Human-Robot Interaction Conversational User Enjoyment Scale (HRI CUES) Dataset - Anonymized

Human-Robot Interaction Conversational User Enjoyment Scale (HRI CUES) and this corresponding dataset aim to provide tools for measuring user enjoyment from an external perspective to supplement self-reported user enjoyment responses in human-robot interaction research, with future potential application for autonomous detection of user enjoyment in real-time in robots and agents for adapting conversations contingently to provide enjoyable and long-lasting interactions.

The dataset consists of 25 older adults' (12 men, 13 women) open-domain dialogue with an autonomous companion robot with an integrated large language model (GPT-3.5, text-davinci-003) from participatory design workshops conducted in March 2023. The conversations are annotated for user enjoyment based on HRI CUES by 3 expert annotators, as described in the paper (arXiv:2405.01354). Robot architecture and participatory design workshops are described in DOI: 10.21203/rs.3.rs-2884789/v1.

Exchanges file contains the participant ID, the number of the turn (conversation exchange by Robot-Participant response), the start and end of the turn, the anonymized transcript for the turn, and three annotator scores for the user enjoyment in the exchange. 

Overall file contains the participant ID, self-reported user perception scores from the questionnaire ("I was satisfied with my conversation with the robot", "It was fun talking to the robot", "The conversation with the robot was interesting", "It felt strange talking to the robot") and three annotator scores for the user enjoyment in the overall interaction.

The conversations are in Swedish. Participants' mean age is 74.6 (SD=5.8). 20 participants had no prior interaction with a robot, and only one had previously talked with a robot. The average interaction duration is 7.4 min (SD=1.5) with 12 to 29 turns. Each turn lasts 5 to 61 seconds (M=17.7, SD=7.2). The total duration of the interactions is 174 min, corresponding to 590 turns. 

Videos of the interactions are available upon request, contingent upon a signed agreement to maintain data confidentiality in accordance with GDPR regulations.

Anonymization macros:

[P_NAME]: Participant's name (may include surname). The robot always uses the first name even when the surname is given.

[NAME_REMOVED]: A name of another person mentioned by the participant.

[LOCATION_REMOVED]: Small town/village/area where the participant lives or lived.

[MEDICAL_INFO_REMOVED]: Medical information shared by the participant.

[AGE_REMOVED]: Participant's or other person's age.

[INFORMATION_REMOVED]: Sensitive information shared by the participant.

[MISTAKEN_NAME]: Speech recognition error resulted in the name being misunderstood.