Search

With the tremendous growth in the number of electronic documents, it is becoming challenging to manage the volume of information. Much research has focused on automatically summarizing the information available in the documents. Multi-Document Summarization (MDS) is one approach that aims to extract the information from the available documents in such a concise way that none of the important points are missed from the summary while avoiding the redundancy of information at the same time. This study presents an extensive survey of extractive MDS over the last decade to show the progress of research in this field. We present different techniques of extractive MDS and compare their strengths and weaknesses. Research work is presented by category and evaluated to help the reader understand the work in this field and to guide them in defining their own research directions. Benchmark datasets and standard evaluation techniques are also presented. This study concludes that most of the extractive MDS techniques are successful in developing salient and information-rich summaries of the documents provided.

Neural headline generation models have recently shown great results since neural network methods have been applied to text summarization. In this paper, we focus on news headline generation. We propose a news headline generation model based on a generative pre-training model. In our model, we propose a rich features input module. The headline generation model we propose only contains a decoder incorporating the pointer mechanism and the n-gram language features, while other generation models use the encoder-decoder architecture. Experiments on news datasets show that our model achieves comparable results in the field of news headline generation.

The emergence of social media platforms like Twitter has become a prominent communication source in disaster outbreak. NGOs, Government agencies leverage twitter’s open and public features to provide immediate relief. Nevertheless, situational information gets immersed in millions of tweets with varying characteristics. Examining each tweet can be cumbersome and time-consuming. Thus, the efficient extraction of disaster-related situational tweets and getting information from all the extracted tweets is required. In the current paper, we have developed a novel framework that uses a deep learning-based classification model to separate the situational tweets from others and summarize them in real-time. Our system is a three-phase process: (a) Creating tweet clusters using a representative set of tweets from the initial set of extracted tweets using a multi-objective optimization concept; (b) When a new tweet arrives, the clusters are updated. The new tweet is classified as situational vs. non-situational. If situational, it is assigned to the closest cluster or new cluster. This assignment is based on its weighted average of syntactic and semantic distances and relevancy to the cluster; (c) Summary is formulated by extracting tweets from each cluster. The proposed approach’s superior performance on four datasets related to different disaster-related events indicates the developed framework’s efficiency over the state-of-the-art techniques.

Community search is a fundamental problem in graph analysis. In many applications, network nodes have specific properties that are essential for making sense of communities. In these networks, attributes are associated with nodes to capture their properties. The community influence is a key property of the community that can be employed to sort the communities in a network based on the relevance/importance of certain attributes. Unfortunately, most of the previously introduced community search algorithms over attributed networks neglected the community influence. In this paper, we study the influential attributed community search problem. Different factors for measuring the influence are discussed. Also, different Influential Attributed Community (InfACom) algorithms based on the concept of k-clique are proposed. Two techniques are presented one for sequential implementation with three variations and one for parallel implementation. In addition, we propose efficient algorithms for maintaining the proposed algorithms on dynamic graphs. The proposed algorithms are evaluated on different real datasets. The experimental results show that the summarization technique reduces the size of the graph by approximately half. In addition, it shows that the proposed algorithms $EnhancedExact$ and $Approximate$ outperform the state-of-the-art approaches Incremental Time efficient $(Inc-T)$ , Incremental Space efficient $(Inc-S)$ , $Exact$ , and 2-Approximation $(AppInc)$ in both efficiency and effectiveness. For the $EnhancedExact$ algorithm, the results show that the efficiency is at least 7 times faster than the $Inc-S$ algorithm, at least 4.5 times faster than the $Inc-T$ algorithm, and 2 times faster than the $AppInc$ algorithm. For the $Approximate$ algorithm, the results show that its efficiency is at least 10 times faster than the $Inc-S$ algorithm, at least 6.4 times faster than the $Inc-T$ algorithm, and 3 times faster than the $AppInc$ algorithm. Finally, the results show that the proposed algorithms retrieve cohesive communities with a smaller diameter than all the state-of-the-art approaches.

Text summarization is an information condensation technique that abbreviates a source document to a few representative sentences with the intention to create a coherent summary containing relevant information of source corpora. This promising subject has been rapidly developed since the advent of deep learning. However, summarization models based on deep neural network have several critical shortcomings. First, a large amount of labeled training data is necessary. This problem is standard for low-resource languages in which publicly available labeled data do not exist. In addition, a significant amount of computational ability is required to train neural models with enormous network parameters. In this study, we propose a model called Learning Free Integer Programming Summarizer (LFIP-SUM), which is an unsupervised extractive summarization model. The advantage of our approach is that parameter training is unnecessary because the model does not require any labeled training data. To achieve this, we formulate an integer programming problem based on pre-trained sentence embedding vectors. We also use principal component analysis to automatically determine the number of sentences to be extracted and to evaluate the importance of each sentence. Experimental results demonstrate that the proposed model exhibits generally acceptable performance compared with deep learning summarization models although it does not learn any parameters during the model construction process.

Towards tackling the phenomenon of textual information overload that is exponentially pumping with redundancy over the Internet, this paper investigates a solution depending on the Automatic Text Summarization (ATS) method. The idea of ATS is to assist, e.g., online readers, in getting a simplified version of texts for preserving their time/effort required to skim a given large body of text. However, ATS is deemed as one of the most complex NLP applications, particularly for the Arabic language that has not been intelligently developed like the other Indo-European languages. Thus, we present an extractive-based summarizer (ArDBertSum) for text written in Arabic, relying on the DistilBERT model. Besides, we propose a domain-specific sentence-clauses segmentater (SCSAR) to support our ArDBertSum in further shortening long/complex sentences. The results of our experiments illustrate that our ArDBertSum yields the best performance, compared with non-heuristic Arabic summarizers, in producing an acceptable quality of candidate summaries. These experiments have been conducted on EASC-dataset (along with our proposed dataset) to report on (1) a statistical evaluation utilizing ROUGE metrics and (2) a specific human-based evaluation. The human evaluation results revealed promising perceptions; however, further works are needed to ameliorate the coherence and punctuation of the automatic summaries.

Conventional Wireless capsule endoscopy (WCE) video summary generation techniques apprehend an image by extracting hand crafted features, which are not essentially sufficient to encapsulate the semantic similarity of endoscopic images. Use of supervised methods for extraction of deep features from an image need an enormous amount of accurate labelled data for training process. To solve this, we use an unsupervised learning method to extract features using convolutional auto encoder. Furthermore, WCE images are classified into similar and dissimilar pairs using fixed threshold derived through large number of experiments. Finally, keyframe extraction method based on motion analysis is used to derive a structured summary of WCE video. Proposed method achieves an average F-measure of 91.1% with compression ratio of 83.12%. The results indicate that the proposed method is more efficient compared to existing WCE video summarization techniques.

Graph representation learning aims to represent the structural and semantic information of graph objects as dense real value vectors in low dimensional space by machine learning. It is widely used in node classification, link prediction, and recommendation systems. However, directly computing the embeddings for original graphs is prohibitively inefficient, especially for large-scale graphs. To address this issue, we present the GSE (Graph Summarization Embedding) model, a more efficient model that computes the nodes’ embeddings based on graph summarization. Specifically, the model first searches for the minimum information entropy of $k$ groups to transform the original graph into a hypergraph with higher-order structural features. Next, the summarization graph’s connection probabilities are used to determine the biased random walks on the hypergraph, which then generates the sequences of the super-nodes. Finally, the node sequences are fed into the skip-gram to generate the vectors of these nodes. Our proposed model improves the efficiency of graph embedding on big data graphs and effectively alleviates the local optimal problem caused by the random walks. Experimental results demonstrate that GSE outperforms main existing clustering baselines, such as K_Means Clustering, Affinity Propagation Clustering, Canopy Clustering, and ACP Clustering. Moreover, our model can be coupled with the main graph embedding methods and improves the Macro-F1 scores and Micro-F1 scores for classification tasks on a variety of real-world graph data.

In the era of rapid development of social networks, user reviews of restaurant review websites have grown rapidly. In order to allow users to quickly grasp the key points of review information on review sites, this paper provides an abstractive multi-text summary method that can automatically generate template-based review summaries based on predefined topics and sentiments. In particular, for each predefined topic and each type of sentiment (positive or negative), this study uses the TextRank algorithm to find the most representative sentences to form a summary. This method allows users to quickly grasp the positive and negative opinions of each important aspect of the restaurant. The previous research on generating abstracts from reviews either did not generate abstracts based on topics, or they were based on topics generated by random models. However, the latter method cannot guarantee that the topics generated by the random model are really the topics that the user needs. For a restaurant review, some topics are indispensable. In order to ensure that abstracts can be generated for these essential topics, our method predefines the topics that must be generated, and then generates abstracts for these topics. In the evaluation, this study compared the template method with the Refresh and Gensim systems based on criteria such as informativeness, clarity, usefulness and likes. The results show that the method proposed in this paper is superior to the other two summary methods.

Code fragment natural language description generation, also known as code summarization, refers to obtaining a natural language sequence describing a given code fragment’s functionality. It is broadly agreed that applying code summarization into production can significantly improve the efficiency of software development and maintenance. In recent years, syntactic analysis (SA) technology and Latent Dirichlet Allocation (LDA) has been widely used in code summarization and has achieved good results. However, most of the existing techniques focus on core code statements, and thus their generated code summarization lacks a logical description of the code fragment’s holistic information. To this end, we propose a code summarization method based on multiple modules to generate natural language for each code statement by constructing a new type of natural language template. Meanwhile, to utilize the code fragment’s holistic information, we adopt the code statement partition rules and cosine similarity measure to rank and optimize the weight of the overall information of the code fragment, and finally generate the holistic natural language description of the code fragment. The experimental results demonstrate that our method can generate more concise and logical natural language descriptions than existing models.

Aiming at the difficulties in document-level summarization, this paper presents a two-stage, extractive and then abstractive summarization model. In the first stage, we extract the important sentences by combining sentences similarity matrix (only used for the first time) or pseudo-title, which takes full account of the features (such as sentence position, paragraph position, and more.). To extract coarse-grained sentences from a document, and considers the sentence differentiation for the most important sentences in the document. The second stage is abstractive, and we use beam search algorithm to restructure and rewrite these syntactic blocks of these extracted sentences. Newly generated summary sentence serves as the pseudo-summary of the next round. Globally optimal pseudo-title acts as the final summarization. Extensive experiments have been performed on the corresponding data set, and the results show our model can obtain better results.

Recording and sharing of educational or lecture videos has increased in recent years. Within these recordings, we find a large number of math-oriented lectures and tutorials which attract students of all levels. Many of the topics covered by these recordings are better explained using handwritten content on whiteboards or chalkboards. Hence, we find large numbers of lecture videos that feature the instructor writing on a surface. In this work, we propose a novel method for extraction and summarization of the handwritten content found in such videos. Our method is based on a fully convolutional network, FCN-LectureNet, which can extract the handwritten content from the video as binary images. These are further analyzed to identify the unique and stable units of content to produce a spatial-temporal index of handwritten content. A signal which approximates content deletion events is then built using information from the spatial-temporal index. The peaks of this signal are used to create temporal segments of the lecture based on the notion that sub-topics change when large portions of content are deleted. Finally, we use these segments to create an extractive summary of the handwritten content based on key-frames. This will facilitate content-based search and retrieval of these lecture videos. In this work, we also extend the AccessMath dataset to create a novel dataset for benchmarking of lecture video summarization called LectureMath. Our experiments on both datasets show that our novel method can outperform existing methods especially on the larger and more challenging dataset. Our code and data are publicly available.

When people debate, they want to familiarize themselves with a whole range of arguments about a given topic in order to deepen their knowledge and inspire new claims. However, the amount of differently phrased arguments is humongous, making the process of processing them time-consuming. In spite of many works on using arguments (e.g. counter-argument generation), there is only a few studies on argument aggregation. To address this problem, we propose a new task in argument mining – Argument Extraction, which gathers similar arguments into key points, usually single sentences describing a set of arguments for a given debate topic. Such a short summary of related arguments has been manually created in previous research, while in our research key point generation becomes fully automatic, saving time and cost. As the first step of key point generation we explore existing similarity calculation methods, i.e. Sentence-BERT and MoverScore to investigate their performance. Next, we propose a combination of argument similarity and Maximal Marginal Relevance (MMR) for extracting key phrases to be utilized in our novel task of Argument Extraction. Experimental results show that MoverScore-based MMR outperforms strong baselines covering 72.5% of arguments when eleven or more arguments are extracted. This percentage is almost identical with the cover rate of human-made key points.

The automatic generation of a text summary is a task of generating a short summary for a relatively long text document by capturing its key information. In the past, supervised statistical machine learning was widely used for the Automatic Text Summarization (ATS) task, but due to its high dependence on the quality of text features, the generated summaries lack accuracy and coherence, while the computational power involved, and performance achieved, could not easily meet the current needs. This paper proposes four novel ATS models with a Sequence-to-Sequence (Seq2Seq) structure, utilizing an attention-based bidirectional Long Short-Term Memory (LSTM), with added enhancements for increasing the correlation between the generated text summary and the source text, and solving the problem of out-of-vocabulary (OOV) words, suppressing the repeated words, and preventing the spread of cumulative errors in generated text summaries. Experiments conducted on two public datasets confirmed that the proposed ATS models achieve indeed better performance than the baselines and some of the state-of-the-art models considered.

Text summarization is an important task in natural language processing (NLP). Neural summary models summarize information by understanding and rewriting documents through the encoder-decoder structure. Recent studies have sought to overcome the bias that cross-entropy-based learning methods can have through reinforcement learning (RL)-based learning methods or the problem of failing to learn optimized for metrics. However, the ROUGE metric with only $n$ -gram matching is not a perfect solution. The purpose of this study is to improve the quality of the summary statement by proposing a reward function used in text summarization based on RL. We propose ROUGE-SIM and ROUGE-WMD, modified functions of the ROUGE function. ROUGE-SIM enables meaningfully similar words, in contrast to ROUGE-L. ROUGE-WMD is a function adding semantic similarity to ROUGE-L. The semantic similarity between articles and summary text was computed using Word Mover’s Distance (WMD) methodology. Our model with two proposed reward functions demonstrated superior performance on ROUGE-1, ROUGE-2, and ROUGE_L than on ROUGE-L as a reward function. Our two models, ROUGE-SIM and ROUGE-WMD, scored 0.418 and 0.406 for ROUGE-L, respectively, for the Gigaword dataset. The two reward functions outperformed ROUGE-L even in the abstractiveness and grammatical aspects.

While describing visual data is a trivial task for humans, it is an intricate task for a computer. This is even more challenging if the visual data is a video. Comprehending a video and describing it is called Video Captioning. This involves understanding the semantics of a video and then generating human-like descriptions of the video. It requires the collaboration of both research communities of computer vision and natural language processing. The captions generated by video captioning can be further utilized for video retrieval, summarization, question-answering, etc. Video Question-Answering (video-QA) involves querying the system to obtain an answer in response. This paper presents a brief survey of the video captioning techniques and a comprehensive review of existing techniques, datasets, and evaluation metrics for the task of video-QA. Video-QA techniques rely on the attention mechanism to generate relevant results. The presented survey shows that recent works on Memory Networks, Generative Adversarial Networks, and Reinforced Decoders, have the capability to handle the complexities and challenges of video-QA. Additionally, the graph-based methods, although less explored, give very promising results. In this article, we have discussed the emerging research directions and various application areas of video-QA.

The ability to detect elephant flows in the forwarding device itself, i.e., a switch, facilitates the deployment of new advanced applications such as load-balancing, per-flow QoS management, etc. Sketches and Space Saving summarization techniques are used for elephant flow detection. However, their memory and computing requirements force the cooperation of an external controller device, due to the scarce resources of current programmable switches. To overcome this limitation, we adapt Sketch and Space Saving elephant flow detection techniques to operate with instant notification and sampled traffic. We evaluate the performance of the resulting techniques with three real traffic traces. The use of sampling allows the identification of a large share of the total traffic corresponding to the elephant flows with a low memory footprint and a reduction of the computing requirements in two orders of magnitude compared to unsampled versions. In turn, we observe a slight increase in the number of false positives and the number of flow notifications. The work of Alberto García-Martínez and Marcelo Bagnulo was supported by the TRUE5G Project ('Evolución hacia redes y servicios auto-gestionados para el 5G del futuro') by the Spanish National Research Agency under Grant PID2019-108713RB-C52/AEI/10.13039/501100011033.

Dealing with vast amounts of textual data requires the use of efficient systems. Automatic summarization systems are capable of addressing this issue. Therefore, it becomes highly essential to work on the design of existing automatic summarization systems and innovate them to make them capable of meeting the demands of continuously increasing data, based on user needs. This study tends to survey the scientific literature to obtain information and knowledge about the recent research in automatic text summarization specifically abstractive summarization based on neural networks. A review of various neural networks based abstractive summarization models have been presented. The proposed conceptual framework includes five key elements identified as encoder-decoder architecture, mechanisms, training strategies and optimization algorithms, dataset, and evaluation metric. A description of these elements is also included in this article. The purpose of this research is to provide an overall understanding and familiarity with the elements of recent neural networks based abstractive text summarization models with an up-to-date review as well as to render an awareness of the challenges and issues with these systems. Analysis has been performed qualitatively with the help of a concept matrix indicating common trends in the design of recent neural abstractive summarization systems. Models employing a transformer-based encoder-decoder architecture are found to be the new state-of-the-art. Based on the knowledge acquired from the survey, this article suggests the use of pre-trained language models in complement with neural network architecture for abstractive summarization task.

With the evolution of the Internet and multimedia technology, the amount of text data has increased exponentially. This text volume is a precious source of information and knowledge that needs to be efficiently summarized. Text summarization is the method to reduce the source text into a compact variant, preserving its knowledge and the actual meaning. Here we thoroughly investigate the automatic text summarization (ATS) and summarize the widely recognized ATS architectures. This paper outlines extractive and abstractive text summarization technologies and provides a deep taxonomy of the ATS domain. The taxonomy presents the classical ATS algorithms to modern deep learning ATS architectures. Every modern text summarization approach’s workflow and significance are reviewed with the limitations with potential recovery methods, including the feature extraction approaches, datasets, performance measurement techniques, and challenges of the ATS domain, etc. In addition, this paper concisely presents the past, present, and future research directions in the ATS domain.

In the last 70 years, the automatic text summarization work has become more and more important because the amount of data on the Internet is increasing so fast, and automatic text summarization work can extract useful information and knowledge what user's need that could be easily handled by humans and used for many purposes. Especially in people's daily life, news text is the type of text most people are exposed to. In this study, a new automatic summarzation model for news text which based on fuzzy logic rules, multi-feature and Genetic algorithm (GA) is introduced. Firstly, the most important feature is word features, we score each word and extracted words that exceeded the preset score as keywords and because news text is a special kind of text, it contains many specific elements, such as time, place and characters, so sometimes these special news elements can be extracted directly as keywords. Second is sentence features, a linear combination of these features shows the importance of each sentence and each feature is weighted by Genetic algorithm. At last, we use fuzzy logic system to calculate the final score in order to get automatic summarization. The results of the proposed method was compared with other methods including Msword, System19, System21, System 31, SDS-NNGA, GCD, SOM and Ranking SVM by using ROUGE assessment method on DUC2002 dataset show that proposed method outperforms the aforementioned methods.

The increasing usage of the Internet and social networks has produced a significant amount of online textual data. These online textual data led to information overload and redundancy. It is important to eliminate the information redundancy and preserve the time required for reading these online textual data. Thus, there is a persistent need for an automatic text summarization system, which extract the relevant and salient information from a collection of documents, that sharing the same or related topics. Then, presenting this extracted information in a condensed form to preserve the main topics. This paper proposes an automatic, generic, and extractive Arabic multi-document summarization system. The proposed system employs the clustering-based and evolutionary multi-objective optimization methods. The clustering-based method discovers the main topics in the text, while the evolutionary multi-objective optimization method optimizes three objectives based on coverage, diversity/redundancy, and relevancy. The performance of the proposed system is evaluated using TAC 2011 and DUC 2002 datasets. The experimental results are compared using ROUGE evaluation measure. The obtained results showed the effectiveness of the proposed system compared to other peer systems. The proposed system outperformed other peer systems for all ROUGE metrics using TAC 2011. We achieved an F-measure of 38.9%, 17.7%, 35.4%, and 15.8% for Rouge-1, Rouge-2, Rouge-L, and Rouge-SU4, respectively. In addition, the proposed system with DUC 2002 dataset achieved an F-measure of 47.1%, 23.7%, 47.1%, 20.4% for Rouge-1, Rouge-2, Rouge-L, and Rouge-SU4, respectively.

Text Summarization is a process which efficiently retrieves the relevant information from documents. The objective of the proposed, unsupervised approach is to summarize bug reports (software artefacts) with complete content and diversified information. The proposed approach utilizes Rapid Automatic Keyword Extraction and term frequency-inverse document frequency method to extract meaningful keywords and key-phrases with a relevant score. For sentence extraction, fuzzy C-means clustering is used to extracts sentences having high degree of membership from each cluster above a set threshold value. A rule-engine is used for sentence selection. The rules are generated with the domain knowledge and based on the extracted information by the keywords and sentences selected by the clustering method. Cohesive and coherent summary is generated by the proposed method on apache bug reports. For redundancy removal and to re-rank generated summary, hierarchical clustering is presented to enrich the extracted summary. The proposed approach is evaluated on newly constructed Apache project Bug Report Corpus (APBRC) and existing Bug Report Corpus (BRC). The results are compared on the basis of performance metrics such as precision, recall, pyramid precision and F-score. The experimental results depict that our proposed approach attains significant improvement over other baseline approaches such as BRC and LRCA. It also attains significant improvement over existing state-of-art unsupervised approaches such as Hurried, centroid and others. It extracts significant keyword phrases and sentences from each cluster to achieve full coverage and coherent summary. The results evaluated on APBRC corpus attains an average value of 78.22%, 82.18%, 80.10% and 81.66% for precision, recall, f-score and pyramid precision respectively.

With the advent of the information age, excessive information collection leads to information overload. Automatic text summarization technology has become an effective way to solve information overload. This paper proposes an automatic text summarization model, which extends traditional sequence-to-sequence (Seq2Seq) neural text summarization model by using a syntax-augmented encoder and a headline-aware decoder. The encoder encodes both syntactic structure and word information of a sentence in the sentence embedding. A hierarchical attention mechanism is proposed to pay attentions to syntactic units. The decoder is improved by a headline attention mechanism and a Dual-memory-cell LSTM network to achieve a higher quality of generated summaries. We designed experiments to compare the proposed method with baseline models on the CNN/DM datasets. The experiment results show that the proposed method is superior to abstractive baseline models in terms of the scores on ROUGE evaluation metrics, and achieve a summary generation performance comparable to the extractive baseline method. Though qualitative analysis, the summary quality of the propose method is more readable and less redundant, which agrees well with our intuition.

Over the last decade, discourse relations, also referred to as rhetorical or coherence relations, have been used to improve a range of natural language processing applications. Researchers have devised several theories, including rhetorical structure theory and cross-document structure theory, to examine relations between generic text units in single and multiple documents, respectively. In this paper, we propose a cross-article structure theory (CAST), that extends the benefit of discourse relations to multi-scientific article applications. It is based on the rhetorical structure theory (RST) and the cross-document structure theory (CST). The insight that underpins CAST is to consider both intra-section and cross-section relations. At the outset, these relations are classified based on the structural features of the article (that is, their appearance within each section type) and then the relations between text portions across multiple articles are classified. The practicality of the theory is showcased by solving a problem that consists to identify the types of relations which exist between each pair of sentences in related sections of different articles. A CAST bank was created and the k-nearest neighbors algorithm was used to develop two classifiers based on CAST and CST, respectively. The performance results obtained markedly demonstrate the role of the specific relations to scientific articles in CAST. Other applications of CAST could address the redundancy and readability problems, which represent main issues for different tasks, such as the summarization of multiple articles.

Code summarization has long been viewed as a challenge in software engineering because of the difficulties of understanding source code and generating natural language. Some mainstream methods combine abstract syntax trees with language models to capture the structural information of the source code and generate relatively satisfactory comments. However, these methods are still deficient in code understanding and limited by the long dependency problem. In this paper, we propose a novel model called Fret, which stands for Functional REinforced Transformer with BERT. The model provides a new way to generate code comments by learning code functionalities and deepening code understanding while alleviating the problem of long dependency. For this purpose, a novel reinforcer is proposed for learning the functional contents of code so that more accurate summaries to describe the code functionalities can be generated. In addition, a more efficient algorithm is newly designed to capture the source code structure. The experimental results show that the effectiveness of our model is remarkable. Fret significantly outperforms all the state-of-the-art methods we examine. It pushes the BLEU-4 score to 24.32 for Java code summarization (14.23% absolute improvement) and the ROUGE-L score to 40.12 for Python. An ablation test is also conducted to further explore the impact of each component of our method.

Automatic related work generation is a new challenge in multi-document scientific summarization focusing on refining a related work section for a given scientific paper. In this paper, we propose a brand new framework ToC-RWG for related work generation by incorporating topic model and citation information. We present an unsupervised generative probabilistic model, called QueryTopicSum, which utilizes a LDA-style model to characterize the generative process of both the scientific paper and its reference papers. We also take advantage of citations of reference papers to identify Cited Text Spans (CTS) from reference papers. This approach provides us with a perspective of annotating the importance of the reference papers from the academic community. With QueryTopicSum and the identified CTS as candidate sentences, an optimization framework based on minimizing KL divergence is exerted to select the most representative sentences for related work generation. Our evaluation results on a set of 50 scientific papers along with their corresponding reference papers show that ToC-RWG achieves a considerable improvement over generic multi-document summarization and scientific summarization baselines.

Bidirectional Encoder Representations from Transformers represents the latest incarnation of pre-trained language models which have been obtained a satisfactory effect in text summarization tasks. However, it has not achieved good results for the generation of Chinese short text summaries. In this work, we propose a novel short text summary generation model based on keyword templates, which uses templates found in training data to extract keywords to guide summary generation. The experimental results of the LCSTS data set show that our model performs better than the baseline model. The analysis shows that the methods used in our model can generate high-quality summaries.

In natural language processing, text summarization is an important application used to extract desired information by reducing large text. Existing studies use keyword-based algorithms for grouping text, which do not give the documents' actual theme. Our proposed dynamic corpus creation mechanism combines metadata with summarized extracted text. The proposed approach analyzes the mesh of multiple unstructured documents and generates a linked set of multiple weighted nodes by applying multistage Clustering. We have generated adjacency graphs to link the clusters of various collections of documents. This approach comprises of ten steps: pre-processing, making multiple corpuses, first stage clustering, creating sub-corpuses, interlinking sub-corpuses, creating page rank keyword dictionary of each sub-corpus, second stage clustering, path creation among clusters of sub-corpuses, text processing by forward and backward propagation for results generation. The outcome of this technique consists of interlinked sub-corpuses through clusters. We have applied our approach to a News dataset, and this interlinked corpus processing follows step by step clustering to search the most relevant parts of the corpus with less cost, time, and improve content detection. We have applied six different metadata processing combinations over multiple text queries to compare results during our experimentation. The comparison results of text satisfaction show that Page-Rank keywords give 38% related text, single-stage Clustering gives 46%, two-stage Clustering gives 54%, and the proposed technique gives 67% associated text. Furthermore, this approach covers/searches the relevant data with a range of most to less relevant content. It provides the systematic query-relevant corpus processing mechanism, which automatically selects the most relevant sub-corpus through dynamic path selection. We used the SHAP model to evaluate the proposed technique, and our evaluation results proved that the proposed mechanism improved text processing. Moreover, combining text summarization features, shown satisfactory results compared to the summaries generated by general models of abstractive & extractive summarization.

With the continuous development of deep learning, pre-trained models have achieved sound effects in the field of natural language processing. However, text summarization research is far from what people want, especially in abstractive summarization. A high-quality summarization system needs to focus on the topic content of the document and the similarity between the summary and the source document. In this paper, we propose a topic information fusion and semantic relevance for text summarization based on Fine-tuning BERT(TIF-SR). Primarily, considering the critical role of topic information in summary generation, we extract topic keywords and fusion them with source documents as part of the input. Secondly, make the summary closer to the source document by calculating the semantic similarity between the generated summary and the source document, the quality of the abstract is improved. The experimental data indicate that the ROUGE index and readability have improved in this model, so these shreds of evidence suggest that the method proposed by our model is sufficient.

A good document summary should summarize the core content of the text. Research on automatic text summarization attempts to solve this problem. The encoder-decoder model is widely used in text summarization research. Soft attention is used to obtain the required contextual semantic information during decoding. However, due to the lack of access to the key features, the generated summary deviates from the core content. In this paper, we proposed an encoder-decoder model based on a double attention pointer network (DAPT). In DAPT, the self-attention mechanism collects key information from the encoder, the soft attention and the pointer network generate more coherent core content, and the fusion of both generates accurate and coherent summaries. In addition, the improved coverage mechanism is used to address the repetition problem and improve the quality of the generated summaries. Simultaneously, scheduled sampling and reinforcement learning (RL) are combined to generate new training methods to optimize the model. Experiments on the CNN/Daily Mail dataset and the LCSTS dataset show that our model performs as well as many state-of-the-art models. The experimental analysis shows that our model achieves higher summarization performance and reduces the occurrence of repetition.

The automatic text summarization (ATS) task consists in automatically synthesizing a document to provide a condensed version of it. Creating a summary requires not only selecting the main topics of the sentences but also identifying the key relationships between these topics. Related works rank text units (mainly sentences) to select those that could form the summary. However, the resulting summaries may not include all the topics covered in the source text because important information may have been discarded. In addition, the semantic structure of documents has been barely explored in this field. Thus, this study proposes a new method for the ATS task that takes advantage of semantic information to improve keyword detection. This proposed method increases not only the coverage by clustering the sentences to identify the main topics in the source document but also the precision by detecting the keywords in the clusters. The experimental results of this work indicate that the proposed method outperformed previous methods with a standard collection.

The exponential growth of the Web documents has constituted the need for automatic document summarization. In this context, extractive document summarization, i.e., that task of extracting the most relevant information, removing redundancy and presenting the remained data in a coherent and cohesive structure, is a challenging task. In this article, we propose a novel intelligent approach, namely ExDoS, that harvests benefits of both supervised and unsupervised algorithms simultaneously. To the best of our knowledge, ExDoS is the first approach to combine both supervised and unsupervised algorithms in a single framework and an interpretable manner for document summarization purpose. ExDoS iteratively minimizes the error rate of the classifier in each cluster with the help of dynamic local feature weighting. Moreover, this approach specifies the contribution of features to discriminate each class, which is a challenging issue in the summarization task. Therefore, in addition to summarizing text, ExDoS is also able to measure the importance of each feature in the summarization process. We evaluate our model both automatically (in terms of ROUGE factor) and empirically (human analysis) on the benchmark datasets: the DUC2002 and CNN/DailyMail. Results show that our model obtains higher ROUGE scores comparing to most state-of-the-art models. The human evaluation also demonstrates that our model is capable of generating informative and readable summaries.

The determination of semantic similarity between sentences is an important component in natural language processing (NLP) tasks such as text retrieval and text summarization. Many approaches have been proposed for estimating sentence similarity, and Siamese neural networks (SNN) provide a better approach. However, the sentence semantic representation, generated by sharing weights in the SNN without any attention mechanism, ignores the different contributions of different words to the overall sentence semantics. Furthermore, the attention operation within only a single sentence neglects interactive semantic influence on similarity estimation. To address these issues, an interactive self-attention (ISA) mechanism is proposed in this paper and integrated with an SNN, named an interactive self-attentive Siamese neural network (ISA-SNN) which is used to verify the effectiveness of ISA. The proposed model obtains the weights of words in a single sentence by means of self-attention and extracts inherent interactive semantic information between sentences via interactive attention to enhance sentence semantic representation. It achieves better performances without feature engineering than other existing methods on three biomedical benchmark datasets (a Pearson correlation coefficient of 0.656 and 0.713/0.658 on DBMI and CDD-ful/-ref, respectively).

Group activity recognition has received a great deal of interest because of its broader applications in sports analysis, autonomous vehicles, CCTV surveillance systems and video summarization systems. Most existing methods typically use appearance features and they seldom consider underlying interaction information. In this work, a technology of novel group activity recognition is proposed based on multi-modal relation representation with temporal-spatial attention. First, we introduce an object relation module, which processes all objects in a scene simultaneously through an interaction between their appearance feature and geometry, thus allowing the modeling of their relations. Second, to extract effective motion features, an optical flow network is fine-tuned by using the action loss as the supervised signal. Then, we propose two types of inference models, opt-GRU and relation-GRU, which are used to encode the object relationship and motion representation effectively, and form the discriminative frame-level feature representation. Finally, an attention-based temporal aggregation layer is proposed to integrate frame-level features with different weights and form effective video-level representations. We have performed extensive experiments on two popular datasets, and both have achieved state-of-the-art performance. The datasets are the Volleyball dataset and the Collective Activity dataset, respectively.

In today’s age, a massive amount of videos are produced every day, which contains audio, visual, and textual data. This constant increase is due to the ease of recording service in portable devices, such as mobile phones, tablets, or cameras. The major challenge is to understand the visual semantics and convert it into a condensed format, such as caption or summary to save storage space, enables users to index, navigate, and help gain information in less time. We propose an innovative joint end-to-end solution, Abstractive Summarization of Video Sequences, which uses the deep neural network to generate the natural language description and abstractive text summarization of an input video. This provides a text-based video description and abstractive summary, enabling users to discriminate between relevant and irrelevant information according to their needs. Furthermore, our experiments show that the joint model can attain better results than the baseline methods in separate tasks with informative, concise, and readable multi-line video description and summary in a human evaluation.

Metaphor is the central issue of language and thinking. Metaphor identification plays a significant preliminary role in the field of machine translation, reading comprehension, and automatic summarization, making it a focus of natural language processing. Recently, research into Chinese verb metaphor identification has become a widespread concern. The main problem is that the usage of semantic resources is relatively simple, and there is a lack of deep semantic support. Therefore, this paper proposes a word representation method suitable for metaphor classification tasks, which combines the traditional word vector with structural information from the Synonym Thesaurus, so that the word vector can contain the abstraction degree of the word in the metaphor. On this basis, we propose a verb metaphor attention network based on subject-predicate and verb-object relationships, which gives full consideration to the global syntactic information when we perform LSTM encoding and calculate the weight of each word. At the same time, it can facilitate the understanding of literalness and non-literalness. The experimental results show that the identification effect improves on the existing results, indicating that word representation combining semantic resources and attention network can improve the verb metaphor identification performance.

Scientific literature summarization aims to summarize related papers into a survey. It can help researchers, especially newcomers, to quickly know the current situation of their professional area from massive literature. Since the structure of documents is very important for scientific literature summarization, and it is obviously observed that there is a relationship and semantic information hidden in document structures of scientific literature, therefore, we employ a hierarchical attention model to learn document structure from all the papers for summarization. In particular, we utilize attention mechanism to capture relations among document structure and learn semantic information on document discourse levels, and judge the importance of each sentence according to its surroundings by the information obtained. Moreover, we automatically construct a scientific literature data set consisting of surveys and their references. We evaluate our proposed model on this dataset with ROUGE metrics, experiments prove that our approach is effective, and our model outperforms several baselines.

