| People are dazzled by the endless array of beautiful products in online shops, and many people are used to commenting upon what they purchased,which results in information overload. It is very necessary for the online shop owners to dig and show valuable information from the huge amount of data efficiently in pursuit of recognition from consumers. Meanwhile,as the safety and portability performance of mobile payment improves dramatically, consumers are diverting to online shopping utilizing mobile devices in a large scale. Considering the constraints of the size of mobile devices and the real payment scenarios, consumers prefer collecting more useful information from fewer product reviews. Therefore, it is a popular topic to study how to summarize the primary reviews sets by a brief and accurate reviews sets.This dissertation will discuss the model of online review validity based on big data platform in depth, referring to previous studies. The main works and features of this dissertation are as follows.It will firstly demonstrate a Balanced Review Choosing algorithm. To begin with, it analyses the features of reviews with a high value. Hence, it leads to the Balanced Review Choosing algorithm based on an Unsupervised Topic Opposite and Sentiment Unification (UTOSU) model.Secondly, it designs UTOSU model based on LDA. UTOSU is a Topic Sentiment mixture model, which suggests that each word not only defines the review theme distribution, but also affects sentiment distribution. It assumes that every word in a single sentence suggests the same or corresponding theme. Therefore, it samples the theme tag from sentences and sentiment tag from words. Each abstract review created is probability distribution and sentiment distribution, the theme of which can choose either positive or negative sentiment.This dissertation verifies the UTOSU model and the Balanced ReviewChoosing algorithm and the result proves that the model and algorithm mentioned above are feasible and valid. |
PDF Full Text Request