Disclosure Governance in Repeated Interactions with an Embodied Conversational Agent: Questionnaire, Language, Acoustic, and Interview Data

Description

This dataset accompanies a laboratory-based repeated-interaction study of disclosure governance during communication with Xuefei 2.0, a Mandarin-speaking, photorealistic embodied conversational agent. It includes 252 participants, each of whom completed two interaction rounds, yielding 504 participant-round observations. In each round, participants completed a brief attention-balancing exercise and pre-session questionnaire, followed by an approximately 13-minute structured interaction consisting of a greeting, icebreaker, personally consequential topic discussion, and closing exchange. A post-session questionnaire followed each interaction, and a brief semi-structured interview was conducted after both rounds. The Excel workbook is organized into three participant-linked tables. Table 1 contains demographic information, self-reported AI-literacy scores, response configurations, topic assignments, and questionnaire measures. The measures cover disclosure readiness, disclosure breadth and depth, social-mask lowering, felt epistemic recognition, reflective reappraisal, privacy–interpretive authority delegation, boundary control, evaluation concern, affective states, topic activation, unresolved meaning, and willingness to continue interacting. Table 2 contains derived participant-language and acoustic indicators for both rounds. Language variables describe participant-only production during the icebreaker and topic-discussion stages, including utterance volume, token and type counts, lexical diversity, and transparent dictionary-based indicators. Acoustic variables cover speech share, articulation, pauses, fillers, pitch, loudness, and spectral change when usable audio was available. Table 3 reports interview availability and participant-level consensus codes concerning disclosure experiences, recognition, reflection, interpretive boundaries, authorization, correction, refusal, and interaction conditions. Rows are aligned by participant identifier, with round-specific variables presented in wide format. Supplementary language, acoustic, and interview fields are populated only when the corresponding source material was available and analysable. The workbook contains de-identified analytic variables supporting the reported questionnaire, corpus, acoustic, and interview analyses; raw audio and video are not included.

Steps to reproduce

1) Download the dataset and accompanying analysis scripts. 2) Import the three worksheets and link records using participant_id. 3) Reshape round-specific questionnaire variables into participant-round format. 4) Run the specified questionnaire, language, acoustic, and interview analyses using modality-eligible records. 5) Execute the figure and table scripts to reproduce the reported outputs.

Institutions

• Macau University of Science and Technology

  Macau, Macao

Categories

Licence