Self-supervised learning on graphs has recently drawn a lot of attention due to its independence from labels and its robustness in representation. Current studies on this topic mainly use static information such as graph structures but cannot well capture dynamic information such as timestamps of edges. Realistic graphs are often dynamic, which means the interaction between nodes occurs at a specific time. This article proposes a self-supervised dynamic graph representation learning framework DySubC, which defines a temporal subgraph contrastive learning task to simultaneously learn the structural and evolutional features of a dynamic graph. Specifically, a novel temporal subgraph sampling strategy is firstly proposed, which takes each node of the dynamic graph as the central node and uses both neighborhood structures and edge timestamps to sample the corresponding temporal subgraph. The subgraph representation function is then designed according to the influence of neighborhood nodes on the central node after encoding the nodes in each subgraph. Finally, the structural and temporal contrastive loss are defined to maximize the mutual information between node representation and temporal subgraph representation. Experiments on five real-world datasets demonstrate that (1) DySubC performs better than the related baselines including two graph contrastive learning models and five dynamic graph representation learning models, especially in the link prediction task, and (2) the use of temporal information cannot only sample more effective subgraphs, but also learn better representation by temporal contrastive loss.

With the rapid growth of scientific papers, understanding the changes and trends in a research area is rather time-consuming. The first challenge is to find related and comparable articles for the research. Comparative citations compare co-cited papers in a citation sentence and can serve as good guidance for researchers to track a research area. We thus go through comparative citations to find comparable objects and build a comparative scientific summarization corpus (CSSC). And then, we propose the comparative graph-based summarization (CGSUM) method to create comparative summaries using citations as guidance. The comparative graph is constructed using sentences as nodes and three different relationships of sentences as edges. The relationship that sentences occur in the same paper is used to calculate the salience of sentences, the relationship that sentences occur in two different papers is used to calculate the difference between sentences, and the relationship that sentences are related to citations is used to calculate the commonality of sentences. Experiments show that CGSUM outperforms comparative baselines on CSSC and performs well on DUC2006 and DUC2007.

Summary News image captioning aims to generate captions or descriptions for news images automatically, serving as draft captions for creating news image captions manually. News image captions are different from generic captions as news image captions contain more detailed information such as entity names and events. Therefore, both images on news and the accompanying text are the source of generating caption of news image. Pointer‐generator network is a neural method defined for text summarization. This article proposes the Multimodal pointer‐generation network by incorporating visual information into the original network for news image captioning. The multimodal attention mechanism is proposed by splitting attention into visual attention paid to the image and textual attention paid to the text. The multimodal pointer mechanism is proposed by using both textual attention and visual attention to compute pointer distributions, where visual attention is first transformed into textual attention via the word‐image relationships. The multimodal coverage mechanism is defined to reduce repetitions of attentions or repetitions of pointer distributions. Experiments on theDailyMailtest dataset and the out‐of‐domainBBCtest dataset show that the proposed model outperforms the original pointer‐generator network, the generic image captioning method, the extractive news image captioning method, and theLDA‐based method accordingBLEU,METEOR, andROUGL‐Levaluations. Experiments also show that the proposed multimodal coverage mechanisms can improve the model, and that transforming visual attention to pointer distributions can improve the model.

Image plays an important role in news articles. Appropriate image arrangements can well transmit scene information and help readers understand more clearly. This paper proposes a neural model based on the cross-modal RNN equipped with an encoding updating mechanism to insert images into proper positions of a news document sequentially. For each image, our model selects the position of which the surrounding sentences are most relevant to the image. Two methods are employed to score the relevance between the image and surrounding sentences: the one employs a layer of convolution network before scoring, and the other employs the pooling operation after the scoring. It is reasonable to believe that inserting an image in news will render the information of the news. We hence introduce an updating mechanism to update the encoding of the news document after the image is inserted. To the best of our knowledge, we are the first to focus on the news image locating problem, as well as provide a feasible solution. We create the dataset for the problem by extending the DailyMail corpora and carry out experiments. Experiments show that the proposed model outperforms four baselines including the pointer network according to two evaluation metrics. The code is released on Github11https://github.com/njupt-summ/ImageLocating.

