Minzheng Wang, Run Luo, Yanbo Wang, Zichen Liu, Yuqiao Tan, Tao Tan, Nan Xu, Lu Wang, Wenji Mao. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Interlingual subtitling, which translates subtitles of visual media into a target language, is essential for entertainment localization but has not yet been explored in machine translation. Although Large Language Models (LLMs) have significantly advanced the general capabilities of machine translation, the distinctive characteristics of subtitle texts pose persistent challenges in interlingual subtitling, particularly regarding semantic coherence, pronoun and terminology translation, and translation expressiveness. To address these issues, we present Hermes, an LLM-based automated subtitling framework. Hermes integrates three modules: Speaker Diarization, Terminology Identification, and Expressiveness Enhancement, which effectively tackle the above challenges. Experiments demonstrate that Hermes achieves state-of-the-art diarization performance and generates expressive, contextually coherent translations, thereby advancing research in interlingual subtitling.

Zhaoxin Yu, Xinglin Xiao, Wenji Mao. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Graph representation learning is a fundamental research theme and can be generalized to benefit multiple downstream tasks from the node and link levels to the higher graph level. In practice, it is desirable to develop task-agnostic graph representation learning methods that are typically trained in an unsupervised manner. However, existing unsupervised graph models, represented by the variational graph auto-encoders (VGAEs), can only address node- and link-level tasks while manifesting poor generalizability on the more difficult graph-level tasks because they can only keep low-orderisomorphic consistencywithin the subgraphs of one-hop neighborhoods. To overcome the limitations of existing methods, in this paper, we propose the Isomorphic-Consistent VGAE (IsoC-VGAE) for multi-level task-agnostic graph representation learning. We first devise an unsupervised decoding scheme to provide a theoretical guarantee of keeping the high-order isomorphic consistency within the VGAE framework. We then propose the Inverse Graph Neural Network (Inv-GNN) decoder as its intuitive realization, which trains the model via reconstructing the node embeddings and neighborhood distributions learned by the GNN encoder. Extensive experiments on multi-level graph learning tasks verify that our model achieves superior or comparable performance compared to both the state-of-the-art unsupervised methods and representative supervised methods with distinct advantages on the graph-level tasks.

Metaphor detection aims to identify whether a linguistic expression in text is metaphorical or literal.Most existing research tackles this problem either using word-pair or token-level information as input, and thus treats word-pair and token-level metaphor detection as distinct subtasks.Benefited from the simplified structure of word pairs, recent methods for wordpair metaphor detection can provide intermediate explainable clues for the detection results, which remains a challenging issue for tokenlevel metaphor detection.To mitigate this issue in token-level metaphor detection and take advantage of word pairs, in this paper, we make the first attempt to bridge word-pair and tokenlevel metaphor detection via modeling word pairs within a sentence as explainable intermediate information.As the central role of verb in metaphorical expressions, we focus on tokenlevel verb metaphor detection and propose a novel explainable Word Pair based Domain Mining (WPDM) method.Our work is inspired by conceptual metaphor theory (CMT).We first devise an approach for conceptual domain mining utilizing semantic role mapping and resources at cognitive, commonsense and lexical levels.We then leverage the inconsistency between source and target domains for core word pair modeling to facilitate the explainability.Experiments on four datasets verify the effectiveness of our method and demonstrate its capability to provide the core word pair and corresponding conceptual domains as explainable clues for metaphor detection.

The development of large language models (LLMs) raises the importance of assessing the fairness and completeness of various evaluation benchmarks. Regrettably, these benchmarks predominantly utilize uniform manual prompts, which may not fully capture the expansive capabilities of LLMs—potentially leading to an underestimation of their performance. To unlock the potential of LLMs, researchers pay attention to automated prompt search methods, which employ LLMs as optimizers to discover optimal prompts. However, previous methods generate the solutions implicitly, which overlook the underlying thought process and lack explicit feedback. In this paper, we propose a novel prompt introspective search framework, namely PromISe, to better release the capabilities of LLMs. It converts the process of optimizing prompts into an explicit chain of thought, through a step-by-step procedure that integrates self-introspect and self-refine. Extensive experiments, conducted over 73 tasks on two major benchmarks, demonstrate that our proposed PromISe significantly boosts the performance of 12 well-known LLMs compared to the baseline approach. Moreover, our study offers enhanced insights into the interaction between humans and LLMs, potentially serving as a foundation for future designs and implementations. Keywords: large language models, prompt search, self-introspect, self-refine

Article comment generation is a novel and challenging task in natural language generation, which has attracted widespread attention from researchers in recent years. High-quality article comments such as relevant, diverse, and informative ones can greatly promote user interactions and enhance the user experience. However, current research works generally overlook the relevance between comments and the source article, which may generate mediocre and dull comments. To address this problem, a variational multi-layer fusion model (VMFM) based on variational auto-encoder (VAE) is proposed in this paper. The posterior distribution of the proposed VMFM is employed to supervise the prior network in selecting context-related latent variables from the source article, which are further integrated into the decoder to increase the relevance between generated comments and the source article. Due to the sequential nature of text generation, the influence of those latent variables on the decoder gradually diminishes during auto-regressive decoding. To mitigate this issue, we propose a multi-layer fusion method, which fuses a series of context-related latent variables extracted from the source article into every decoder layer. Experiments on four datasets show that our model significantly outperforms strong baselines in relevance, diversity, informativeness and fluency of generated comments based on automatic and human evaluations.

