Search

The ordering of sentence is an important part in multi-document summarization system,and the quality of ordering will directly affect the coherence and readability of the generated abstract. The current sentence ordering algorithm based on time is dependent on the time tag of the document, and has poor universality; sentence ordering algorithm based on large corpus is so complex in training that the quality of ordering is low. For these problems,this paper proposed an ordering algorithm based on topic document collection, the aim was to solve the problem of sentence ordering without time tag. Firstly, it used conditional entropy and context proximity algorithm to get the adjacent degree and connected relationship between sentence pairs from the source document set. Then, it used the maximum weight pruning ordering algorithm to form the final results. Because the proposed algorithm depends only on the source documents collection of the summary sentence,so it has stronger domain generality. The experimental results show that the proposed algorithm provides a higher precision than former sentence ordering strategies. [ABSTRACT FROM AUTHOR]