Sorting Online Reviews of Restaurant Merchants Based on Narrativity

Abstract

With the rapid development of e-commerce, online reviews have become the core reference information for consumers to make purchase decisions. Traditional review sorting methods mostly rely on the number of likes, timeliness and other factors, ignoring the textual characteristics of the reviews themselves. In this paper, we innovatively propose to take the "narrativity" of reviews as an important dimension of review sorting. We present a measurement method for "narrativity" based on NLP, and combine it with other review metrics (e.g., timeliness, text length, rating deviation, user identity, etc.) to build a comprehensive review sorting model. We crawled 52,500 reviews from 500 stores on the Dianping platform and trained a classification model based on the transformer architecture to categorize these reviews into high-value and low-value groups. Subsequently, we developed a narrative-based sorting model to reorder the reviews of these stores, and observing whether the ranks of high-value reviews have improved relative to the original order. The experimental results reveal that more high-value reviews have been prioritized to the forefront, leading to an overall improvement of 39.4% compared to the original sorting. Our experimental demonstrated that the narrative-based sorting model can optimize the review sorting, so that high-value reviews can get more exposure and improve consumers' shopping experience.