Main path analysis is a famous network‐based method for understanding the evolution of a scientific domain. Most existing methods have two steps, weighting citation arcs based on search path counting and exploring main paths in a greedy fashion, with the assumption that citation networks are acyclic. The only available proposal that avoids manual cycle removal is to preprint transform a cyclic network to an acyclic counterpart. Through a detailed discussion about the issues concerning this approach, especially deriving the “de‐preprinted” main paths for the original network, this article proposes an alternative solution with two‐fold contributions. Based on the argument that a publication cannot influence itself through a citation cycle, the SimSPC algorithm is proposed to weight citation arcs by counting simple search paths. A set of algorithms are further proposed for main path exploration and extraction directly from cyclic networks based on a novel data structure main path tree . The experiments on two cyclic citation networks demonstrate the usefulness of the alternative solution. In the meanwhile, experiments show that publications in strongly connected components may sit on the turning points of main path networks, which signifies the necessity of a systematic way of dealing with citation cycles.

Summary Related work is a component of a scientific paper, which introduces other researchers' relevant works and makes comparisons with the current author's work. Automatically generating the related work section of a writing paper provides a tool for researchers to accomplish the related work section efficiently without missing related works. This paper proposes an approach to automatically generating a related work section by comparing the main text of the paper being written with the citations of other papers that cite the same references. Our approach first collects the papers that cite the reference papers of the paper being written and extracts the corresponding citation sentences to form a citation document. It then extracts keywords from the citation document and the paper being written and constructs a graph of the keywords. Once the keywords that discriminate the two documents are determined, the minimum Steiner tree that covers the discriminative keywords and the topic keywords is generated. The summary is generated by extracting the sentences covering the Steiner tree. According to ROUGE evaluations, the experiments show that the citations are suitable for related work generation and our approach outperforms the three baseline methods of MEAD, LexRank, and ReWoS. This work verifies the general summarization method based on connotation and extension through citation.

Rapid growth of multi-modal documents containing images on the Internet expresses strong demand on multi-modal summarization. The challenge is to create a computing method that can uniformly process text and image. Deep learning provides basic models for meeting this challenge. This paper treats extractive multi-modal summarization as a classification problem and proposes a sentence–image classification method based on the multi-modal RNN model. Our method encodes words and sentences with the hierarchical RNN models and encodes the ordered image set with the CNN model and the RNN model, and then calculates the selection probability of sentences and the sentence–image alignment probability through a logistic classifier taking text coverage, text redundancy, image set coverage, and image set redundancy as features. Two methods are proposed to compute the image set redundancy feature by combining the important scores of sentences and the hidden sentence–image alignment. Experiments on the extended DailyMail corpora constructed by collecting images and captions from the Web show that our method outperforms 11 baseline text summarization methods and that adopting the two image-related features in the classification method can improve text summarization. Our method is able to mine the hidden sentence–image alignments and to create informative well-aligned multi-modal summaries.

Rapid growth of multi-modal documents on the Internet makes multi-modal summarization research necessary.Most previous research summarizes texts or images separately.Recent neural summarization research shows the strength of the Encoder-Decoder model in text summarization.This paper proposes an abstractive text-image summarization model using the attentional hierarchical Encoder-Decoder model to summarize a text document and its accompanying images simultaneously, and then to align the sentences and images in summaries.A multi-modal attentional mechanism is proposed to attend original sentences, images, and captions when decoding.The DailyMail dataset is extended by collecting images and captions from the Web.Experiments show our model outperforms the neural abstractive and extractive text summarization methods that do not consider images.In addition, our model can generate informative summaries of images.

Summary People often read with aims and reading process significantly influences understanding. This paper defines a new measure of information in text named Aimed Information Quantity in Text by simulating human reading process with reading aims. Aimed Information Quantity in Text reflects not only human memory of words but also relevancy between words emerging in aimed reading process. We demonstrate its application in recommendation. Copyright © 2014 John Wiley & Sons, Ltd.