为了同时挖掘商品口碑数据中所谈论的对象、对象的某个方面以及评论者对这个方面的观点,用于指导消费者消费和生产厂家对商品的改进,该文面向口碑数据提出一个无监督对象方面情感联合模型。该模型假设方面分布依赖于对象分布,情感分布依赖于方面分布和对象分布,词是采样的最小单位。在汽车口碑数据上进行了多组实验,实验结果表明:无监督对象方面情感联合模型不仅可以判别文本方面和文本情感的类别,还可以获取文本对象信息。
This paper presents a probabilistic graphical model to simultaneously extract objects, aspects and sentiments from commodity reputation data. The underlying assumption is that the aspect distribution depends on the object distribution, while the sentiment distribution depends on the aspect distribution. The model further assumes that words are the smallest sampling units and is fully unsupervised. Tests on car reputation data show that this model can predict the aspect and sentiment categories of commodity reviews and can simultaneously extract object information from the reviews.