Abstract: The social platform represented by Weibo has become an important way for people to obtain and release information, and it has also become a breeding ground for false information. Weibo containing false information often contains obvious emotional bias. From the perspective of sentiment analysis, this paper proposes the LableBert model: a false information recognition model based on Bert combined with BI-LSTM. It uses the emotional dictionary to add weights to emotional words, and improves Bert's pre-training tasks, and enhances the model's ability to extract implicit emotional features. And batch mark the text emotional polarity of the masked words, and strengthen the model's ability to acquire emotional features of the text context, and combine BI-LSTM to identify false information. Experimental results show that the accuracy of the model has reached 91.36%, and the F₁ value has reached 91.03%. Compared with the original Bert model, the accuracy and F₁ value of the model have been improved.