Quantile summarization is a useful tool in data streams management and mining that can efficiently capture the distribution of the data. A quantile of a sequence of points is the point with a given rank in the sequence. Given a sequence of uncertain points S on the real line, each represented by one-dimensional probability density function (pdf), we study the problem of incrementally maintaining quantile summaries on S such that for any query with a given rank, the summaries can provide a point as the quantile within a given error. We define quantile on uncertain data with discrete or continuous pdf in terms of two error metrics under possible worlds semantics. For an answer to a quantile query on uncertain data, we give the methods for calculating the value of the error and thereby discussing the high-level features of the summaries that can answer approximate quantile query under the two error metrics. We propose an online, space efficient algorithm to compute such summary data on uncertain data streams. The experimental results show that our algorithm substantially outperforms other techniques, such as Monte Carlo and averaging methods, in terms of query error and space for storing the summary data.

ive text summarization plays an important role in the field of natural language processing. However, the abstractive text summary adopts deep learning research method to predict words often appears semantic inaccuracy and repetition and so on. at the present stage, in order to solve the problem that semantic inaccuracy, we propose an MS-Pointer Network that based on the multi-head self-attention mechanism, which a multi-head self-attention mechanism is introduced in the basic encoder-decoder model. Since multi-head self-attention can combine input words into the encoder-decoder arbitrarily, and given a higher weight of these words that combination of the semantics, thereby achieving the purpose of enhancing the semantic features of the text, so that the abstractive text summary is more semantically structured, And the multi-head self-attention mechanism add the position information of the input text, which can enhance the semantic representation of the text. At the same time, in order to solve the problem of out of vocabulary, a pointer network is introduced on the seqtoseq with a multi-head attention mechanism. The model is referred to as MS-Pointer Network. We used CNN/Daily Mail and Gigaword datasets to validate our model, and uses the ROUGE metric to measure model. Experiments have shown that abstractive text summaries generated using the multi-head self-attention mechanism outperforming current open state-of-the-art two points averagely.

Topic models have been widely utilized in Topic Detection and Tracking tasks, which aim to detect, track, and describe topics from a stream of broadcast news reports. However, most existing topic models neglect semantic or syntactic information and lack readable topic descriptions. To exploit semantic and syntactic information, Language Models (LMs) have been applied in many supervised NLP tasks. However, there are still no extensions of LMs for unsupervised topic clustering. Moreover, it is difficult to employ general LMs (e.g., BERT) to produce readable topic summaries due to the mismatch between the pretraining method and the summarization task. In this paper, noticing the similarity between content and summary, first we propose a Language Model-based Topic Model (LMTM) for Topic Clustering by using an LM to generate a deep contextualized word representation. Then, a new method of training a Topic Summarization Model is introduced, where it is not only able to produce brief topic summaries but also used as an LM in LMTM for topic clustering. Empirical evaluations of two different datasets show that the proposed LMTM method achieves better performance over four baselines for JC, FMI, precision, recall and F1-score. Additionally, the generated readable and reasonable summaries also validate the rationality of our model components.

Code comments are a key software component for program comprehension and software maintainability. High-quality code and comments are urgently needed by data-driven models widely used in tasks like code summarization. Many existing approaches for assessing the quality of comments are machine learning based classification algorithms or rely on heuristic rules. These approaches are difficult to capture the complicated features of text data and are often limited in accuracy, efficiency, and generalization ability. In this paper, we convert the quality assessment of code comments into a classification problem based on the multi-input neural network. We summarize the input, the code and comments, into vectors using the attention-based Bi-LSTM model and the weighted GloVe model, respectively, and concatenate the code vectors and the comment vectors as the input of the Multiple-Layer Perceptron classifier for the comment quality assessment. Experimental results show that our approach, in general, outperforms the previous technique, on both our labeled dataset and the public dataset, with the F1-score of 96.91% and 91.90%, respectively. Using the training set and the testing set from distinct sources, our approach can still achieve reasonable performance, which demonstrates its generalization ability.

Identifying interesting patterns from a huge amount of data is a challenging task across a wide range of application domain. Especially, for cyber security being able to identify rare types of network activities or anomalies from network traffic data (a.k.a. Big Data!) is an important but time-consuming data analysis task having moderate computing resources. Existing research has shown that it is possible to detect rare anomalies from the summarized version of big data. Therefore, summarization is an effective preprocessing function before applying anomaly detection techniques. This aim of this paper is to improve and quantify the scalability and accuracy of the anomaly detection techniques by using summarization. Hence, we propose a sampling-based summarization technique (SUCh: Summarization Using Chernoff-Bound) which is computationally effective than the existing techniques and also performs better in identifying rare anomalies from twelve benchmark network traffic datasets. The experimental results show that, instead of using original dataset, a summary of the data yields better performance in terms of true positive and false positive rates, when used for anomaly detection with less time required.

Video summarization (VS) is to identify important content from a given video, which can help users quickly comprehend video content. Recently, sparse dictionary selection (SDS) has demonstrated to be an effective solution for VS problems, which generally assumes a linear relationship between keyframes and non-keyframes. However, this assumption is not always true for video frames which possess intrinsic nonlinear structures and properties. In this paper, by exploiting the nonlinearity between video frames, a nonlinear SDS model is formulated for VS, in which the nonlinearity is transformed to linearity by projecting a video to a high-dimensional feature space induced by a kernel function. We also propose two greedy optimization algorithms to solve the resulting model, namely the standard kernel SDS (KSDS) greedy algorithm and the robust KSDS greedy algorithm with a backtracking strategy. In order to achieve an intuitive and flexible configuration of the VS process, an adaptive criterion, namely energy ratio, is devised to produce video summaries with different lengths for different video contents. Experimental results on two different benchmark video datasets demonstrate that the proposed algorithm outperforms several state-of-the-art VS algorithms.

