Abstractive Multi-document Summarization Research Articles

Compressed Heterogeneous Graph for Abstractive Multi-Document Summarization

Multi-document summarization (MDS) aims to generate a summary for a number of related documents. We propose HGSum — an MDS model that extends an encoder-decoder architecture to incorporate a heterogeneous graph to represent different semantic units (e.g., words and sentences) of the documents. This contrasts with existing MDS models which do not consider different edge types of graphs and as such do not capture the diversity of relationships in the documents. To preserve only key information and relationships of the documents in the heterogeneous graph, HGSum uses graph pooling to compress the input graph. And to guide HGSum to learn the compression, we introduce an additional objective that maximizes the similarity between the compressed graph and the graph constructed from the ground-truth summary during training. HGSum is trained end-to-end with the graph similarity and standard cross-entropy objectives. Experimental results over Multi-News, WCEP-100, and Arxiv show that HGSum outperforms state-of-the-art MDS models. The code for our model and experiments is available at: https://github.com/oaimli/HGSum.

Improvements in Multi-Document Abstractive Summarization using Multi Sentence Compression with Word Graph and Node Alignment

The present work proposes a scheme for multi-document abstractive text summarization using node-aligned Word Graph based representation of clustered sentences. In the first step, the proposed scheme uses SBERT embedding for representing the sentences as fixed-size vectors. The sentences belonging to the same cluster are then represented using Word Graph, in which words of different sentences are aligned based on their semantic and syntactic similarities. The advantage of the above representation is that it utilizes alignment information of words between pairs of similar sentences to merge nodes in the Word Graph, and thereby facilitating the generation of sentences with multiple chunks of information. A sentence scoring function assisted by an intensification function is used to measure the grammaticality and informativeness of the generated sentences. Integer Linear Programming has been used to make the final selection of the scored sentences for generating the abstract. Experiments conducted for the task of sentence fusion and multi-document summarization demonstrate superior performance in comparison with the state-of-the-art techniques available in the literature.

Unsupervised Multi-Document Abstractive Summarization Using Recursive Neural Network with Attention Mechanism

Text summarization is an active field of research with a goal to provide short and meaningful gists from large amount of text documents. Extractive text summarization methods have been extensively studied where text is extracted from the documents to build summaries. There are various type of multi document ranging from different formats to domains and topics. With the recent advancement in technology and use of neural networks for text generation, interest for research in abstractive text summarization has increased significantly. The use of graph based methods which handle semantic information has shown significant results. When given a set of documents of English text files, we make use of abstractive method and predicate argument structures to retrieve necessary text information and pass it through a neural network for text generation. Recurrent neural networks are a subtype of recursive neural networks which try to predict the next sequence based on the current state and considering the information from previous states. The use of neural networks allows generation of summaries for long text sentences as well. This paper implements a semantic based filtering approach using a similarity matrix while keeping all stop-words. The similarity is calculated using semantic concepts and Jiang–Conrath similarity and making use of a recurrent neural network with an attention mechanism to generate summary. ROUGE score is used for measuring accuracy, precision and recall scores.

ViMs: a high-quality Vietnamese dataset for abstractive multi-document summarization

Automatic text summarization is important in this era due to the exponential growth of documents available on the Internet. In the Vietnamese language, VietnameseMDS is the only publicly available dataset for this task. Although the dataset has 199 clusters, there are only three documents in each cluster, which is small compared to typical datasets in English. This motivates us to construct ViMs—a big and high-quality Vietnamese dataset for abstractive multi-document summarization. To that end, we recruited 29 annotators and enhanced MDSWriter—an open-source annotation tool, to support the annotators in creating gold standard summaries. As a result, ViMs has 600 summaries corresponding to 300 clusters of 1,945 documents. We have verified the reliability of our dataset by using a variety of metrics including conventional Cohen’s $$\kappa $$ , relaxed Cohen’s $$\kappa $$ —a new metric that we propose to make it more suitable for abstractive summarization, and ROUGE scores. A relaxed $$\kappa $$ score of 0.55 indicate that ViMs could attain moderate agreement between annotators. Meanwhile, ROUGE scores are 0.729 of ROUGE-1, 0.507 of ROUGE-2 and 0.524 of ROUGE-SU4. We have further evaluated ViMs by using three different summarization systems: TextRank, CFVi and MUSEEC. Their performances are 0.628, 0.711 and 0.732 of ROUGE-1, respectively. These results show that the ViMs dataset is suitable for both training and evaluating multi-document summarization systems. We have made the dataset and evaluation results of this work publicly available for research community. It is noted that unlike previous work that only published the final summarization dataset, we also publish intermediate annotation results, which can be used in other NLP problems such as sentence classification.

Neural sentence fusion for diversity driven abstractive multi-document summarization

The lack of multi-document based models and the inaccuracy in representing multiple long documents into a fixed size vector inspired us to solve abstractive multi-document summarization. Also, there is lack of good multi-document based human-authored datasets to train any encoder-decoder models. To overcome this, we have designed complementary models for two different tasks such as sentence clustering and neural sentence fusion. In this work, we minimize the risk of producing incorrect fact by encoding a related set of sentences as an input to the encoder. We have applied our complementary models to implement a full abstractive multi-document summarization system which simultaneously considers importance, coverage, and diversity under a desired length limit. We conduct extensive experiments for all the proposed models which bring significant improvements over the state-of-the-art methods across different evaluation metrics.

An Optimized Semantic Technique for Multi-Document Abstractive Summarization

Background/Objective: Multi-document summarization produces a concise summary from several online topically related documents. A major challenge in this domain is usually the information overlap in documents emanating from various sources. This paper introduces an optimized semantic technique for multi-document abstractive summarization. Methods/Statistical Analysis: Linguistic and semantic approaches are usually employed for abstractive summarization of multiple documents. Linguistic approaches lack semantic representation of source text while semantic approaches mostly rely on human experts to construct domain ontology and rules; which require immense time and effort. The technique in this paper utilizes the benefits of semantic role labeling, clustering and Particle Swarm Optimization (PSO) to rank predicate argument structures (semantic representation) in each cluster using optimized features. Findings: The summary quality is susceptible to the text features i.e., different features have varied importance towards summary generation. Therefore, optimal features weights obtained using PSO integrated in the semantic technique to rank semantic representation improved summarization results. The performance of the technique is evaluated against the benchmark summarization systems using pyramid evaluation measures (mean coverage score, precision and F-measure). A Paired- Samples T-test is carried out to validate the summarization results. Applications/Improvements: Experiment of this research is performed with DUC-2002, a benchmark data set for text summarization. Experimental results confirm that the proposed technique yields better results than other comparison summarization models in terms of mean coverage score and average F-measure.

A framework for multi-document abstractive summarization based on semantic role labelling

We propose a framework for abstractive summarization of multi-documents, which aims to select contents of summary not from the source document sentences but from the semantic representation of the source documents. In this framework, contents of the source documents are represented by predicate argument structures by employing semantic role labeling. Content selection for summary is made by ranking the predicate argument structures based on optimized features, and using language generation for generating sentences from predicate argument structures. Our proposed framework differs from other abstractive summarization approaches in a few aspects. First, it employs semantic role labeling for semantic representation of text. Secondly, it analyzes the source text semantically by utilizing semantic similarity measure in order to cluster semantically similar predicate argument structures across the text; and finally it ranks the predicate argument structures based on features weighted by genetic algorithm (GA). Experiment of this study is carried out using DUC-2002, a standard corpus for text summarization. Results indicate that the proposed approach performs better than other summarization systems.