Entailment and Spectral Clustering based Single and Multiple Document Summarization

Abstract

Text connectedness is an important feature for content selection in text summarization methods. Recently, Textual Entailment (TE) has been successfully employed to measure sentence connectedness in order to determine sentence salience in single document text summarization. In literature, Analog Textual Entailment and Spectral Clustering (ATESC) is one such method which has used TE to compute inter-sentence connectedness scores. These scores are used to compute salience of sentences and are further utilized by Spectral Clustering algorithm to create segments of sentences. Finally, the most salient sentences are extracted from the most salient segments for inclusion in the final summary. The method has shown good performance earlier. But the authors observe that TE has never been employed for the task of multi-document summarization. Therefore, this paper has proposed ATESC based new methods for the same task. The experiments conducted on DUC 2003 and 2004 datasets reveal that the notion of Textual Entailment along with Spectral Clustering algorithm proves to be an effective duo for redundancy removal and generating informative summaries in multi-document summarization. Moreover, the proposed methods have exhibited faster execution times.

Investigating Entropy for Extractive Document Summarization

“…The vast majority of extractive, non-neural summarization algorithms use four data-sets for performance evaluation, exhibiting an interesting pattern. Unsupervised summarization methods majorly evaluate performance on DUC datasets (Fang et al (2017) (Huang, 2017;Li et al, 2019), graph-based methods for sentence selection (Zheng & Lapata, 2019;Gupta et al, 2019), multiobjective optimization for sentence selection (Saini et al, 2019;Mishra et al, 2021), BART for abstraction (Chaturvedi et al, 2020;Dou et al, 2021), using embedding based similarity for reducing redundancy (Hailu et al, 2020;Zhong et al, 2020b) etc.…”

Divide and Conquer: From Complexity to Simplicity for Lay Summarization

“…Among non-neural models, a popular approach is to capture relations between sentences or word phrases via a weighted graph. Gupta et al (2014Gupta et al ( , 2019 model the sentences of the document as nodes of a weighted directed graph and compute idf based entailment scores between sentence pairs. They use weighted minimum vertex cover to extract most salient sentences.…”

“…(ii) Unsupervised: Entailment based Weighted Minimum Vertex Cover (wMVC) is an unsupervised network based approach proposed by Gupta et al (2014). The sentences are modelled as vertices of the graph, and Inverse Document Frequency (IDF) based entailment is employed to link sentences Gupta et al (2019). The algorithm considers those sentences important, which entail many sentences.…”

Towards Unlocking the Linguistic Code of Post-Traumatic Stress Disorder: A Comprehensive Analysis and Diagnostic Approach

“…Clinical notes 22 and non-structural interviews [23][24][25] are also used and often associated with a more precise diagnosis using self-questionnaire PCL-5 based on the DSM-5 criterion 26 or semi-structured interview SCID (American Psychiatric Association 2013). To build NLP models, many kinds of linguistic features are extracted: statistical (number of words, number of words per sentence), morpho-syntactic (proportion of rst-person pronoun, verb tense), topic modeling (LDA, LSA); word vector representation (Word2Vec, Doc2Vec, Glove, Fasttext), contextual embeddings vectors (BERT, Roberta), graph-based features 28 , coherence 29 and readability features 30 , external resources such as LIWC 31 , sentiment analysis scores like LabMT 32 , TexBlob (Loria, 2018) or FLAIR 34 and transfer learning methods like DLATK 35 that used pre-trained models on social media data. The models used for the classi cation task, which consists of separating in people with and without PTSD, are mainly Random Forest (RF), 36 , Logistic Regression (LR), CNN, LSTM, and transformers 16,17 .…”