Yuan Tian, Nan Xu, Wenji Mao. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Adversarial vulnerability remains a major obstacle to the construction of reliable NLP systems. When imperceptible perturbations are added to raw input text, the performance of a deep learning model may drop dramatically under attacks. Recent work has argued that the adversarial vulnerability of a model is caused by non-robust features in supervised training. Thus, in this paper, we tackle the adversarial robustness challenge by means of disentangled representation learning, which is able to explicitly disentangle robust and non-robust features in text. Specifically, inspired by the variation of information (VI) in information theory, we derive a disentangled learning objective composed of mutual information to represent both the semantic representativeness of latent embeddings and the differentiation of robust and non-robust features. On the basis of this, we design a disentangled learning network to estimate the mutual information for realization. Experiments on the typical text-based tasks show that our method significantly outperforms the representative methods under adversarial attacks, indicating that discarding non-robust features is critical for improving model robustness.

Multimodal sarcasm detection is an important research topic in natural language processing and multimedia computing, and benefits a wide range of applications in multiple domains.Most existing studies regard the incongruity between image and text as the indicative clue in identifying multimodal sarcasm.To capture cross-modal incongruity, previous methods rely on fixed architectures in network design, which restricts the model from dynamically adjusting to diverse imagetext pairs.Inspired by routing-based dynamic network, we model the dynamic mechanism in multimodal sarcasm detection and propose the Dynamic Routing Transformer Network (DynRT-Net).Our method utilizes dynamic paths to activate different routing transformer modules with hierarchical co-attention adapting to cross-modal incongruity.Experimental results on a public dataset demonstrate the effectiveness of our method compared to the stateof-the-art methods.Our codes are available at https://github.com/TIAN-viola/DynRT.

Metaphor detection is an important and challenging task in natural language processing, which aims to distinguish between metaphorical and literal expressions in text.Previous studies mainly leverage the incongruity of source and target domains and contextual clues for detection, neglecting similar attributes shared between source and target concepts in metaphorical expressions.Based on conceptual metaphor theory, these similar attributes are essential to infer implicit meanings conveyed by the metaphor.Under the guidance of conceptual metaphor theory, in this paper, we model the likeness of attribute for the first time and propose a novel Attribute lIkeness and Domain Inconsistency Learning framework (AIDIL) for word-pair metaphor detection.Specifically, we propose an attribute siamese network to mine similar attributes between source and target concepts.We then devise a domain contrastive learning strategy to learn the semantic inconsistency of concepts in source and target domains.Extensive experiments on four datasets verify that our method significantly outperforms the previous state-of-the-art methods, and demonstrate the generalization ability of our method.

Stance detection aims to identify the user's attitude toward specific targets from text, which is an important research area in text mining and benefits a variety of application domains.Existing studies on stance detection were conducted mainly in English.Due to the low-resource problem in most non-English languages, crosslingual stance detection was proposed to transfer knowledge from high-resource (source) language to low-resource (target) language.However, previous research has ignored the practical issue of no labeled training data available in target language.Moreover, target inconsistency in cross-lingual stance detection brings about the additional issue of unseen targets in target language, which in essence requires the transfer of both language and target-oriented knowledge from source to target language.To tackle these challenging issues, in this paper, we propose the new task of cross-lingual cross-target stance detection and develop the first computational work with dual knowledge distillation.Our proposed framework designs a cross-lingual teacher and a cross-target teacher using the source language data and a dual distillation process that transfers the two types of knowledge to target language.To bridge the target discrepancy between languages, cross-target teacher mines target category information and generalizes it to the unseen targets in target language via category-oriented learning.Experimental results on multilingual stance datasets demonstrate the effectiveness of our method compared to the competitive baselines 1 .

Adversarial training is the dominant strategy towards model robustness.Current adversarial training methods typically apply perturbations to embedding representations, whereas actual text-based attacks introduce perturbations as discrete tokens.Thus there exists a gap between the continuous embedding representations and discrete text tokens that hampers the effectiveness of adversarial training.Moreover, the continuous representations of perturbations cannot be further utilized, resulting in the suboptimal performance.To bridge this gap for adversarial robustness, in this paper, we devise a novel generative adversarial training framework that integrates gradient-based learning, adversarial example generation and perturbed token detection.Our proposed framework consists of generative adversarial attack and adversarial training process.Specifically, in generative adversarial attack, the embeddings are shared between the classifier and the generative model, which enables the generative model to leverage the gradients from the classifier for generating perturbed tokens.Then, adversarial training process combines adversarial regularization with perturbed token detection to provide token-level supervision and improve the efficiency of sample utilization.Extensive experiments on five datasets from the AdvGLUE benchmark demonstrate that our framework significantly enhances the model robustness, surpassing the state-of-the-art results of ChatGPT by 10% in average accuracy.

Security-related events are continuously emerging around the world, such as social unrest, armed conflicts and natural disasters. Although event knowledge graphs provide a concise and intuitive way to organize event information, they often suffer from the incompleteness problem. To solve this problem, the knowledge graph reasoning technique is usually adopted to infer new facts based on existing relations of entities. However, existing reasoning methods often concentrate on the entity relations, without considering the attributes of nodes for event knowledge graphs, such as event descriptions and event time, which contain rich information of events and are vital for event relation reasoning. Existing methods are also deficient in interpretability, which significantly benefits security-related decision-making scenarios. In this paper, we propose a Neuro-Logic Learning (NLL) method for relation reasoning over event knowledge graph. Our method first adopts different event attributes to learn the semantic representations of events, and ranks candidate event relation triples according to their similarity scores. The proposed method then develops the neural logic rule leaner to obtain interpretable reasoning rules. We construct a new large-scale event relation dataset EventKG-R based on EventKG to evaluate the effectiveness of our proposed method. Experimental results show that our proposed method achieves superior performances compared to the state-of-the-art baselines.

