The Research On Evaluating Topic Models

Topic models not only can help our researchers in-depth study in the field of machine learning, can help us solve many problems in the real world. This content relates to three kinds of topic models including LDA, CTM and OnlineLDA, then three kinds of topic models, which model is more popular?The main content of research in this thesis is the method of evaluating topic models, proposed by different methods to evaluate topic models. Firstly, evaluateing topic models by behavioral experiments, before each experiment,run each topic model,and we will get topic model word cloud, the cloud according to the probability of the corresponding vocab-ulary words, with different size to be displayed, the probability of the emergence of large-size high, so that the probability of subjects clearly see the big keywords, the133experi-menters were tested, with the option to select123to choose their subjective view, and then analyze the evaluation of the optimum topic models. Because of this behavior experiments, subjective, entirely depends on the choice of the experimenters, then in objective sense, the experimenters how to evaluate the theme of them? This has become a functional magnetic resonance experiments in the field of brain research. Brain functional magnetic resonance imaging experiment, this experiment can objectively evaluate the merits of the topic models, so the experimenters look at the topic models and determine the topic models, the reaction they made in their brains. After two experiments, the research allows us to want to get more datas,through the behavioral and brain experiments, experimenters are in limited number, if we have different educational backgrounds, different ages of people do the same options, whether the results will be different from them? We introduce micro-channel public plat-form that has a very large user groups, this article build such a public platform to encourage more people to participate and get results to be more significance in statistical.