Improving social relationships in face-to-face human-agent interactions: when the agent wants to know user's likes and dislikes

Abstract

This paper tackles the issue of the detection of user's verbal expressions of likes and dislikes in a human-agent interaction. We present a system grounded on the theoretical framework provided by (Martin and White, 2005) that integrates the interaction context by jointly processing agent's and user's utterances. It is designed as a rule-based and bottom-up process based on a symbolic representation of the structure of the sentence. This article also describes the annotation campaign-carried out through Amazon Mechanical Turk-for the creation of the evaluation dataset. Finally, we present all measures for rating agreement between our system and the human reference and obtain agreement scores that are equal or higher than substantial agreements.

Grounding the detection of the user’s likes and dislikes on the topic structure of human-agent interactions

“…The analysis (see Fig. 2 ) consists in the process described in [8] . 30 sessions of the corpus have been used for the development set.…”

“…The first version of our system [8] was designed for processing jointly, both agent and user's speech turns belonging to same adjacency pairs. Through the context of an adjacency pair, it was possible to deal with the expressions of like or dislike built in a collaborative way (such as question -answering about the user's preferences, see example below).…”

“…The system processes jointly each adjacency pair, which allows us to model the agent's utterances in order to help the detection of the user's sentiment-related phenomena. Even though the system presented in [8] shows interesting results, a larger conversational context needs to be considered for improving the efficiency of the system [10] . In [8] , the adjacency pairs were processed without considering neither the progression of the conversation nor the topic structure of the ongoing dialogue.…”

“…Even though the system presented in [8] shows interesting results, a larger conversational context needs to be considered for improving the efficiency of the system [10] . In [8] , the adjacency pairs were processed without considering neither the progression of the conversation nor the topic structure of the ongoing dialogue. A contextualisation of the likes and dislikes with respect to the topic structure of the conversation -defined by the interaction scenario -can be very helpful for the detection system.…”

“…In [8] , we address some of these issues: the selection of the relevant sentiment-related expressions, the likes and the dislikes, and the integration of a first level of dialogic context, the adjacency pairs. We delimit and specify the linguistic phenomenon to detect by focusing on one specific aspect required by ECAs for modelling social relationships: the user's interests that are given by the expressions of the user's likes and dislikes in the verbal content.…”