Stance detection is an important research area in social media analytics and text mining and has a number of security-related applications, which aims to identify the user's viewpoint towards specific targets in domains such as social security, politics and public events. Previous research on stance detection has mainly focused on monolingual scenarios with the limited number of targets, while little attention was paid to cross-lingual stance detection. In contrast to the abundant labeled data in source language (typically English), the labeled data are often scarce in the non-English target language. Meanwhile, the diverse expressions of target further complicate the task and bring additional challenges. In this paper, we focus on cross-lingual stance detection in practical applications, where target is generally expressed as a topic and no labeled data are available for the target language. To tackle the above challenges, we propose an adversarial topic-aware memory network (ATOM) for cross-lingual stance detection. Specifically, our method first mines the generalized topic representations across source and target languages and utilizes them as the guidance to transfer knowledge from the high-resource source language to the low-resource target language. We further develop an iterative memory network to facilitate knowledge transfer across languages, which adaptively generates language-invariant topic-aware clues via adversarial training. Experimental results on three multilingual datasets in the politics domain demonstrate the effectiveness of our proposed method.

News comments provide a convenient way for people to express opinions and exchange ideas. Positive comments en-courage a harmonious discussion atmosphere within news media communities. In contrast, offensive or insulting comments may result in cyberbullying and personal psychological trauma, which have particular practical impacts in security-related domains. The automatic generation of news comments with controllable attributes (e.g. sentiment) to assist users and news platform administrators is greatly needed. However, existing research for news comment generation has not addressed the controllable issue yet. On the other hand, existing methods for controllable text generation focus on token-level constraints, which are not applicable to controlling the sentence level attributes for news comment generation. To address this challenging issue, in this paper, we propose an attribute-level contrastive learning method for controllable news comment generation. To apply attribute level constraints on the generated text, our method considers the attributes of the generated comments and the pre-defined attributes as different views of the same attribute, and maximizes their similarity during the training process. We conduct experiments on two publicly available news comment datasets, and the experimental results show that our model achieves competitive performance in news comment generation and attribute controllability.

The multimodality nature of web data has necessitated complex multimodal information retrieval for a wide range of web applications. Deep neural networks (DNNs) have been widely employed to extract semantic features from raw samples to improve retrieval accuracy. In addition, hashing is widely used to improve computational and storage efficiency. As such, deep hashing frameworks have been applied for multimodal retrieval tasks. However, there is still a great recognitive gap between primate brain structure-inspired DNNs and humans. On computer vision tasks, well-crafted DNN models can be easily defeated by invisible small attacks, and this phenomenon indicates a large recognition gap between DNN models and humans. Recently, adversarial defense methods have been shown to improve the human–machine recognition alignment in several classification tasks. However, the robustness problem on the retrieval tasks, especially on the deep hashing-based multimodal retrieval models, is still not well studied. Therefore, in this article, we present an adversarially robust training mechanism to improve model robustness for the purpose of human–machine recognition alignment on retrieval tasks. Through extensive experimental results on several social multimodal retrieval benchmarks, we show that the robust training hashing framework proposed can mitigate the recognition gap on retrieval tasks. Our study highlights the necessity of robustness enhancement on deep hashing models.

The wide application of deep neural networks (DNNs) has significantly improved the performance of hashing models on multimodal retrieval issues. DNN-based deep models can automatically learn semantic features from raw data to make human-level decisions. However, the superior generalization leads to potential privacy leakage risks. Strong DNN-based retrieval models enable malicious crawlers to search for nontag private information based on semantic similarity matching. Hence, executing effective privacy protection mechanisms against those retrieval software is essential for reliable social website construction. In this article, we propose a retrieval task-based adversarial perturbation generation method called Hashing Fake to meet this request. Specifically, DNNs are recently found to be vulnerable to a specific set of attacks called adversarial perturbations, which denote some magnitude-restricted signals added on objective samples to misguide well-crafted DNN models, and perturbations’ magnitudes are small enough that will not induce humans’ perception. Moreover, since existing adversarial perturbation generation methods are designed for supervised tasks, Hashing Fake constructs a differential approximation substitution for perturbation production on unsupervised retrieval tasks. Through extensive experiments on several deep retrieval benchmarks, we demonstrate that well-crafted perturbations using Hashing Fake can effectively misguide objective models’ recognitions to make false predictions. The added norm-restricted perturbations on objective samples will not alter humans’ perception; hence, Hashing Fake can be applied on real-world social websites to protect subscribers’ privacy against malicious retrieval software.

With the explosive increase of multimodal data, cross-modal correlation classification has become an important research topic and is in great demand in many cross-modal applications. A variety of classification schemes and predictive models have been built based on the existing cross-modal correlation categorization. However, these classification schemes typically follow the prior assumption that the paired cross-modal samples are strictly related, and thus pay great attention to the fine-grained relevant types of cross-modal correlation, ignoring the high volume of implicitly relevant data which are often wrongly classified into irrelevant types. Even more, previous predictive models fall short of reflecting the essence of cross-modal correlation according to their definitions, especially in the modeling of network structure. Thus in this paper, by comprehensively investigating the current image-text correlation classification research, we redefine a new classification scheme for cross-modal correlation based on the implicit and explicit relevance. To predict the types of image-text correlation based on our proposed definition, we further devise the Association and Alignment Network (namely AnANet) to model the implicit and explicit relevance, which captures both the implicit association of global discrepancy and commonality between image and text and explicit alignment of cross-modal local relevance. Experimental studies on our constructed new image-text correlation dataset verify the effectiveness of our proposed model.

To inhibit the spread of rumorous information, fact checking aims at retrieving evidence to verify the truthfulness of a given statement. Fact checking methods typically use knowledge graphs (KGs) as external repositories and develop reasoning methods to retrieve evidence from KGs. As real-world statement is often complex and contains multiple claims, multi-claim fact verification is not only necessary but more important for practical applications. However, existing methods only focus on verifying a single claim (i.e. a single-claim statement). Multiple claims imply rich context information and modeling the interrelations between claims can facilitate better verification of a multi-claim statement as a whole. In this paper, we propose a computational method to model inter-claim interactions for multi-claim fact checking. To focus on relevant claims within a statement, our method first extracts topics from the statement and connects the triple claims in the statement to form a claim graph. It then learns a policy-based agent to sequentially select topic-related triples from the claim graph. To fully exploit information from the statement, our method further employs multiple agents and develops a hierarchical attention mechanism to verify multiple claims as a whole. Experimental results on two real-world datasets show the effectiveness of our method for multi-claim fact verification.

With the rapid development of the Internet, mining opinions and emotions from the explosive growth of user-generated content is a key field of social media analysis. However, the expression forms of the central opinion which strongly expresses the essential points and converges the main sentiments of the overall document are diverse in practice, such as sequential sentences, a sentence fragment, or an individual sentence. Previous research studies on sentiment analysis based on document level and sentence level fail to deal with this actual situation uniformly. To address this issue, we propose a Central Opinion Extraction (COE) framework to boost performance on sentiment analysis with social media texts. Our framework first extracts a span-level central opinion text, which expresses the essential opinion related to sentiment representation among the whole text, and then uses extracted textual span to boost the performance of sentiment classifiers. The experimental results on a public dataset show the effectiveness of our framework for boosting the performance on document-level sentiment analysis task.

Identifying hidden nodes in social networks is a critical issue in security-related applications. In contrast to the conventional node classification on graphs with all nodes being observable, it is more challenging to classify the hidden nodes that are unobservable during the training process, also known as the “inductive learning” in previous research. Existing approaches for inductive node classification mainly adopt graph neural network models to learn node representations. Although these methods are advantageous to modeling the topology of graph-structured data, they rely heavily on node features which may vary significantly in different specific application scenarios. In addition, the inherently changeable graph structure induced by hidden nodes may cause the over-fitting problem. To address the above issues and boost the performances of hidden node classification, we propose a deep generative model based on variational auto-encoders. Specifically, we design a novel graph neural network to aggregate the multi-hop neighbor information of each node. Meanwhile, to better utilize the graph structure information as a supplement to node features, we consider the heterogeneous node influences and introduce a gated attention mechanism using node degrees. Moreover, our proposed model can be trained by minibatches and thus is applicable to large social networks. We conduct experiments on four real-world datasets, and verify the effectiveness of our method for hidden graph node classification.

The rapid development of Internet and online data explosions elicit strong aspirations for users to search for semantic relevant information based on available samples. While the data online are always released with different modalities like images, videos or texts, effective retrieval models should discover latent semantic information with different structures. Recently, the state-of-the-art deep cross modal retrieval frameworks have effectively enhanced the performance on commonly-used platforms using the deep neural networks (DNNs). Yet DNNs have been verified to be easily misguided by small perturbations, and there are already several attack generation methods proposed on DNN-based models for real-world tasks, but they are all focused on supervised tasks like classification or object recognition. To effectively evaluate the robustness of deep cross-modal retrieval frameworks, in this paper, we propose a retrieval-based adversarial perturbation generation method, and demonstrate that our perturbation could effectively attack the state-of-the-art deep cross-modal and single image retrieval hashing models.

Cross-modal retrieval is an important multimedia research area which aims to take one type of data as the query to retrieve relevant data of another type. Most of the existing methods follow the paradigm of pair-wise learning and class-level learning to generate a common embedding space, where the similarity of heterogeneous multimodal samples can be calculated. However, in contrast to large-scale cross-modal retrieval applications which often need to tackle multiple modalities, previous studies on cross-modal retrieval mainly focus on two modalities (i.e., text-image or text-video). In addition, for large-scale cross-modal retrieval with modality diversity, another important problem is that the available training data are considerably modality-imbalanced. In this paper, we focus on the challenging problem of modality-imbalanced cross-modal retrieval, and propose a Multimodal Coordinated Clustering Network (MCCN) which consists of two modules, Multimodal Coordinated Embedding (MCE) module to alleviate the imbalanced training data and Multimodal Contrastive Clustering (MCC) module to tackle the imbalanced optimization. The MCE module develops a data-driven approach to coordinate multiple modalities via multimodal semantic graph for the generation of modality-balanced training samples. The MCC module learns class prototypes as anchors to preserve the pair-wise and class-level similarities across modalities for intra-class compactness and inter-class separation, and further introduces intra-class and inter-class margins to enhance optimization flexibility. We conduct experiments on the benchmark multimodal datasets to verify the effectiveness of our proposed method.

In practical applications of cross-modal retrieval, test queries of the retrieval system may vary greatly and come from unknown category. Meanwhile, due to the cost and difficulty of data collection as well as other issues, the available data for cross-modal retrieval are often imbalanced over different modalities. In this paper, we address two important issues to increase the robustness of cross-modal retrieval system for real-world applications: handling test queries from unknown category and modality-imbalanced training data. The first issue has not been addressed by existing methods and the second issue was not well addressed in the related research. To tackle the above issues, we take the advantage of prototype learning, and propose a prototype-based adaptive network (PAN) for robust cross-modal retrieval. Our method leverages a unified prototype to represent each semantic category across modalities, which provides discriminative information of different categories and takes unified prototypes as anchors to learn cross-modal representations adaptively. Moreover, we propose a novel prototype propagation strategy to reconstruct balanced representations which preserves the semantic consistency and modality heterogeneity. Experimental results on the benchmark datasets demonstrate the effectiveness of our method compared to the SOTA methods, and further robustness tests show the superiority of our method in solving the above issues.

