Movie Box-office Fraud Detection Based On Anomaly Detection

The Chinese movie market has developed into the second largest movie market in the world.Movie box-office is an important indicator of the popularity of films and it has attracted more and more attention.However,in order to obtain more benefits,box-office fraud has been common in recent years,which has become a major factor hindering the development of Chinese movie market.Box-office fraud detection becomes the top priority for box-office supervision,but there has been no systematic technical study for it.Therefore,in this paper,starting from the original data of box-office,a series of studies around the box-office fraud has been done which has a pioneering significance at the box-office fraud detection.Understanding the characteristics of the data is an important prerequisite for all the work,given that there was no priori analysis of box-office data,therefore,analyzing the box-office data becomes a primary task.Through the analysis of data characteristics,this paper get the rules of box-office data,and lay a solid foundation for box-office fraud detection later.Dealing with the box-office data is an important work in this paper,especially how to extract the useful information that reflects the box-office fraud and convert it into a classifiable form.Because there is no method for reference,this paper puts forward the 'data score' method based on the experience of the film industry and the processing of other data.It converts box-office data into numerical data and sets rules to mark the box-office data.Another focus of this paper is to propose the non-kernel function multi classification QSSVM-OAA model.The model not only exploits the advantages of the traditional Support Vector Machine,but also avoids the selection of the kernel function.Through the experiment on the public datasets,it verifies the excellent performance of the QSSVM-OAA model.Compared with traditional multi classification method,this model has higher classification accuracy and stability.It is more suitable for small training set of dataset and thus it fits the characteristics of movie box-office data.Finally,the transformed labeled box-offic data are input to the QSSVM-OAA model for classification,the result achieved the expected effect of movie box-office fraud detection.This paper has theoretical and scientific basis for the exploration and experiment of movie box-office fraud detection,and the experimental results further illustrate the feasibility of the method.