Method for detecting deceptive e-commerce reviews based on sentiment-topic joint probability
Provided is a method for detecting deceptive e-commerce reviews based on a sentiment-topic joint probability, which belongs to the fields of natural language processing, data mining and machine learning. In the data of different fields, a STM model is superior to other reference models; compared with other models, the STM model belongs to a completely un-supervised (no label information) statistic learning method and shows great advantages in processing unbalanced large sample dataset. Thus, the STM model is more suitable for application in a real e-commerce environment.
1. A method for detecting deceptive e-commerce reviews based on a sentiment-topic joint probability, wherein
A STM model is a sentiment-topic joint probability model which is a 9-tuple, STM=(α, β, μ, , , , z m,n , s m,n , w m,n ), wherein:
α is a hyper parameter that reflects a relative strength hidden between topic and sentiment;
μ is a hyper parameter that reflects a sentiment probability distribution over topic;
β is a hyper parameter that reflects a word probability distribution;
is a K-dimensional Dirichlet random variable, which is a topic probability distribution matrix;
is a K*T-dimensional Dirichlet random variable, which is a sentiment probability distribution matrix;
is a K*T*N-dimensional Dirichlet random variable, which is a word probability distribution matrix;
z m,n is a topic to which the n-th word of a document m belongs;
s m,n is a sentiment to which the n-th word of the document m belongs;
w m,n is a basic unit of discrete data, which is defined as a word indexed by n in the document m;
the method for detecting deceptive e-commerce reviews based on a sentiment-topic joint probability comprising the following steps:
at step 1: initializing the hyper parameters α, β, μ of the STM model;
at step 2: setting the appropriate numbers of the topic and the sentiment, and maximum iterations of Gibbs sampling;
at step 3: training the STM model until the model stabilizes and converges;
at step 4: inputting the sentiment probability distribution matrix calculated by the STM model as a feature into a classifier for training;
at step 5: inputting new unlabeled samples into the STM model and training the STM model to calculate the sentiment probability distribution matrix of the new unlabeled samples as the feature;
at step 6: inputting the sentiment probability distribution matrix of the new unlabeled samples into the trained classifier for prediction; and
at step 7: outputting new sample labels through the classifier.