Online product reviews play important roles in the word-of-mouth marketing of e-commerce enterprises, but only helpful reviews actually influence customers’ purchase decisions. Current research focuses on how to predict the helpfulness of a review but lacks a thorough analysis of why it is helpful. In this paper, feature sets covering review text and context cues are firstly proposed to represent review helpfulness. Then, a set of gradient boosted trees (GBT) models is introduced, and the optimal one, which as implemented in eXtreme Gradient Boosting (XGBoost), is chosen to predict and explain review helpfulness. Specially, by including the SHAP (Shapley) values method to quantify feature contribution, this paper presents an integrated framework to better interpret why a review is helpful at both the macro and micro levels. Based on real data from Amazon.cn, this paper reveals that the number of words contributes the most to the helpfulness of reviews on headsets and is interactively influenced by features like the number of sentences or feature frequency, while feature frequency contributes the most to the helpfulness of facial cleanser reviews and is interactively influenced by the number of adjectives used in the review or the review’s entropy. Both datasets show that individual feature contributions vary from review to review, and individual joint contributions gradually decrease with the increase of feature values.

In the e-commerce environment, it is very common for consumers to select goods or services based on online reviews from social platforms. However, the behavior of some unscrupulous merchants who hire a “water army” to brush up on reviews of their products has been continuously exposed, which seriously misleads consumers’ purchasing decisions and undermines consumer trust. Until now, it has been a challenging task to accurately detect the “water army”, who could easily alter their behaviors or writing styles. The focus of this paper is on some collusive clues between members of the same social platform to propose a new graph model to detect the “water army”. First is the extraction of six kinds of user collusive relationships from two aspects: user content and user behavior. Further, the use of three aggregation methods on such collusive relationships generates a user collusive relationship factor (CRF), which is then used as the edge weight value in our graph-based water army detection model. In the combination of the graph grouping method and evaluation rules on candidate subgraphs, the graph model effectively detects multiple collusive groups automatically. The experimental results based on the Mafengwo platform show that the CRF generated from the coefficient of variation (CV) method demonstrates the best performance in detecting collusive groups, which provides some practical reference for the detection of “water armies” in an e-commerce environment.