Two-sided receptivity to conversational AI agents in online dating: Bilingual survey data from Fledge.Love
Using a two-dimensional graded response model on survey data (N=2499), the study quantifies user receptivity to conversational AI in online dating, revealing high correlation between deployment and encounter roles.
Key Findings
Methodology
This study employs a two-dimensional graded response model (GRM) to measure user receptivity towards autonomous conversational agents, differentiating between deploying one's own agent (principal role) and encountering others' agents (counterpart role). Data from two surveys (N=2617 and N=2894) include ordinal items and covariates, processed through anonymization, time coarsening, and row shuffling. Model estimation uses maximum likelihood with Gauss-Hermite quadrature, capturing latent dimensions (θ_send and θ_recv) with a correlation of 0.92. Cross-language measurement invariance is tested via differential item functioning (DIF). The analysis supports the model's robustness, with model fit indices (ΔBIC=51.8) and predictive validation (AUC=0.89). Latent class analysis identifies four user preference groups, indicating heterogeneity in acceptance patterns.
Key Results
- The model reveals a strong correlation (0.92) between acceptance in deploying and encountering roles, with all items showing high discrimination (1.72–4.20). Non-invariant items (Y4, Y5) exhibit language and gender differences, highlighting cultural measurement challenges. The latent class analysis uncovers four distinct preference groups, with stable class structures across samples. Predictive validation confirms the model's internal consistency, with Y4 endorsement prediction AUC of 0.89. Parameter stability analysis via bootstrap demonstrates robustness, supporting practical deployment.
- Model fit indices and DIF tests indicate that the measurement instrument reliably captures the two roles, though some items show partial invariance. The model's ability to distinguish user segments suggests potential for personalized AI interaction strategies. The high correlation between dimensions underscores the intertwined nature of deployment and reception attitudes, while the latent classes reveal diverse user profiles, informing tailored recommendation systems.
- Overall, the findings provide a detailed quantitative understanding of user attitudes towards conversational AI, with implications for platform design, cross-cultural adaptation, and ethical deployment. The robustness and predictive power of the model make it a valuable tool for future research and industry applications.
Significance
This research offers a pioneering quantitative framework for understanding user acceptance of conversational AI in online dating, emphasizing the importance of dual-role measurement. It bridges gaps in existing literature by providing micro-level, cross-cultural data, enabling more nuanced AI deployment strategies. The methodological rigor—combining graded response modeling, invariance testing, and latent class analysis—sets a new standard for measurement in human-AI interaction studies. Practically, the results guide platform developers in designing AI features that align with user preferences, fostering trust and engagement. The privacy-preserving anonymization pipeline demonstrates responsible data sharing, encouraging open science. Overall, this work advances both theoretical understanding and practical implementation of AI acceptance in diverse cultural contexts.
Technical Contribution
The study introduces a novel application of a two-dimensional graded response model to measure dual roles in AI acceptance, integrating Bayesian latent regression with multi-item ordinal data. The approach supports cross-cultural invariance testing through DIF analysis, identifying non-invariant items and enabling partial invariance modeling. The pipeline includes rigorous anonymization with k-anonymity, ensuring data privacy. The model's latent dimensions are validated via bootstrap and latent class analysis, revealing stable user segments. This comprehensive framework enhances measurement precision and interpretability, offering a replicable template for future human-AI interaction research, especially in multilingual, multicultural settings.
Novelty
This is the first study to simultaneously measure user receptivity to deploying and encountering conversational AI in a real-world online dating context using a two-dimensional graded response model. It uniquely combines cross-language measurement invariance testing with latent class segmentation, revealing heterogeneity in user preferences. The integration of privacy-preserving data sharing with advanced psychometric modeling represents a significant methodological innovation, setting new standards for microdata sharing in AI acceptance research. These contributions collectively push the frontier of nuanced, scalable measurement of human-AI interaction attitudes.
Limitations
- Sample bias towards active platform users may limit generalizability to broader populations. The cross-sectional design cannot capture dynamic changes in attitudes over time. Some items exhibit partial invariance across languages and genders, complicating cross-cultural comparisons. The model assumes linear relationships between latent traits and responses, potentially oversimplifying complex behaviors. Future work should include longitudinal data, broader samples, and refined measurement invariance techniques to address these issues.
Future Work
Future research should expand to longitudinal studies tracking attitude changes, incorporate behavioral data for validation, and explore adaptive measurement models. Developing more invariant items and refining cross-cultural equivalence will enhance comparability. Integrating experimental interventions could test causal effects of AI features on user acceptance. Additionally, extending the framework to other platforms and languages will broaden applicability. Ethical considerations, such as privacy and transparency, must remain central as AI deployment scales. These directions will deepen understanding and improve AI integration in human social contexts.
AI Executive Summary
The rapid integration of generative AI and autonomous conversational agents into online dating platforms has transformed digital romance, yet understanding user acceptance remains a challenge. Existing surveys often treat AI acceptance as a monolithic construct, ignoring the nuanced roles users play—either deploying agents or encountering them. This study addresses this gap by analyzing survey data from Fledge.Love, an international dating platform, using a sophisticated two-dimensional graded response model. The model distinguishes between users’ willingness to deploy their own AI agents and their reactions to encountering others’ agents, revealing a high correlation (0.92) and strong discriminative capacity across items.
The research involved two large-scale surveys—one in Russian and English, with 2617 responses, and another with 2894 responses—covering diverse demographic groups. Data preprocessing included anonymization, time coarsening, and row shuffling, ensuring privacy compliance. The core analytical method, the graded response model, estimates latent dimensions representing deployment and reception receptivity. Model fit indices (ΔBIC=51.8) and predictive validation (AUC=0.89) confirm robustness. Differential item functioning tests identified some non-invariant items (Y4, Y5), highlighting cultural measurement challenges. Latent class analysis uncovered four distinct user segments, indicating heterogeneity in AI acceptance.
These findings have significant implications for designing AI features that cater to diverse user preferences, especially in cross-cultural contexts. The model’s predictive accuracy and stability support its use in personalized recommendation systems and user behavior analysis. The study also emphasizes privacy, employing a rigorous anonymization pipeline aligned with k-anonymity standards, facilitating open data sharing. Overall, this work advances the quantitative understanding of human-AI interaction, providing a foundation for more inclusive, trustworthy AI deployment in social platforms. Future directions include longitudinal studies, behavioral validation, and expanding to other cultural settings, aiming to foster broader acceptance and responsible AI integration.
Deep Dive
Abstract
Autonomous conversational agents and generative-AI features are being added to online dating platforms faster than public evidence about user attitudes can accumulate, and the scarcest evidence concerns the receiving side: how people react when the profiles, messages, or conversation partners they encounter are machine-generated. We release two anonymized survey datasets collected from active users of Fledge.Love, a dating platform serving an international user base. The first (N = 2,617; Russian and English forms) measures receptivity to autonomous conversational agents with a seven-item battery that separates the principal role (deploying one's own agent) from the counterpart role (encountering someone else's), plus six ordinal covariates and two auxiliary items. The second (N = 2,894) measures interest in three passive generative-AI features. The release includes model-derived scores for 2,499 complete cases, a bilingual codebook, a documented anonymization pipeline with a k-anonymity audit, executable analysis notebooks, and canonical outputs, supporting reuse in human-AI communication, recommender-systems, and cross-cultural technology-acceptance research.