ...