LexRank: Graph-based Lexical Centrality as Salience in Text Summarization

Abstract

We introduce a stochastic graph-based method for computing relative importance of textual units for Natural Language Processing. We test the technique on the problem of Text Summarization (TS). Extractive TS relies on the concept of sentence salience to identify the most important sentences in a document or set of documents. Salience is typically defined in terms of the presence of particular important words or in terms of similarity to a centroid pseudo-sentence. We consider a new approach, LexRank, for computing sentence importance based on the concept of eigenvector centrality in a graph representation of sentences. In this model, a connectivity matrix based on intra-sentence cosine similarity is used as the adjacency matrix of the graph representation of sentences. Our system, based on LexRank ranked in first place in more than one task in the recent DUC 2004 evaluation. In this paper we present a detailed analysis of our approach and apply it to a larger data set including data from earlier DUC evaluations. We discuss several methods to compute centrality using the similarity graph. The results show that degree-based methods (including LexRank) outperform both centroid-based methods and other systems participating in DUC in most of the cases. Furthermore, the LexRank with threshold method outperforms the other degree-based techniques including continuous LexRank. We also show that our approach is quite insensitive to the noise in the data that may result from an imperfect topical clustering of documents.

The Nexus of FDI, R&D, and Human Capital on Chinese Sustainable Development: Evidence from a Two-Step Approach

“…Depending on the attributes of the connection relationship, centrality can be interpreted in various ways (e.g., degree, closeness, and betweenness) [58]. Degree centricity refers to the degree of strong connection and attention in the network, which can be useful as the simplest and most effective indicator of the power relationship across nodes [59,60]. Degree centrality is measured to the extent to which a node is connected to another node in the network.…”

Timeline Summarization from Relevant Headlines

“…Beside the difference in the approach (using headlines instead of the full text), our framework differs from MDS in that it takes the relations among events across dates into account. As there is already a rich body of research on multi-document summarization (for example, [22], [21], [16], [8], [17]), in this study we also investigate how good they are in producing daily summaries using only headlines, in the same setting as our approaches.…”

“…Since ETS and Chieu et al extract sentences from the full text of news articles for timeline summaries, as shown in the experiments of Yan et al [27], we also would like to see how good (multi-)document summarization would work on the headlines dataset. We consider the following state-of-the-art methods: Centroid [22], LexRank [8], TextRank [19] SumSim selects top news reports and non-duplicated headlines that maximize the sum of TF-IDF similarity with other headlines that are published in the previous and next 10 days. Conceptually, it works similarly to Chieu et al, but on the headline level.…”

Building Document Graphs for Multiple News Articles Summarization: An Event-Based Approach

“…Different measurements are used to determine how to represent sentence and how to define connections between sentences. The similarity between two sentences according to their term vectors is used to generate links and define link strength in [4]. Similarly, [3] weighed links by the content overlap of two sentences normalized by the length of each sentence.…”