With the popularity of smartphones, many apps have emerged in the app store for users to download. Most app stores allow users who download an app to post reviews and ratings on this app. These reviews are not only a major factor in determining the ranking of an app but also a major reference for users in choosing whether to download the app and an important way for developers to get feedback from users. However, many meaningless reviews (spam reviews) have severely damaged the normal ecology of the app store and are one of the urgent problems to be solved in maintaining the regular order of the mobile app market. This paper proposes a weakly supervised spam detection framework called WeSSRD. It can mine app reviews for relevance to the app by unsupervised topic modeling methods and then train a weakly supervised detector to detect spam in application stores using minimal prior knowledge. We tested this framework on a real dataset with 14,052 reviews. The detector trained by our proposed framework can achieve a precision rate of 80.97% and a recall rate of 81.89% on the test set, far exceeding the detection method based on similarity.