Automatic keyword extraction for scientific literatures using references

References provide some important clues for detecting keywords of the scientific literatures. We propose a unified framework based on word co-occurrence and topic distribution using references to extract top-k single keywords, and remove words within a range of topics. For those multiword keywords, we use LocalMaxs algorithm and apply the Co-occurrence Cohesion Degree to measure the “glue” of the n-gram. Experimental results show that our keyword extraction method by using references can obviously improve the performance of precision, recall and F-measure compared to other keyword extraction methods.