Background: Questionnaires have been used in the past 2 decades to predict the diagnosis of vertigo and assist clinical decision-making. A questionnaire-based machine learning model is expected to improve the efficiency of diagnosis of vestibular disorders. Objective: This study aims to develop and validate a questionnaire-based machine learning model that predicts the diagnosis of vertigo. Methods: In this multicenter prospective study, patients presenting with vertigo entered a consecutive cohort at their first visit to the ENT and vertigo clinics of 7 tertiary referral centers from August 2019 to March 2021, with a follow-up period of 2 months. All participants completed a diagnostic questionnaire after eligibility screening. Patients who received only 1 final diagnosis by their treating specialists for their primary complaint were included in model development and validation. The data of patients enrolled before February 1, 2021 were used for modeling and cross-validation, while patients enrolled afterward entered external validation. Results: A total of 1693 patients were enrolled, with a response rate of 96.2% (1693/1760). The median age was 51 (IQR 38-61) years, with 991 (58.5%) females; 1041 (61.5%) patients received the final diagnosis during the study period. Among them, 928 (54.8%) patients were included in model development and validation, and 113 (6.7%) patients who enrolled later were used as a test set for external validation. They were classified into 5 diagnostic categories. We compared 9 candidate machine learning methods, and the recalibrated model of light gradient boosting machine achieved the best performance, with an area under the curve of 0.937 (95% CI 0.917-0.962) in cross-validation and 0.954 (95% CI 0.944-0.967) in external validation. Conclusions: The questionnaire-based light gradient boosting machine was able to predict common vestibular disorders and assist decision-making in ENT and vertigo clinics. Further studies with a larger sample size and the participation of neurologists will help assess the generalization and robustness of this machine learning method.
This paper examines how game developers can effectively design player-versus-player (PvP) games while balancing revenue and player engagement. In many PvP games, the ability to purchase powerful items creates a “pay-to-win” environment, often frustrating nonpaying players. To address this issue, this paper proposes a tournament design model that allows developers to create different game versions with varying advantages, influencing players’ chances of winning. The findings indicate that this approach enhances developers’ pricing flexibility and mitigates competition between versions. By offering multiple free versions alongside paid options, developers can engage players who may be unwilling to spend money, thereby translating their willingness to play into other players’ willingness to pay. Furthermore, the model shows that tournament design can significantly increase profits, particularly among players with limited budgets. Ultimately, this strategy benefits both developers and players, fostering a win-win outcome for the gaming community. This work elucidates the rationale of the freemium model that offers differentiated free versions in games with PvP battles.
Generative AI tools can transform software artifacts so extensively that downstream use remains technically valuable while the connection to the source becomes difficult to verify. This creates a governance problem that conventional software-provenance systems do not fully address: they can authenticate a project's declared repository and build process, but they cannot establish that the declared history includes every contributing predecessor. We develop a game-theoretic model in which an original project chooses how much reusable knowledge to disclose and a later project decides whether to credit the source or claim independent origin. When the originator's return from openness depends strongly on visible credit, it limits sharing to the maximum safe disclosure-the largest amount that does not induce a false origin claim-and may close entirely. When broad use remains valuable without credit, the originator can remain open at a lower disclosure level even as reuse is concealed. We then study a governance rule that combines a signed origin claim, independent records, and a bounded penalty. Evidence and penalties reinforce one another, and AI creates a countervailing effect: it reduces disclosure by weakening code-based detection, yet can raise the marginal value of process records that do not rely on final-code similarity. Incomplete or overlapping records, limited penalties, and insufficient platform investment create a governance ceiling. Our results also show how protecting visible lineage can sustain voluntary disclosure in open-source ecosystems.
Data science contests enable seekers to crowdsource predictive and algorithmic solutions from heterogeneous solvers, but releasing training data often raises the need to protect sensitive information, such as user-side privacy and firm-side proprietary business information. A common response is to poison the released data by injecting noise. Prior work predominantly treats such poisoning as a data-protection necessity that inherently degrades data utility. We reconceptualize data poisoning as a strategic contest-design lever to manage solvers' incentives. We develop a model wherein a seeker chooses the noise scale and two heterogeneous solvers choose their effort. Data poisoning generates a fundamental tradeoff: it worsens data quality but levels the playing field and stimulates effort. We establish a moderate-asymmetry principle: poisoning is optimal only when solvers' expertise asymmetry is moderate-large enough to warrant leveling the playing field, but not so large that the requisite noise destroys the data's intrinsic value. Furthermore, we identify a complementarity between data poisoning and prize design. Strategically injecting noise restores competitive pressure, optimally inducing the seeker to augment the prize purse and potentially achieving a Pareto improvement that strictly benefits all parties. Finally, extending the analysis to a multi-solver environment reveals that optimal poisoning depends crucially on expertise skewness in addition to the dispersion. Overall, our results recast data poisoning from a defensive data-protection instrument into a proactive economic design choice.
Streamer emotion is central to audience engagement, but it may invite incivility when it appears inappropriate for the content being communicated. We conceptualize such inappropriateness as inconsistency between the valence of a streamer's textual content and nonverbal emotional signals. Drawing primarily on Emotions as Social Information (EASI) theory, we argue that viewers use cross-channel fit to infer whether a streamer's reaction is appropriate, with this inference varying by mismatch direction and avatar artificiality. We develop a temporally aligned face-voice-text analytics framework that retains cue source and mismatch direction and links each emotional configuration to audience responses in the next synchronized window. Applying the framework to 21,654 ten-second windows from Bilibili virtual livestreams, we find that greater nonverbal-textual inconsistency predicts a higher subsequent rate of incivility in danmaku comments. The association is concentrated in misplaced positivity rather than misplaced negativity. As avatar artificiality increases, the face-text association weakens whereas the voice-text association strengthens. The findings identify contextual fit as a boundary condition for the effects of streamer emotion and show how avatar mediation changes the diagnostic value of emotional channels.