While scholars and managers have long queried how a review message can be useful, the e-word-of-mouth (eWOM) literature yields inconclusive findings. Our study approaches this classic research question from the novel angle of diversity and investigate how topic diversity of online reviews affect review usefulness. Specifically, topic diversity manifests in both topic variety, the extent to which different topics are discussed in a review message, and content distribution across topics, the extent to which content is distributed across different topics. We leverage the aspect-based topic modeling technique to uncover the extent of different topics discussed in individual review messages as the nested subunits and operationalize topic variety and content distribution as two main unit-level predictors of review usefulness. This text-based approach helps analyze about 200k restaurant reviews from Yelp.com and generates six different topics relating to restaurant reviews. Results reveal that (1) on average, topic variety positively affected review usefulness, while content distribution across topics did not; (2) however, when topic variety is high (low), the impact of content distribution on review usefulness is significant and positive (significant and negative). Our study contributes to the e-WOM literature by offering new knowledge regarding topic diversity and review usefulness.

Health infomediaries are online interactive platforms where patients learn about diseases, procedures, and treatments, but how the learning dynamics evolve to help patients make decisions is unknown. Investigating patient learning in health information is essential to enabling patient empowerment using these platforms. In this study, we theoretically anchor patients’ learning processes in a health infomediary to Nonaka-Takeuchi’s Socialization, Externalization, Combination, and Internalization (SECI) learning dynamics model, with the infomediary as the ba or collaborative space for learning. We posit that questions reflect externalization for a focal learning patient, while reviews and comments reflect socialization, driving patients’ knowledge internalization. Learning with more information aids patients in health-relevant choices and actions such as a doctor consultation in a health infomediary. Three hypotheses about continued learning choices are presented and tested using data from 8,900 reconstructive surgery health infomediary users, using a multichoice mixed logit model with panel estimations. Findings show patients continue to review and comment more than questions to learn about trendy surgeries, which is counterintuitive to existing insights, whereas questions lead to learning sentiment-laden information. Exploratory analyses suggest that learning through reviews regarding post-procedural services may deter, but other learning modes motivate patients to consult a doctor. We discuss the theoretical contributions and implications for infomediary design and research.

Saliency detection that utilizes deep convolutional neural networks to obtain high level features from original images has achieved considerable progress during the past years. However, few methods consider learning saliency cues from hand-crafted features. In this paper, we demonstrate that deep learning can produce good enough saliency detection results using only hand-crafted features. We propose a novel multi-context deep learning saliency detection algorithm, where only hand-crafted features are taken into account and modeled in a unified deep learning framework. Extensive experiments on benchmark datasets indicate significant and consistent improvements over the representative deep learning framework based saliency detection methods.