Knowledge Graph Assisted Automatic Writing of Football Match News
Yunkang Zou, Tao Xu, Yugang Dai · 2022
The football news style generated by the extraction method will be colloquial and informal, the background information in the live text is lacking, and the football news generation combined with the template lacks the generalization ability and flexibility. This paper proposes an extraction and generation model integrating knowledge atlas to realize automatic writing of football match news. The extraction model BertSum based on BERT pre training is adopted, and combined with the time information in the football match, candidate sentences are extracted from the live text to form the first press release. The players' names in the first news draft are replaced with special tags, and then the first news draft is rewritten using the generative model PGNet. The coverage mechanism is combined to reduce the duplication of generation. At the same time, a knowledge graph is built to supplement the background knowledge that is missing in the live text broadcast, so as to improve the quality of news generation. The evaluation results show that the proposed method can effectively generate football match news from live text broadcast, and the extraction generation method integrating knowledge graphs is obviously better than the single extraction or generation method.