Unsupervised Text Segmentation Using Semantic Relatedness Graphs

Proceedings of the Fifth Joint Conference on Lexical and Computational Semantics (*SEM 2016), pages 125–130, Berlin, Germany, August 11-12, 2016. Unsupervised Text Segmentation Using Semantic Relatedness Graphs Goran Glavaˇs, Federico Nanni, Simone Paolo Ponzetto Data and Web Science Group University of Mannheim B6 26, DE-68161 Mannheim, Germany {goran,federico,simone}@.informatik.uni-mannheim.de Abstract Segmenting text into semantically coher- ent fragments improves readability of text and facilitates tasks like text summariza- tion and passage retrieval. In this pa- per, we present a novel unsupervised al- gorithm for linear text segmentation (TS) that exploits word embeddings and a mea- sure of semantic relatedness of short texts to construct a semantic relatedness graph of the document. Semantically coherent segments are then derived from maximal cliques of the relatedness graph. The al- gorithm performs competitively on a stan- dard synthetic dataset and outperforms the best-performing method on a real-world (i.e., non-artificial) dataset of political man- ifestos. 1 Introduction Despite the fact that in mainstream natural lan- guage processing (NLP) and information retrieval (IR) texts are modeled as bags of unordered words, texts are sequences of semantically coherent seg- ments, designed (often very thoughtfully) to ease readability and understanding of the ideas con- veyed by the authors. Although authors may explic- itly define coherent segments (e.g., as paragraphs), many texts, especially on the web, lack any explicit segmentation. Linear text segmentation aims to represent texts as sequences of semantically coherent segments. Besides improving readability and understandabil- ity of texts for readers, automated text segmenta- tion is beneficial for NLP and IR tasks such as text summarization (Angheluta et al., 2002; Dias et al., 2007) and passage retrieval (Huang et al., 2003; Dias et al., 2007). Whereas early approaches to unsupervised text segmentation measured the co- herence of segments via raw term overlaps between sentences (Hearst, 1997; Choi, 2000), more recent methods (Misra et al., 2009; Riedl and Biemann, 2012) addressed the issue of sparsity of term-based representations by replacing term-vectors with vec- tors of latent topics. A topical representation of text is, however, merely a vague approximation of its meaning. Con- sidering that the goal of TS is to identify seman- tically coherent segments, we propose a TS algo- rithm aiming to directly capture the semantic re- latedness between segments, instead of approxi- mating it via topical similarity. We employ word embeddings (Mikolov et al., 2013) and a measure of semantic relatedness of short texts (ˇSari´c et al., 2012) to construct a relatedness graph of the text in which nodes denote sentences and edges are added between semantically related sentences. We then derive segments using the maximal cliques of such similarity graphs. The proposed algorithm displays competitive performance on the artifically-generated bench- mark TS dataset (Choi, 2000) and, more im- portantly, outperforms the best-performing topic modeling-based TS method on a real-world dataset of political manifestos. 2 Related Work Automated text segmentation received a lot of at- tention in NLP and IR communities due to its use- fulness for text summarization and text indexing. Text segmentation can be performed in …

Content truncated to 500 words