Abstract:

In order to overcome the shortcomings of the Chronological Ordering and the Majority Ordering methods for summarization sentences, a new ordering algorithm that combines the mutual cohesion among themes and the Majority Ordering method is proposed. Based on the statistical data about the relative position in each pair of themes, a directed graph of the themes is built and the corresponding mutual cohesion is computed. In the ordering process, when a vertex is output from the directed graph, the vertex possessing the greatest cohesion with the vertex is searched from the remaining vertexes. If the cohesion is bigger than the threshold value, the sentences from the two themes corresponding to the two above-mentioned vertexes are placed on adjacent locations in the summarization. Experimental results show that the summarization generated by the proposed ordering algorithm is more coherent and readable.

Key words: artificial intelligence, multi-document summarization, local topic, sentence ordering