[Purpose/significance] Scientific and efficient domain keywords screening and extraction methods are of great help in identifying and analyzing research fronts and hotspots in domain knowledge. [Method/process] Based on the keyword co-occurrence network, this article compares the keyword extraction methods from four dimensions: individual statistical information, knowledge association breadth, correlation quality, and network structure. [Result/conclusion] The results show that the keyword extraction method based on word frequency has the greatest impact on network robustness and is suitable for identify research hotspots. The method based on degree value emphasizes local features and is applicable to identify core knowledge in unweighted network. The method based on relationship frequency pays attention to the correlation quality, and is suitable for exploring related research of knowledge relevance. The method based on eigen decomposition considers both the global network structural characteristics and the weight of the link, and has comprehensive applicability.