Maximizing total yield in safety hazard monitoring of online reviews

Many firms face challenges in monitoring their products for evidence of safety hazards, which can have pro-foundly negative effects both on consumers and on firms' financial standings. Monitoring online reviews has the potential to unveil vital safety insights, but the volume of the data poses practical challenge. A popular recent approach to safety surveillance has involved utilizing smoke terms customized to the safety-related language in each product category. In this work, we suggest two new methods for deriving smoke terms: piecewise smoke terms and sum of ranks smoke terms. In addition, we address the question of whether smoke terms can be trained to perform well across product categories. We find that our new smoke term methods show improvement in performance upon traditional approaches, such as sentiment analysis, for detecting mentions of safety hazards. We find that piecewise smoke terms in particular performed well across all multiple product categories. We also found that while it is difficult for a cross-category technique to match the performance of a category-specific technique, our novel set of cross-category smoke terms performed with great promise and offer substantial improvement upon sentiment analysis.

Automated summary

Many firms struggle with monitoring product safety due to the potential negative impacts on consumers and financial standings. Monitoring online reviews can provide important safety insights, but the large volume of data poses practical challenges. This study proposes two new methods for identifying safety hazards, which show improvement over traditional approaches and demonstrate promise for cross-category analysis.