Generating the football news from live webcast scripts is one kind of automatic text summarization in the field of natural language processing (NLP). Two typical characteristics in the live webcast scripts make the traditional methods can’t well-generate the football news. Short sentences in live webcast scripts make it difficult to identify all of the keywords. And the sentence category identification is more crucial than the other kinds of automation text summarization. In this paper, the custom dictionary, forward dictionary and reverse dictionary in the football news domain are constructed, which aim to identify all import information from the live webcast. And the character-CNN with five strokes model is designed to classify the sentences category. Finally, the scanning algorithm based on time window is proposed to select the sentences to assemble into football news. Experiments results show that our designed Character-CNN model achieved higher accuracy rate on sentence classification than the state-of-art models. Evaluation conducted on ROUGE-1.5.5 toolkit showed that generating the football news with our proposed scanning algorithm outperforms than some baseline methods.