The problem of identifying rumors is practical importance especially in online social networks, since information can diffuse more rapidly and widely than the offline counterpart. In this paper, we identify characteristics by examining following three aspects diffusion: temporal, structural, linguistic. For temporal characteristics, propose a new periodic time series model that considers daily external shock cycles, where demonstrates rumor likely have fluctuations over time. We also key...

Influence maximization is the problem of selecting top k seed nodes in a social network to maximize their influence coverage under certain diffusion models. In this paper, we propose novel algorithm IRIE that integrates advantages ranking (IR) and estimation (IE) methods for both independent cascade (IC) model its extension IC-N incorporates negative opinion propagations. Through extensive experiments, demonstrate matches other algorithms while scales much better than all algorithms....

This study determines the major difference between rumors and non-rumors explores rumor classification performance levels over varying time windows—from first three days to nearly two months. A comprehensive set of user, structural, linguistic, temporal features was examined their relative strength compared from near-complete date Twitter. Our contribution is at providing deep insight into cumulative spreading patterns as well tracking precise changes in predictive powers across features....

Speech emotion recognition is a challenging task, and extensive reliance has been placed on models that use audio features in building well-performing classifiers. In this paper, we propose novel deep dual recurrent encoder model utilizes text data signals simultaneously to obtain better understanding of speech data. As emotional dialogue composed sound spoken content, our encodes the information from sequences using neural networks (RNNs) then combines these sources predict class. This...

Neural question generation (NQG) is the task of generating a from given passage with deep neural networks. Previous NQG models suffer problem that significant proportion generated questions include words in target, resulting unintended questions. In this paper, we propose answer-separated seq2seq, which better utilizes information both and target answer. By replacing answer original special token, our model learns to identify interrogative word should be used. We also new module termed...

In this paper, we are interested in exploiting textual and acoustic data of an utterance for the speech emotion classification task. The baseline approach models information from audio text independently using two deep neural networks (DNNs). outputs both DNNs then fused classification. As opposed to knowledge modalities separately, propose a framework exploit tandem with lexical data. proposed uses bi-directional long short-term memory (BLSTM) obtaining hidden representations utterance....

In this paper, we propose a framework for isolating text regions from natural scene images. The main algorithm has two functions: it generates region candidates, and verifies of the label candidates (text or non-text). are generated through modified K-means clustering algorithm, which references texture features, edge information color information. candidate labels then verified in global sense by Markov Random Field model where collinearity weight is added as long most texts aligned....

In this paper, for overlapping community detection, we propose a novel framework of the link-space transformation that transforms given original graph into graph. Its unique idea is to consider topological structure and link similarity separately using two distinct types graphs: line For structure, each mapped node graph, which enables us discover communities non-overlapping detection algorithms as in similarity, it calculated on carried over keep transformed Thus, our transformation, by...

The context-dependent nature of online aggression makes annotating large collections data extremely difficult. Previously studied datasets in abusive language detection have been insufficient size to efficiently train deep learning models. Recently, Hate and Abusive Speech on Twitter, a dataset much greater reliability, has released. However, this not comprehensively its potential. In paper, we conduct the first comparative study various models discuss possibility using additional features...

In this paper, we propose a novel method for sentence-level answer-selection task that is fundamental problem in natural language processing. First, explore the effect of additional information by adopting pretrained model to compute vector representation input text and applying transfer learning from large-scale corpus. Second, enhance compare-aggregate proposing latent clustering within target corpus changing objective function listwise pointwise. To evaluate performance proposed...

In this paper, we propose an attention-based classifier that predicts multiple emotions of a given sentence. Our model imitates human’s two-step procedure sentence understanding and it can effectively represent classify sentences. With emoji-to-meaning preprocessing extra lexicon utilization, further improve the performance. We train evaluate our with data provided by SemEval-2018 task 1-5, each which has several labels among 11 emotions. achieves 5th/1st rank in English/Spanish respectively.

