Home

Survey research has a measurement problem it cannot see. A telephone or web survey records the answer and nothing else: not the hesitation before it, not where the respondent was looking, not whether they understood the question. At the same time, conventional face-to-face interviewing is under pressure from attrition, recruitment difficulty and rising cost.

FACES responds to both. It builds a multimodal data space for survey research in which interviews are conducted between avatars in virtual reality or over video, with the full interaction (speech, gaze, posture, facial expression, answers and their timing) captured on a single clock and stored as a linked, relational dataset. The environment is variable by design: avatars, situational parameters, interfaces and AI components can all be manipulated as experimental conditions.

The project is run jointly by the Text Technology Lab at Goethe University Frankfurt and Departments 2 and 3 of the Leibniz Institute for Educational Trajectories (LIfBi) in Bamberg.

Three research questions

FACES asks whether avatar-mediated interviewing is feasible, whether respondents accept it, and whether the resulting data quality holds up.

01

Avatars versus video

What are the advantages of avatar-based interviews compared to video-based interviews in terms of acceptance, feasibility and data quality?

02

Reducing interviewer effects

Which combinations of features reduce interviewer effects, and how do they interact?

03

Virtual interviewer training

How can the results be integrated into a theory for training virtual interviewers?

What has been delivered

Four pieces of open-source software, two completed studies, a reproducible analysis pipeline and two papers. Everything below is public.

Released

InterView

The interview platform. VR and browser clients, a WebRTC media layer, a room server and a logging API, shipped as one Docker Compose stack.

Platform → Released

Va.Si.Li-Lab

The Unity VR framework InterView is built on, together with its backend: scene and role definitions, the Ubiq room server and the logging API.

Framework → Released

Janus-Gateway

A pinned, reproducible build of the Janus WebRTC server, so you can state which software produced a dataset.

Infrastructure → Complete

Avatar pre-study

99 respondents rated 20 avatars on preference, trust, comfort and similarity, to select the interviewer avatars used in the main study.

Study 1 → Complete

VR interview study

27 full interviews conducted in virtual reality across two institutions, with the complete multimodal record of each one.

Study 2 → Complete

Analysis pipeline

Every figure, table and reported number regenerates from cache with one command, and the prose is read back out of the written tables.

Reproducibility → Accepted

ReEmote

Emotion representation in VR through avatar facial expression. Accepted at XR Salento 2026, Springer LNCS.

Publication → Under review

InterView paper

Towards a Unified VR-Capable Interview Environment: the system, its data model, and the evaluation study. Under review at IJHCS (Elsevier); preprint on SSRN.

Publication →

4,567,743

records captured across six linked modalities in 27 interviews (eye, body and head tracking, audio, transcribed speech, questionnaire state and self-report), at a measured median of 18.5 Hz with a 96.6 % capture duty cycle. Not a nominal specification: a measurement, read back out of the stored data.

How it works

An interviewee wears a VR headset and sits across a desk from an avatar in a shared virtual room. The interviewer sits at an ordinary web browser. A live audio and video link connects them, a questionnaire is embedded in the environment itself, and everything that happens is recorded time-aligned into a single relational dataset.

The interview room. A deliberately unremarkable medium-sized room. The panel on the desk carries the answer options for the current question. It is part of the instrument, and it measurably changes how people answer. See Answer format.

The approach runs in three stages. The platform was built first and is now released; the preliminary experiments on avatar characteristics and immersion are complete; validation with former NEPS panel participants is now under way.

Screenshot placeholder
InterView-W, the browser client

Not headset only. Everything above describes the VR side. InterView also runs entirely in an ordinary browser, with no headset required. It was not part of the published evaluation study, but it is the interface the ongoing NEPS validation study is currently being fielded on. See InterView.

Read about the project See the platform