Search for: All records

Auditing is critical to ensuring the fairness and reliability of decision-making systems. However, auditing a black-box system for bias can be challenging due to the lack of transparency in the model’s internal workings. In many web applications, such as Yelp, it is challenging, if not impossible, to manipulate their inputs systematically to identify bias in the output. Yelp connects users and businesses, where users identify new businesses and simultaneously express their experiences through reviews. Yelp recommendation software moderates user-provided content by categorizing it into recommended and not-recommended sections. The recommended reviews, among other attributes, are used by Yelp’s ranking algorithm to rank businesses in a neighborhood. Due to Yelp’s substantial popularity and its high impact on local businesses’ success, understanding the bias of its algorithms is crucial.This data-driven study, for the first time, investigates the bias of Yelp’s business ranking and review recommendation system. We examine three hypotheses to assess if Yelp’s recommendation software shows bias against reviews of less established users with fewer friends and reviews and if Yelp’s business ranking algorithm shows bias against restaurants located in specific neighborhoods, particularly in hotspot regions, with specific demographic compositions. Our findings show that reviews of less-established users are disproportionately categorized as not-recommended. We also find a positive association between restaurants’ location in hotspot regions and their average exposure. Furthermore, we observed some cases of severe disparity bias in cities where the hotspots are in neighborhoods with less demographic diversity or higher affluence and education levels. 

Abstract—Innovation is a fundamental driver of economic growth, regional competitiveness, and societal well-being. Nonetheless, its benefits remain unevenly distributed, with many regions, so called “innovation deserts,” struggling to translate potential into productivity due to systemic barriers and limited insight into the drivers of innovation. Motivated by this disparity, this article presents a novel data-driven framework for analyzing and benchmarking innovation productivity across the U.S. counties, identifying key determinants that drive regional innovation performance. The study covers over 3200 U.S. counties and includes 108 attributes spanning innovation outputs, demographics, education levels, STEM participation, internet connectivity, business sizes, and economic indicators. The framework comprises three integrated modules: 1) a hybrid feature selection mechanism combining random forest importance scores with mutual information criteria to identify influential factors; 2) a nonlinear dimensionality reduction module employing kernel-based principal component analysis to mitigate the curse of dimensionality; and 3) a particle swarm optimized XGBoost model (PSO-XGBoost) enabling sensitivity analysis and scenario-based forecasting. The framework demonstrates superior predictive performance (R2 > 0.81) compared to five other state-of-the art boosting algorithms. The analysis uncovers nonlinear relationships between key attributes and innovation outcomes, facilitating benchmarking of innovation productivity and quantifying targeted interventions to foster inclusive innovation ecosystems. The proposed framework offers policymakers interpretable, actionable insights for evidence-based regional innovation development. In particular, this study contributes a novel framework to identify innovation deserts and benchmark regional performance at a national scale. 

This article presents a data-driven framework for policy-oriented benchmarking and catalyzation of innovation productivity across 65 U.S. metropolitan areas. The study achieves a high level of research rigor by integrating multiple complementary analytical modules and systematically validating results through robust statistical and diagnostic tests. Specifically, the methodological design synthesizes three components: First, a multifeature selection pipeline that combines random forest, select K-best, and recursive feature elimination to ensure statistically reliable identification of innovation determinants; second, kernel principal component analysis with parameterized kernel functions optimized to capture complex nonlinear interdependencies among innovation factors; and third, a particle swarm optimization-enhanced gradient boosting machine that delivers exceptional predictive accuracy (R2 = 0.984, RMSE = 358.0) while demonstrating minimal overfitting (R2 differential between training and testing = 0.016). Rigor is further reinforced through systematic residual analysis and comprehensive sensitivity analysis, which together provide robust validation of the framework’s reliability. These diagnostics reveal significant regional disparities in innovation performance relative to model predictions, with striking counterintuitive results. Several metropolitan areas substantially outperform expectations through strategic ecosystem alignment and policy coherence, while others exhibit considerable innovation deficits despite apparent structural advantages. Sensitivity analysis identifies STEM education infrastructure as the most influential driver of innovation, challenging conventional policy assumptions. The empirically validated framework equips engineering managers and policymakers with actionable, quantitative insights for designing targeted interventions, allocating resources effectively, and transforming under-performing regions into resilient innovation ecosystems through evidence-based strategies. 

We are being constantly judged by automated decision systems that have been widely criticised for being discriminatory and unfair. Since an algorithm is only as good as the data it works with, biases in the data can significantly amplify unfairness issues. In this paper, we take initial steps towards integrating fairness conditions into database query processing and data management systems. Specifically, we focus on selection bias in range queries. We formally define the problem of fairness-aware range queries as obtaining a fair query which is most similar to the user's query. We propose a sub-linear time algorithm for single-predicate range queries and efficient algorithms for multi-predicate range queries. Our empirical evaluation on real and synthetic datasets confirms the effectiveness and efficiency of our proposal.