Internet inquiry is playing an increasingly important role as the complement of the traditional medical service system,especially the similar cases recommendation.It can not only save the patients waiting time,but also make use of the historical resources,for many cases with the same purpose have been solved perfectly.However,because of the diversity and non-standard of the patients descriptions,the inquiry platform cannot find the cases with similar semantic easily.Most traditional retrieval methods require the overlap of two sentences,and this is not suitable with the diversity and non-standard descriptions.In this paper,we try to utilize the sentences semantic representation in a continuous space to understand the cases,and then recommend the similar cases.We also incorporate it into query likelihood language models,trying to get better results.Our experimental data are all collected from a real internet inquiry platform,and the results show that our methods significantly outperform the state-of-the-art translation based methods for similar cases recommendation.