Research
1. Novel Information in Restaurant Reviews: The Impact on Helpfulness and Restaurant Check-Ins
Dicle Yagmur Ozdemir, Harpreet Singh, Sumit Sarkar
Abstract: The popularity of electronic word-of-mouth has increased dramatically over the years, and it has become important to understand how online reviews impact relevant stakeholders such as platforms, businesses, and consumers. While prior works have examined the impact of different review characteristics, the impact of novel information in reviews has not received attention. To study how novel information in reviews impacts different stakeholders, we use a dataset from Yelp that provides reviews and other related information for restaurants. We examine how novel information in reviews influences restaurant check-ins and review helpfulness, and how this relationship changes as review volume changes. We show that novel information positively impacts helpfulness of reviews and restaurant check-ins – however, the impact becomes negative when the review volume becomes high. We examine the mechanism behind our findings and show that review helpfulness partially mediates the effect of novel information on restaurant check-ins. The indirect effect (through helpfulness) is moderated by review volume. We also show that the impact of novel information on check-ins changes with restaurant type, with the impact higher for high-priced restaurants.
2. Tell Me Something New: Selecting Novel Online Reviews
Dicle Yagmur Ozdemir
Abstract: Reviews can help consumers decrease the uncertainty regarding the quality of a product or service. However, reading reviews can become time-consuming if there exists a large number of reviews for each product. To decrease the time a reader spends reading enough reviews to make a decision, platforms can identify a subset of informative reviews to present to the reader. Such a subset can help the consumer obtain as much new information as possible in a short time. The goal of this study is to identify a subset of reviews that maximizes the novel information offered to a reader. To solve this problem, a novelty formulation for a review set is presented, and a relevant class of novelty measures is identified. The problem is difficult to solve, and efficient heuristics exploiting the structure of the problem are proposed. Experiments are conducted on restaurant reviews from Yelp to examine the effectiveness of the proposed methods. The results show that the heuristics identify review sets with high amounts of novel information in a computationally efficient way. Heuristics designed for real-time environments are shown to be very scalable while still generating excellent solutions. We also show the proposed approaches can be extended to scenarios preserving the average opinion and aspect coverage in a corpus.
3. Credibility, Novelty, and Helpfulness in Online Reviews
Dicle Yagmur Ozdemir, Harpreet Singh, Sumit Sarkar
Abstract: User-generated contents such as reviews and ratings are very important for online platforms. To better leverage such content, platforms enable users to vote on the helpfulness of reviews. The importance of helpful votes (received by a review) to the platforms, consumers, and reviewers, is substantive. Factors found to impact the helpfulness of a review include, among others, the novelty of the content in the review and the review’s credibility characteristics (i.e., source credibility and rating credibility). To better understand how consumers’ perceptions of review helpfulness are affected by these factors, we investigate the moderating impact of credibility on the influence of review novelty on helpfulness. We find that source credibility and review novelty are substitutes in terms of their contribution to review helpfulness. On the other hand, rating credibility positively moderates the effect of a review’s novelty on its helpfulness and complements review novelty. These findings are very important for the different stakeholders. Platforms would benefit by knowing in a more nuanced manner what is driving helpful votes. This should help them refine their procedures for recognizing elite contributors. From a reviewer’s perspective, a better understanding of these interactions can help them craft reviews that are more likely to receive helpful votes.
4. Axiomatic Characterization of Novelty for Textual Content
Dicle Yagmur Ozdemir, Sumit Sarkar
Abstract: With the rapid increase in digitization, detecting novelty in textual content has become important in many applications. A variety of computational techniques have been developed for this purpose, typically tailored to the context of interest. While these approaches have been shown to work well in their specific contexts, they have been developed using many diverse methodological underpinnings. As a result, there is no consistent underlying conceptual framework applicable to the various methods presented in the literature. The goal of this work is to identify specific requirements that are desirable for novelty measures. We first state assumptions that help characterize the objects (i.e., textual contents or documents) of interest. We then identify a set of axioms for such objects, providing intuitive justifications for them. Existing approaches typically satisfy only a subset of the axioms. To obtain additional insights regarding how the measures may perform for different tasks, we conduct a set of experiments. The experiments compare the performance of some of the commonly used novelty measures. The axioms and the experimental findings can help researchers identify appropriate measures for their task. Importantly, the axioms should also help develop new measures for specific applications.