Recent studies show that neural models for natural language processing are usually fragile under adversarial attacks (e.g., character-level insertion and word-level synonym substitution), which exposes the lack of robustness. Most defense techniques are tailored to specific semantic level attacks and do not possess the ability to mitigate multi-level attack simultaneously. Adversarial training has been shown the effectiveness of increasing model robustness. However, it often suffers from degradation on normal data, especially when the proportion of adversarial examples increase. To address this, we propose mixup regularized adversarial training (MRAT) against multi-level attack. Our method can utilize multiple adversarial examples to increase model intrinsic robustness without sacrificing the performance on normal data. We evaluate our method on text classification and entailment tasks. Experimental results on different text encoders (BERT, LSTM and CNN) with multi-level attack show that our method outperforms adversarial training consistently.

There is a great need for effective summarization methods to absorb the key points of large amounts of opinions expressed on the Web. In this paper, we study the problem of opinionated text summarization, which aims to generate a coherent summary for a set of opinionated texts towards a specific topic (e.g., a movie or a controversial issue). The main characteristic of this problem is that the input set contains an arbitrary number of texts, which brings about redundant opinions and useless texts. Further, informative opinions to be summarized are scattered over different opinionated texts, thus it is vital to avoid focusing only on partial opinions. However, previous work can not tackle the above two issues effectively. To address such issues, we propose a two-stage graph-to-sequence learning framework for summarizing opinionated texts. The first stage selects summary-worthy texts from all input opinionated texts, and we construct an opinion relation graph to help estimate salience via exploiting the relationships among the input texts. Given the selected texts, the second stage generates an opinion summary via a maximal marginal relevance guided graph-to-sequence model, which gives consideration to both salient and non-redundant opinions. Experimental results on two benchmark datasets show that our framework outperforms the existing state-of-the-art methods. Human evaluation further verifies that our framework can generate more informative and compact opinion summaries than previous methods.

Emotion-cause pair extraction aims to extract all emotion clauses coupled with their cause clauses from a given document.Previous work employs two-step approaches, in which the first step extracts emotion clauses and cause clauses separately, and the second step trains a classifier to filter out negative pairs.However, such pipeline-style system for emotion-cause pair extraction is suboptimal because it suffers from error propagation and the two steps may not adapt to each other well.In this paper, we tackle emotion-cause pair extraction from a ranking perspective, i.e., ranking clause pair candidates in a document, and propose a onestep neural approach which emphasizes interclause modeling to perform end-to-end extraction.It models the interrelations between the clauses in a document to learn clause representations with graph attention, and enhances clause pair representations with kernel-based relative position embedding for effective ranking.Experimental results show that our approach significantly outperforms the current two-step systems, especially in the condition of extracting multiple pairs in one document.Clause I.

Sarcasm is a sophisticated linguistic phenomenon to express the opposite of what one really means.With the rapid growth of social media, multimodal sarcastic tweets are widely posted on various social platforms.In multimodal context, sarcasm is no longer a pure linguistic phenomenon, and due to the nature of social media short text, the opposite is more often manifested via cross-modality expressions.Thus traditional text-based methods are insufficient to detect multimodal sarcasm.To reason with multimodal sarcastic tweets, in this paper, we propose a novel method for modeling cross-modality contrast in the associated context.Our method models both cross-modality contrast and semantic association by constructing the Decomposition and Relation Network (namely D&R Net).The decomposition network represents the commonality and discrepancy between image and text, and the relation network models the semantic association in cross-modality context.Experimental results on a public dataset demonstrate the effectiveness of our model in multimodal sarcasm detection.

Despite extensive research on cross-modal retrieval, existing methods focus on the matching between image objects and text words. However, for the large amount of social media, such as news reports and online posts with images, previous methods are insufficient to model the associations between long text and image. As long text contains multiple entities and relationships between them, as well as complex events sharing a common scenario of the text, it poses unique research challenge to cross-modal retrieval. To tackle the challenge, in this paper, we focus on the retrieval task on long text and image, and propose an event-driven network for cross-modal retrieval. Our approach consists of two modules, namely the contextual neural tensor network (CNTN) and cross-modal matching network (CMMN). The CNTN module captures both event-level and text-level semantics of the sequential events extracted from a long text. The CMMN module learns a common representation space to compute the similarity of image and text modalities. We construct a multimodal dataset based on the news reports in People's Daily. The experimental results demonstrate that our model outperforms the existing state-of-the-art methods and can provide semantic richer text representations to enhance the effectiveness in cross-modal retrieval.

The massive spread of false information has brought about severe security-related problems to individuals and society. To debunk misinformation automatically, fact checking has become an important task that aims at retrieving evidence from external sources to verify the truthfulness of a given claim. As knowledge graph (KG) is a classic external source for retrieving relevant evidence. Previous methods typically check a claim by making inferences over it. Entity category information can be utilized to strengthen both the learning and verification process. However, this information was largely ignored in previous research. To make better use of the category information, in this paper, we propose a category-based framework for improving the performance of fact checking with KGs. We first learn prototypes for each category as their representatives, and then propose a prototype-based learning technique for effectively modeling the entity dependency in KG. We further develop a prototype matching technique to explore the category-level relations between head and tail entities for more robust verification. Experimental results on two benchmark datasets and a real-world dataset show that our framework can significantly improve the reasoning abilities of KG reasoning methods on Fact Checking task.

Social emotion refers to the emotion evoked to the reader by a textual document. Compared to classical sentiment analysis conducted from the author's perspective, emotion analysis from the reader's perspective (i.e. social emotion mining) is an important task in web-based social media analytics, and particularly meaningful for security related applications. Mining the causes of social emotions, i.e. Social Emotion Cause Extraction (SECE), is a new challenging task in social emotion mining, which can help better explain the elicited social emotions in text. In this paper, we propose the SECE task for the first time, and construct a new SECE dataset to support our study. We develop the first computational method for this new task, and conduct experimental studies to evaluate the effectiveness of our proposed method based on the dataset we construct.

