Answering unique topic queries with dynamic threshold

摘要

Queries with threshold are common when dealing with unstructured data such as text corpus. It often requires several exploring attempts for users to achieve final results. In this work, we propose an automatic sampling method for threshold determination without any interaction with users, in which two optimizing algorithms are introduced to reach the lower-bound time complexity in each sampling trial. We evaluate our methods using several experiments and demonstrate the effectiveness of it, which can be an enormously powerful tool for ordinary users.