Despite the fact that large language models (LLMs) show exceptional skill in instruction following tasks, this strength can turn into a vulnerability when are required to disregard certain instructions. Instruction-following tasks typically involve clear task description and input text containing target data be processed. However, itself resembles an instruction, confusion may arise, even if there is explicit prompting distinguish between input. We refer phenomenon as instructional...

Accurate and timely prediction of streamflow is critical for managing the increasing risks associated with floods, particularly in developing countries where traditional in-situ monitoring systems are often sparse or non-existent. This study introduces a novel probabilistic multi-step ahead model that leverages Graph Neural Networks (GNNs), self-attention mechanisms via Informer network, distributional output layer to enhance predictive accuracy uncertainty quantification time series. By...

Seunghyun Yoon, Joongbo Shin, Kyomin Jung. Proceedings of the 2018 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.

Some news headlines mislead readers with overrated or false information, and identifying them in advance will better assist choosing proper stories to consume. This research introduces million-scale pairs of headline body text dataset incongruity label, which can uniquely be utilized for detecting misleading headlines. On this dataset, we develop two neural networks hierarchical architectures that model a complex textual representation articles measure the between text. We also present data...

In this paper, we propose an evaluation metric for image captioning systems using both and text information. Unlike the previous methods that rely on textual representations in evaluating caption, our approach uses visiolinguistic representations. The proposed method generates image-conditioned embeddings each token ViLBERT from generated reference texts. Then, these contextual of two sentence-pair are compared to compute similarity score. Experimental results three benchmark datasets show...

Hwanhee Lee, Seunghyun Yoon, Franck Dernoncourt, Trung Bui, Kyomin Jung. Proceedings of the 59th Annual Meeting Association for Computational Linguistics and 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). 2021.

We define the notion of a transitive-closure spanner directed graph. Given graph G = (V, E) and an integer k ≥ 1, k-transitive-closure-spanner (k-TC-spanner) is H EH) that has (1) same as (2) diameter at most k. These spanners were studied implicitly in access control, property testing, data structures, properties these have been rediscovered over span 20 years. bring areas under unifying framework TC-spanners. abstract common task tackled diverse applications problem constructing sparse...

We consider optimizing the coalition structure in Coalitional Skill Games (CSGs), a succinct representation of coalitional games. In CSGs, value depends on tasks its members can achieve. The require various skills to complete them, and agents may have different skill sets. optimal is partition coalitions, that maximizes sum utilities obtained by coalitions. show CSGs represent any characteristic function, generation this representation. provide hardness results, showing general as well very...

In this work, we explore the impact of visual modality in addition to speech and text for improving accuracy emotion detection system. The traditional approaches tackle task by independently fusing knowledge from various modalities performing classification. contrast these approaches, problem introducing an attention mechanism combine information. regard, first apply a neural network obtain hidden representations modalities. Then, is defined select aggregate important parts video data...

Motivated by applications of distributed linear estimation, control, and optimization, we consider the question designing iterative algorithms for computing average numbers in a network. Specifically, our interest is such an algorithm with fastest rate convergence given topological constraints As main result this paper, design possible using nonreversible Markov chain on network graph. We construct transforming standard chain, which obtained Metropolis-Hastings method. call novel...

Given a directed graph $G = (V,E)$ and an integer $k \geq 1$, $k$-transitive-closure-spanner ($k$-TC-spanner) of $G$ is $H (V, E_H)$ that has (1) the same transitive-closure as (2) diameter at most $k$. These spanners were implicitly studied in context circuit complexity, data structures, property testing, access control, properties these have been rediscovered over span 20 years. We abstract common task tackled diverse applications problem constructing sparse TC-spanners. initiate study...

Nowadays, recommendation systems are widely used on various services. This system predicts what item a user will use next using large amounts of stored history. Recommendation commonly applied in fields as movies, e-commerce, and social However, previous researches overlooked the importance usage sequence time intervals series data from users. We provide novel that incorporates these temporal properties design recurrent neural network (RNN) model with hierarchical structure so user's history...