Abstract: Topic model is an effective modeling method for extracting implied themes in large-scale text. In this paper, Latent Dirichlet Allocation(LDA) topic model is introduced to patent content analysis for patent theme extraction, which solves the problems in previous patent subject classification, such problems as too rough classifications, lack of time-effectiveness and scientific nature, etc. And then, based on the original LDA model, an extended institution-topic model is developed in this paper. By joint modeling of patent subject and object, the internal relationships between patent themes and corresponding institutions is identified. Finally, a case study is carried out in the LTE technology of the telecommunication industry, and it is found out that these models can be used effectively for patent subject classification and identifying competition situation of patent subjects under each corresponding theme.

Key words: topic model (LDA), LDA institution-topic model, patent content analysis, LTE