To Save Mobile Crowdsourcing from Cheap-talk: A Game Theoretic Learning Approach

TL;DR

This work is the first to study how to save mobile crowdsourcing from cheap-talk and strategically learn from biased users’ reviews as a dynamic Bayesian game, including users’ service-type messaging and the platform's follow-up rating/inference.

Abstract

Today mobile crowdsourcing platforms invite users to provide anonymous reviews about service experiences, yet many reviews are found biased to be extremely positive or negative. The existing methods find it difficult to learn from biased reviews to infer the actual service state, as the state can also be extreme and the platform cannot verify the truthfulness of reviews immediately. Further, reviewers can hide their (positive or negative) bias types and proactively adjust their anonymous reviews against the platform's inference. To our best knowledge, we are the first to study how to save mobile crowdsourcing from cheap-talk and strategically learn from biased users' reviews. We formulate the problem as a dynamic Bayesian game, including users' service-type messaging and the platform's follow-up rating/inference. Our closed-form PBE shows that an extremely-biased user may still honestly message to convince the platform of listening to his review. Such Bayesian game-theoretic learning obviously outperforms the latest common schemes especially when there are multiple diversely-biased users to compete. For the challenging single-user case, we further propose a time-evolving mechanism with the platform's commitment inferences to ensure the biased user's truthful messaging all the time, whose performance improves with more time periods to learn from more historical data.

Figures (6)

• Figure 1: System model on $N$ users' messaging of observed service type $S$$\in$$\{L, H\}$ of service state $\theta$ (e.g., of hotels, restaurants and trips) to a crowdsourcing platform (e.g., TripAdvisor), who takes recommendation action $a(\{m_i\}_{i=1}^N)$ to best infer the actual service state. Here we consider the practice that the actual (continuous) service state $\theta$ cannot be accurately estimated by users, and only its low- or high-quality type is observed by each user.
• Figure 2: The system loss $\bar{L}_{evolving}$ in \ref{['u-r-m']} of the time-evolving mechanism versus the period number $T$ and the mean of low-type PDF $\mu_L$. Here we set $p_H = 0.3$, choose $\phi_H(\theta)$ and $\phi_L(\theta)$ as normal distribution PDFs with $\mu_H$=1 and $\sigma^2_H$=$\sigma^2_L$=1.
• Figure 3: The system loss $\bar{L}_2$ in \ref{['u-r-b-1']} of the benchmark 2 and $\bar{L}_{evolving}$ in \ref{['u-r-m']} of the time-evolving mechanism, versus the high-PDF variance $\sigma_H^2$ and the period number $T$, respectively. Here we set $p_H = 0.3$, choose $\phi_H(\theta)$ and $\phi_L(\theta)$ as normal distribution PDFs with $\mu_H$=1, $\mu_L$=$-$1 and $\sigma^2_L$=1.
• Figure 4: The system loss $\bar{L}_2$ in \ref{['u-r-b-1']} of the benchmark 2 and $\bar{L}_{evolving}$ in \ref{['u-r-m']} of the time-evolving mechanism, versus the period number $T$. Here we set $p_H = 0.3$, choose $\phi_H(\theta)$ and $\phi_L(\theta)$ as Logistic distribution PDFs Logistic($\mu$,$s$) with $\mu_H$=1, $\mu_L$=$-$1, $s_H=\frac{3}{2}$ and $s_L$=1, and Laplace distribution PDFs Laplace($\mu$,$b$) with $\mu_H$=1, $\mu_L$=$-$1, $b_H=\frac{3}{2}$, $b_L$=1.
• Figure 5: The system loss $\bar{L}_{Bayesian}$ in \ref{['N']} of the Bayesian game theoretic learning approach and $\bar{L}_{evolving}$ in \ref{['u-r-m']} of the time-evolving mechanism, versus the period number $T$ and the user number $N$, respectively. Here we set $p_H = 0.3$, choose $\phi_H(\theta)$ and $\phi_L(\theta)$ as normal distribution PDFs with $\mu_H$=1, $\mu_L$=-1, $\sigma^2_H$=$\frac{3}{2}$, and $\sigma^2_L$=1.

Theorems & Definitions (22)

• Definition 1
• Lemma 1
• Lemma 2
• Corollary 1
• Lemma 3
• Proposition 1
• Theorem 1
• Definition 2: The Platform's One-Shot Commitment Mechanism
• Lemma 4
• Definition 3: The Platform's Time-Evolving Commitment Mechanism