Advertisement overflow detection method based on multi-dimensional Gaussian mixture model

The invention relates to the technical field of advertisement note data detection, and particularly discloses an advertisement overflow detection method based on a multi-dimensional Gaussian mixture model, which comprises the following steps of: firstly, establishing the multi-dimensional Gaussian mixture model, listing a probability density function model of the multi-dimensional Gaussian mixture model of a sample note, and establishing a likelihood function to estimate model parameters; secondly, logarithmizing a likelihood function, introducing a hidden variable representing a Gaussian sub-model to which a sample note belongs, and calculating a posterior probability in combination with a Bayesian formula; judging whether the multi-dimensional Gaussian mixture model completes convergence under the current model parameter condition or not, and if not, continuing to update the model parameters through the posterior probability until convergence; and finally, clustering of all the notes is completed by the con