As social media plays a more essential part in expressing emotions in daily life, it also provides a novel approach to predict suicidal tendencies. Meanwhile, we can also borrow the deep learning method to mine the entities of the social networks of microblogs of Weibo with suicidal tendencies and provide professional guidance to suicide attempters. We divide different social scenes into different social networks: original family, new family, campus and social work. First, we use the Bert model to re-pretrain among multiple microblogs to acquire the upstream parameters with more distinctive domain characteristics and use the built-in softmax as the downstream network to extract entities in the tagged microblogs with suicidal tendencies. The precision and recall of this model are 90.03% and 86.71%, respectively. Then, we use the re-pretrained Bert model to predict the entities, and the result of Chi-square analysis shows that there is a significant difference between nonsuicidal tendencies microblogs in all 17,647,172 microblogs of active users in 2018 and 176,578 microblogs with suicidal tendencies. Besides, we list the high-frequency entities in different relations among the social networks in the suicide Weibo data with a frequency of occurrence greater than 5. The result shows that the original family is pivotal in influencing suicide attempters and in different relations in social networks, there exist some entities that are highly relevant to the suicidal tendencies. This research provides critical cues for suicidal tendencies detection.