This work proposes a sentence network-based approach for performing the task of multi-document text summarization. The sentences of the input set of documents are represented by the nodes of the network. Weighted edges are added between the nodes to represent the semantic similarity between the corresponding sentences. The network has a multilayer structure, where each layer corresponds to an individual input document. This helps in effective differentiation between the inter-document and intra-document edges. A hyperparameter, namely layering factor, has been used to alter the strength of inter-document connections through reinforcement or weakening. It is hypothesized that the summary sentence nodes must act as effective information spreaders in the sentence network. Summary generation is performed by identifying the influential nodes of the network using VoteRank scheme. A comparative study with different network measures, such as Weighted Degree, PageRank, Betweenness centrality, and Closeness centrality reveals the efficacy of the proposed VoteSumm technique for multi-document text summarization. Improved performance is observed when an additional pre-processing step of syntactic simplification is applied on the raw text. Performance is further improved when keyword information is included in the simplified texts.

这项工作提出了一种基于句子网络的方法来执行多文档文本摘要任务。输入文档集的句子由网络的节点表示。节点之间添加加权边来表示对应句子之间的语义相似度。该网络具有多层结构，其中每一层对应于单独的输入文档。这有助于有效区分文档间边缘和文档内边缘。一个超参数，即分层因子，已用于通过强化或弱化来改变文档间连接的强度。假设摘要句节点必须充当句子网络中的有效信息传播者。摘要生成是通过使用 VoteRank 方案识别网络中有影响力的节点来执行的。对不同网络度量（例如加权度、PageRank、介数中心性和紧密性中心性）的比较研究揭示了所提出的 VoteSumm 技术对于多文档文本摘要的有效性。当对原始文本应用额外的语法简化预处理步骤时，可以观察到性能的提高。当简化文本中包含关键字信息时，性能会进一步提高。