The popularity of online contents in social media frequently experiences ebb and flow, and thus its evolution often involves different stages, such as burst and valley. Exploring the patterns of popularity evolution, especially how burst forms and decays, and even further, predicting the trends of popularity evolution is both an important research topic and beneficial to support decision making for many applications, such as emergency management, business intelligence, and public security. Previous work on popularity prediction has focused on predicting the popularity volume of online contents, and at most, popularity burst and ignored the exploration of popularity evolution and the prediction of its stages. To fill this gap, in this paper, we propose our method for the popularity stage prediction problem both at the microscopic level and macroscopic level. At the microscopic level, we first extract multiple dynamic factors and infer future evolution stage by considering the contributions of different dynamic factors. At the macroscopic level, we extract the overall evolution patterns of popularity stages and adopt a pattern matching-based method to predict future popularity stages. We evaluate the proposed approach using tweets in SinaWeibo, the most popular Twitter-like social media platform in China. The experimental results show the effectiveness of our proposed approach in predicting popularity evolution stages.

Analyzing public attitudes plays an important role in opinion mining systems. Stance detection aims to determine from a text whether its author is in favor of, against, or neutral towards a given target. One challenge of this task is that a text may not explicitly express an attitude towards the target, but existing approaches utilize target content alone to build models. Moreover, although weakly supervised approaches have been proposed to ease the burden of manually annotating largescale training data, such approaches are confronted with noisy labeling problem. To address the above two issues, in this paper, we propose a Topic-Aware Reinforced Model (TARM) for weakly supervised stance detection. Our model consists of two complementary components: (1) a detection network that incorporates target-related topic information into representation learning for identifying stance effectively; (2) a policy network that learns to eliminate noisy instances from auto-labeled data based on off-policy reinforcement learning. Two networks are alternately optimized to improve each other’s performances. Experimental results demonstrate that our proposed model TARM outperforms the state-of-the-art approaches.

As a fundamental task of sentiment analysis, aspect-level sentiment analysis aims to identify the sentiment polarity of a specific aspect in the context. Previous work on aspect-level sentiment analysis is text-based. With the prevalence of multimodal user-generated content (e.g. text and image) on the Internet, multimodal sentiment analysis has attracted increasing research attention in recent years. In the context of aspect-level sentiment analysis, multimodal data are often more important than text-only data, and have various correlations including impacts that aspect brings to text and image as well as the interactions associated with text and image. However, there has not been any related work carried out so far at the intersection of aspect-level and multimodal sentiment analysis. To fill this gap, we are among the first to put forward the new task, aspect based multimodal sentiment analysis, and propose a novel Multi-Interactive Memory Network (MIMN) model for this task. Our model includes two interactive memory networks to supervise the textual and visual information with the given aspect, and learns not only the interactive influences between cross-modality data but also the self influences in single-modality data. We provide a new publicly available multimodal aspect-level sentiment dataset to evaluate our model, and the experimental results demonstrate the effectiveness of our proposed model for this new task.

Penghui Wei, Nan Xu, Wenji Mao. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

Emotion cause extraction aims at automatically identifying cause clauses for a certain emotion expressed in a document. It is an important task in emotion analysis since it helps form a deeper understanding of emotion text. Detecting potential causes of user emotion in online contents is beneficial to public opinion monitoring, government decision-making, and other security-related applications. Existing studies treat this task as a binary clause-level classification problem, which considers each clause separately and omits the context information of clauses. Moreover, previous work only models emotion-dependent linguistic representations of clauses but ignores emotion-independent features in clauses including cause indicators. To address the above two issues, we formalize this task as a sequence labeling problem and propose the COntext-aware Multi-View attention networks (COMV) for emotion cause extraction. Our proposed model integrates context information and learns multi-view clause representations. Experimental results show that our model outperforms existing state-of-the-art methods.

Targeted stance detection aims to classify the attitude of an opinionated text towards a pre-defined target. Previous methods mainly focus on in-target setting that models are trained and tested using data specific to the same target. In practical cases, the target we concern may have few or no labeled data, which restrains us from training a target-specific model. In this paper we study the problem of cross-target stance detection, utilizing labeled data of a source target to learn models that can be adapted to a destination target. To this end, we propose an effective method, the core intuition of which is to leverage shared latent topics between two targets as transferable knowledge to facilitate model adaptation. Our method acquires topic knowledge with neural variational inference, and further adopts adversarial training that encourages the model to learn target-invariant representations. Experimental results verify that our proposed method is superior to the state-of-the-art methods.

Sub-event detection can help faster and deeper understanding of an event by providing human-friendly clusters, and thus has become an important research topic in Web mining and knowledge management. In existing sub-event detection methods, clustering based methods are brittle for using heuristic similarity metric to judge whether documents belong to the same sub-event, while topic model based methods are limited to the bag of words assumption. To overcome these drawbacks in previous research, in this paper, we propose an encoder-memory-decoder framework for sub-event detection. Our model learns document and sub-event representations suitable for the similarity metric in a data-driven manner, and transforms sub-event detection into selecting the most proper sub-event representation that can maximize text reconstruction probability. Considering the case of over-fitting, we also apply transfer learning in our model. To the best of our knowledge, our model is the first to develop an unsupervised deep neural model for sub-event detection. We use Twitter as an examplar social media platform for our study, and experimental results show that our model outperforms baseline methods for sub-event detection.

Exploring user stances and attitudes is beneficial to a number of Web related research and applications, especially in social media platforms such as Twitter. Stance detection in Twitter aims at identifying the stance expressed in a tweet towards a given target (e.g., a government policy). A key challenge of this task is that a tweet may not explicitly express opinion about the target. To effectively detect user stances implied in tweets, target content information plays an important role. In previous studies, conventional feature-based methods often ignore target content. Although more recent neural network-based methods attempt to integrate target information using attention mechanism, the performance improvement is rather limited due to the underuse of this information. To address this issue, we propose an endto- end neural model, TGMN-CR, which makes better use of target content information. Specifically, our model first learns conditional tweet representation with respect to specific target. It then employs a target-guided iterative process to extract crucial stance-indicative clues via multiple interactions between target and tweet words. Experimental results on SemEval-2016 Task 6.A Twitter Stance Detection dataset show that our proposed method outperforms the state-of-the-art alternative methods, and substantially outperforms the comparative methods when a tweet does not explicitly express opinion about the given target.

