Abstract:

How to make full use of the sentiment information in online word-of-mouth and consumers’ personalized preferences to achieve online merchant ranking with high accuracy is of great practical significance and theoretical value to improve the efficiency of consumers’ decision making in massive information. In this paper, a ranking method is proposed based on deep learning that considers fuzzy sentiment, personalized preferences, incorporates online reviews, ratings, and popularity. First, an aspect-level sentiment classification model based on dual-attention BILSTM (Bi-directional long short-term memory) is constructed to identify the sentiment orientation of online reviews. Then, the sentiment value is converted to IFV (intuitionistic fuzzy values), and the intuitionistic fuzzy TOPSIS (technique for order preference by similarity to an ideal solution) method is used to calculate the closeness degree. After that, the attribute weighted score and popularity value are calculated based on consumer ratings and the number of reviews. Finally, the ranking of alternative restaurants is determined by integrating the three factors. Comparative experiments with standard datasets show that the dual-attention BILSTM model outperforms traditional models such as LR, SVM, BILSTM, and single-attention BILSTM models. The ranking results of eight restaurants in the example analysis have higher average overlap scores with the platform ranking, which indicates the effectiveness of the merchant ranking method proposed in this paper.

Key words: merchant ranking, consumption decision-making, online reviews, intuitionistic fuzzy sets (IFS), aspect-based sentiment analysis

摘要：

如何充分利用在线口碑中的情感信息和消费者的个性化偏好，实现具有较高准确性的在线商家排序，对于提高消费者在海量信息中的决策效率具有重要的现实意义和理论价值。提出了一种基于深度学习的考虑模糊情感、个性化偏好，并融合在线评论、评分及人气的排序方法。首先构建双注意力BILSTM方面级情感分类模型识别在线评论情感，将情感值转换为直觉模糊值，应用直觉模糊TOPSIS方法计算贴近度，然后结合消费者评分、商家人气度量确定排序。以标准数据集进行对比实验，表明双注意力BILSTM模型优于LR、SVM等传统模型以及BILSTM和单注意力BILSTM模型；实例分析中大众点评8家餐厅的排序结果与平台排序平均重叠分数较高，说明本文所提出的商家排序方法的有效性。

关键词: 商家排序, 消费决策, 在线评论, 直觉模糊集, 方面级情感分析