Domination number is of practical interest in several theoretical and applied scenes. In the problem of wireless networking, the dominating idea is used to deduce an efficient route within the adhoc mobilenetworks. It has also been used for the summarization of documents and making a design of secure web-systems for electrical recursive grids. In this paper, we explore an important class of domination numbers, which is the 2-rainbow independent dominating function (2RiDF) on graphs. The minimum weight of a 2RiDF on a graph G is called the 2-rainbow independent domination number of G. A graph G is 2-rainbow independent domination stable if the 2-rainbow independent domination number of G remains unchanged under removal of any vertex. As a result, we characterize 2-rainbow independent domination stable tree-networks and study the effect of edge removal on 2-rainbow independent domination number in trees.

Recent advances in the film industry have given rise to exponential growth in movie/drama production and adaptation of the Big Data concept. Automatic identification and classification of movie characters have received tremendous attention from researchers due to its applications in video semantic analysis, video summarization, and personalized video retrieval for which several methods have been recently presented. However, these methods cannot detect main characters properly due to their variation in pose and style in different scenes of a movie. To address this problem we present DeepStar, a novel framework for starring character identification based on deep high-level robust features. The proposed framework is threefold: the extraction of shots with clear faces from the input video; face clustering using discriminative deep features; and the occurrence matrix generation, helping in the selection of starring characters. The promising results obtained using representative Hollywood movies demonstrate the effectiveness of our method in detecting starring characters over the state-of-the-art methods.

The application of neural networks in natural language processing, including abstractive text summarization, is increasingly attractive in recent years. However, teaching a neural network to generate a human-readable summary that reflects the core idea of the original source text (i.e., semantically similar) remains a challenging problem. In this paper, we explore using generative adversarial networks to solve this problem. The proposed model contains three components: a generator that encodes the long input text into a shorter representation; a discriminator to teach the generator to create human-readable summaries and another discriminator to restrict the output of the generator to reflect the core idea of the input text. The main training process can be carried out in an adversarial learning process. To solve the non-differentiable problem caused by the words sampling process, we use the policy gradient algorithm to optimize the generator. We evaluate the proposed model on the CNN/Daily Mail summarization task. The experimental results show that the model outperforms previous state-of-the-art models.

Teaching activities have nowadays been supported by a variety of electronic devices. Formative assessment tools allow teachers to evaluate the level of understanding of learners during frontal lessons and to tailor the next teaching activities accordingly. Despite plenty of teaching materials are available in the textual form, manually exploring these very large collections of documents can be extremely time-consuming. The analysis of learner-produced data (e.g., test outcomes) can be exploited to recommend short extracts of teaching documents based on the actual learner's needs. This paper proposes a new methodology to recommend summaries of potentially large teaching documents. Summary recommendations are customized to student's needs according to the results of comprehension tests performed at the end of frontal lectures. Specifically, students undergo multiple-choice tests through a mobile application. In parallel, a set of topic-specific summaries of the teaching documents is generated. They consist of the most significant sentences related to a specific topic. According to the results of the tests, summaries are personally recommended to students. We assessed the applicability of the proposed approach in real context, i.e., a B.S. university-level course. The results achieved in the experimental evaluation confirmed its usability.

This paper presents an enhanced artificial bee colony (ABC) algorithm for solving the web service selection problem. The proposed algorithm searches the best possible combination of web services to satisfy user requirements. An adapted neighborhood selection and replacement process and a swapping process are used to improve the ABC behavior. Neighboring nodes are employed to enhance ABC performance by encouraging exploration in early iterations, where bees have no knowledge regarding the search space, and by encouraging exploitation in later iterations to exploit bee knowledge of the search space. The swapping process is used to enhance ABC performance by randomly swapping portions among the best two solutions randomly. The idea behind this swap is to exploit the characteristics of the best solutions to generate new solutions. We compared the proposed algorithm with other algorithms in terms of quality and execution time using 60 different datasets. These datasets have different numbers of tasks and web services. The results indicate that the proposed algorithm finds better solutions compared to other algorithms. In addition, the results' summarization on 60 datasets with 30 different executions shows that the proposed algorithm outperforms the threshold-based algorithm by 6% and the enhanced ABC by 3% in terms of solutions' quality.

Automatic text summarization aims to reduce the document text size by building a brief and voluble summary that has the most important ideas in that document. Through the years, many approaches were proposed to improve the automatic text summarization results; the graph-based method for sentence ranking is considered one of the most important approaches in this field. However, most of these approaches rely on only one weighting scheme and one ranking method, which may cause some limitations in their systems. In this paper, we focus on combining multiple graph-based approaches to improve the results of generic, extractive, and multi-document summarization. This improvement results in more accurate summaries, which could be used as a significant part of some natural language applications. We develop and experiment with two graph-based approaches that combine four weighting schemes and two ranking methods in one graph framework. To combine these methods, we propose taking the average of their results using the arithmetic mean and the harmonic mean. We evaluate our proposed approaches using DUC 2003 & DUC 2004 dataset and measure the performance using ROUGE evaluation toolkit. Our experiments demonstrate that using the harmonic mean in combining weighting schemes outperform the arithmetic mean and show a good improvement over the baselines and many state-of-the-art systems.

In online video communities, an emerging source of comment, such as danmaku, allows viewers to interact when watching. Prior work discussed the feasibility of application using danmaku, while the comprehensive analysis of large-scale data is vacant to be filled in. We here release our danmaku data collection and report interesting observed phenomena in the danmaku comments. This analysis reveals the temporal distributions and user’s access patterns of online time-sync comments. In particular, we distribute two novel natural language processing (NLP) tasks based on our danmaku dataset and provide baseline models. In the first task, we show how the naive models predict positive or negative sentiment given a danmaku comment, which effectively extends the real applications such as opinion poll prediction and marketing investigation. In the second task, we propose to use the NLP summarization model to make video tagging and summarization. The experimental results suggest that danmaku can not only support deeper and richer interactions between viewers and videos but also with high research value.