With the continuous development of social media and the diversity of its contents, rumors can quickly spread widely and cause serious damage, which makes the early detection of rumors a key task in social media analysis. Most existing methods for rumor detection exploit handcrafted features and ignore the different roles original post and retweets play in information production and spreading process. In fact, original post usually contains detailed content information about event itself, while retweets reflect the diffusion of information in online interaction. Recently, several studies manifest the effectiveness of deep neural network in rumor detection. However, these previous research has made no distinction between original post and retweets for rumor detection. In this paper, we propose a Merged Neural Rumor Detection (MNRD) model that consider three aspects of rumor data: content of original post, diffusion of retweet and user information. We propose content attention to aggregate pivotal words in original post and diffusion attention to focus on most informative retweets during the diffusion process. We also incorporate user information by identifying features to capture user's reliability and social influence. Experimental results on real world dataset show the effectiveness of our model to detect rumors in social media.

Social media platforms make the spread of social event information quicker and more convenient. Some of these social events may become hot topics, which highlights the importance of event popularity prediction in public management, decision making and other security related applications. Due to the complexity of social event itself, it has two unique characteristics which most previous popularity prediction work has ignored: (1) the discussion of an event itself may consist of several components, e.g. different sub-events, different stances or different user communities; (2) the popularity of an event can be influenced by other related events. To address its unique characteristics, we propose an event popularity prediction model combining partition and interaction. We employ reinforcement learning to automatically partition an event into components and recognize related events. Then we predict event popularity by modeling component information and interactions between related events. Experimental results on a real world dataset show that our proposed model can outperform the competitive baseline methods.

Event detection for social media is an important social media analytics task in security domain, which can provide valuable information for decision making, intelligence analysis and public management. Traditional event detection methods either rely on simple term frequency based features, or take the bag of words assumption. Recently, some deep neural network based event detection methods are proposed. However, these methods still have some drawbacks. Firstly, they do not provide an effective way to learn the connection between message representation and event representation, but simply use average of message representations as the event representation. Secondly, representations in the hidden space lack interpretability compared to the traditional event keyword representation. To deal with these weaknesses, we propose an event detection approach joint learning with keyword extraction. We provide an episode learning strategy to enable the training of event representation update. In addition, by joint learning with keyword extraction, the model is more explainable, and can achieve a better performance. As the selection of a set of keywords is a combinatorial problem and non-differential, we also employ reinforcement learning in our approach. Experiments on a public available dataset show the superiority of our approach compared with baseline methods.

Stance detection aims at inferring from text whether the author is in favor of, against, or neutral towards a target entity. Most of the existing studies consider different target entities separately. However, in many scenarios, stance targets are closely related, such as several candidates in a general election and different brands of the same product. Multi-target stance detection, in contrast, aims at jointly detecting stances towards multiple related targets. As stance expression regarding a target can provide additional information to help identify the stances towards other related targets, modeling expressions regarding multiple targets jointly is beneficial for improving the overall performance compared to single-target scheme. In this paper, we propose a dynamic memory-augmented network DMAN for multi-target stance detection. DMAN utilizes a shared external memory, which is dynamically updated through the learning process, to capture and store stance-indicative information for multiple related targets. It then jointly predicts stances towards these targets in a multitask manner. Experimental results show the effectiveness of our DMAN model.

With the rapid increase of diversity and modality of data in user-generated contents, sentiment analysis as a core area of social media analytics has gone beyond traditional text-based analysis. Multimodal sentiment analysis has become an important research topic in recent years. Most of the existing work on multimodal sentiment analysis extracts features from image and text separately, and directly combine them to train a classifier. As visual and textual information in multimodal data can mutually reinforce and complement each other in analyzing the sentiment of people, previous research all ignores this mutual influence between image and text. To fill this gap, in this paper, we consider the interrelation of visual and textual information, and propose a novel co-memory network to iteratively model the interactions between visual contents and textual words for multimodal sentiment analysis. Experimental results on two public multimodal sentiment datasets demonstrate the effectiveness of our proposed model compared to the state-of-the-art methods.

People often publish online texts to express their stances, which reflect the essential viewpoints they stand. Stance identification has been an important research topic in text analysis and facilitates many applications in business, public security and government decision making. Previous work on stance identification solely focuses on classifying the supportive or unsupportive attitude towards a certain topic/entity. The other important type of stance identification, multiple stance identification, was largely ignored in previous research. In contrast, multiple stance identification focuses on identifying different standpoints of multiple parties involved in online texts. In this paper, we address the problem of recognizing distinct standpoints implied in textual data. As people are inclined to discuss the topics favorable to their standpoints, topics thus can provide distinguishable information of different standpoints. We propose a topic-based method for standpoint identification. To acquire more distinguishable topics, we further enhance topic model by adding constraints on document-topic distributions. We finally conduct experimental studies on two real datasets to verify the effectiveness of our approach to multiple stance identification.

With the prevalence of more diverse and multiform user-generated content in social networking sites, multimodal sentiment analysis has become an increasingly important research topic in recent years. Previous work on multimodal sentiment analysis directly extracts feature representation of each modality and fuse these features for classification. Consequently, some detailed semantic information for sentiment analysis and the correlation between image and text have been ignored. In this paper, we propose a deep semantic network, namely MultiSentiNet, for multimodal sentiment analysis. We first identify object and scene as salient detectors to extract deep semantic features of images. We then propose a visual feature guided attention LSTM model to extract words that are important to understand the sentiment of whole tweet and aggregate the representation of those informative words with visual semantic features, object and scene. The experiments on two public available sentiment datasets verify the effectiveness of our MultiSentiNet model and show that our extracted semantic features demonstrate high correlations with human sentiments.

