Abstract:

It is an important problem of identifying and quantifying the competition in similar services or products in the research area of competitive relationship mining. A scientific and reasonable evaluation metric of competitive relationship is proposed, and a comprehensive evaluation system of entity competitive relationship is constructed. Dimension reduction and theme extraction on users?? reviews are achieved by using the latent Dirichlet allocation (LDA) model, the similarity function of comments is constructed, and the similarity degree of entity users?? comments is quantified. Based on the geographic location information of entities, the spatial distance of entities is calculated, the adjacent relation of entities is constructed, the distance of entities with adjacent relationship is regarded as the cluster center, and the entities are clustered by using the K-nearest neighbor (KNN) algorithm. The location & topical model (LTM) is proposed by integrating user??s reviews, entity??s geographical attributes, and quantifying the com-petitive relationship between entities. Conducted on a large number of real social network data, the experiments results show that the proposed method has great advantages in quantitative metric formulation, practicability and time performance.

Key words: competitive relationship, relationship quantification, word-of-mouth, geographical location

摘要：

在同类服务或产品中识别和量化竞争是当前竞争关系挖掘领域关注的重要问题。提出科学合理的竞争关系评价指标，构建实体竞争关系综合评价指标体系，使用隐含Dirichlet分布（LDA）模型对消费者口碑评论进行降维和主题提取，构建口碑相似度函数，对实体用户口碑相似度进行量化表示。根据实体地理位置属性，计算实体空间距离，构建实体相邻关系并以具有相邻关系实体的距离作为聚类中心，使用[K]近邻（KNN）算法对其进行聚类。综合上述技术提出LTM模型，融合了用户评论、实体地理位置属性，量化实体间竞争关系。大量真实移动社交网络数据上实验结果表明所提方法在量化指标制定、实用性和时间性能上具有较大优势。

关键词: 竞争关系, 关系量化, 口碑, 地理位置