This paper studies the strategies of automatically extracting the conditional relationships between diseases and symptoms from a Chinese encyclopedia site and the disease-related web pages searched from the Internet. At first, the seed symptoms of a disease are extracted from an online medical encyclopedia automatically. These seed symptoms are utilized as query keywords to automatically find more symptoms of a disease from the unstructured documents of the disease-related search results. Next, a jointly learning method is used to construct the embedded representations of the conditional terms and pattern terms. Besides, the semantic similarity matrix of conditional terms is computed through the co-clustering algorithm to discover the representative conditional terms of the clusters. The result of experiments shows that the proposed method, which discovers the semantically related symptoms of a disease associated with conditionals, achieves high accuracy. Besides, many unusually known symptoms considered by the medical experts are discovered, which may be noticeable symptoms needing further verification in the future.