Roles of topic features in perceived helpfulness of online company reviews

Purpose In this study, we propose a model to forecast the helpfulness of online company reviews and understand the influence of identified topics on this perceived helpfulness. Design/methodology/approach Our approach involves constructing machine learning models to predict the potential helpfulness of the reviews. We performed feature engineering to capture the review topics by employing latent Dirichlet allocation. To identify the factors influencing review helpfulness, we applied an explainable artificial intelligence methodology. We used 649,801 reviews from the JobPlanet website. Findings The light gradient boosting machine outperformed seven alternative models in terms of predictive capability. Furthermore, incorporating topic features significantly enhanced the model performance. Additionally, the overall rating and negative topics related to human relationships, seniors and salaries mentioned in the reviews substantially influenced the perceived helpfulness. Originality/value This study devises effective techniques for extracting variables from company reviews, thereby contributing to the ongoing investigations into identifying the determinants of helpfulness, with a focus on the job seeker perspective.