Online interaction behavior between web users often makes some events go viral. Popularity prediction of events is a key task in many security related applications. It forecasts how widely events would spread based on the information of evolution at an early stage. Existing methods either rely on careful feature engineering, or solely consider time series, ignoring rich information of user and text content. In this paper, we attempt to extract and fuse the rich information of text content, user and time series in a data-driven fashion. To this end, we design a popularity prediction model based on deep neural networks, which uses three encoders to extract high-level representation of text content, users and time series respectively. In addition, we incorporate attention mechanism to make our model focus on important features. Experiments on real world dataset show the effectiveness of our proposed model.

The ever-growing number of users makes social media a valuable information source about recent events. Event detection and tracking plays an important role in decision-making and public management. Despite recent progress, the performance of event detection and tracking is still limited. The majority of existing work lacks an effective way to judge whether a text related to a certain event, due to the limitations of semantic representation and heuristic similarity metric. In this paper, we present an online event detection and tracking method based on similarity metric learning using neural network. Our method first trains a classification model to identify event related texts. To detect and track events, we adopt a clustering-based approach. Specifically, we use neural network to jointly learn a similarity metric and low dimension representation of events, and then use a memory module to store and update event representation. Experiments on Twitter dataset show the effectiveness of our proposed method.

Given a collection of event-related documents, event ranking generates a list of ranked events based on the input query. Ranking news events, which takes event related news documents for the generation of ranked events, is both an essential research issue and important component for many security oriented applications, such as public event monitoring, retrieval, detection and mining. Previous related work solely relies on queries of event relevant aspects, and user relevant aspects of queries that are critical for security applications are totally ignored. In this paper, we deal with the problem of news ranking by incorporating user relevant information into the input query, from the cluster of relevant new documents and comments. Given an input query, which contains event related objective aspects(e.g. actors, locations, date) and user related subjective aspects(e.g. public attention and opinion polarity), we develop a Learning-to-Rank framework to integrate aspect-level correlation between query and event. Experiments on a crawled large news corpus show the effectiveness of our proposed approach compared to several baseline models.

The competitive perspective implied in online texts reflect people's conflicts in their stances and viewpoints. Competitive perspective identification aims to determine people's inclinations to one of multiple competitive perspectives, which is an important research issue and can facilitate many security-related applications. As the word usage of different perspectives is distinct in various topics, in this paper, we first proposes a supervised topic-refined method for competitive perspective identification. Our method refines perspective classifiers with the document-topic distributions mined from texts. To reduce human labor in data annotation, we further extend our work in a semi-supervised manner and propose a user-based bootstrapping framework. As the perspectives people hold are relatively stable, our bootstrapping process leverages the user-level perspective consistency to select high-quality classified texts from unlabeled corpus and boost the perspective classifier iteratively. Experimental studies show the effectiveness of our proposed approach in identifying the competitive perspectives of online texts.

Social media has become an important platform for people to express opinions, share information and communicate with others. Detecting and tracking topics from social media can help people grasp essential information and facilitate many security-related applications. As social media texts are usually short, traditional topic evolution models built based on LDA or HDP often suffer from the data sparsity problem. Recently proposed topic evolution models are more suitable for short texts, but they need to manually specify topic number which is fixed during different time period. To address these issues, in this paper, we propose a nonparametric topic evolution model for social media short texts. We first propose the recurrent semantic dependent Chinese restaurant process (rsdCRP), which is a nonparametric process incorporating word embeddings to capture semantic similarity information. Then we combine rsdCRP with word co-occurrence modeling and build our short-text oriented topic evolution model sdTEM. We carry out experimental studies on Twitter dataset. The results demonstrate the effectiveness of our method to monitor social media topic evolution compared to the baseline methods.

Mining topics in short texts (e.g. tweets, instant messages) can help people grasp essential information and understand key contents, and is widely used in many applications related to social media and text analysis. The sparsity and noise of short texts often restrict the performance of traditional topic models like LDA. Recently proposed Biterm Topic Model (BTM) which models word co-occurrence patterns directly, is revealed effective for topic detection in short texts. However, BTM has two main drawbacks. It needs to manually specify topic number, which is difficult to accurately determine when facing new corpora. Besides, BTM assumes that two words in same term should belong to the same topic, which is often too strong as it does not differentiate two types of words (i.e. general words and topical words). To tackle these problems, in this paper, we propose a non-parametric topic model npCTM with the above distinction. Our model incorporates the Chinese restaurant process (CRP) into the BTM model to determine topic number automatically. Our model also distinguishes general words from topical words by jointly considering the distribution of these two word types for each word as well as word coherence information as prior knowledge. We carry out experimental studies on real-world twitter dataset. The results demonstrate the effectiveness of our method to discover coherent topics compared with the baseline methods.

People write online documents from different personal perspectives. The competitive perspectives they hold reflect the conflicts in their fundamental stances and viewpoints. For many security-related applications, it is both beneficial and critical to identify the competitive perspectives implied in online documents. Previous work on competitive perspective identification is based on word features, which did not consider that the word usage for perspective expression varies with topics in documents. Thus topic information can be incorporated and contribute to a more fine-grained treatment of perspective identification. Motivated by this, this paper proposes an approach for competitive perspective identification in online documents via topic based refinement. Our approach refines the basic word feature-based perspective identification model with latent semantic information. In addition, we develop a self-adaptive process to fit the model parameters automatically. Experimental study shows the effectiveness of our approach compared to the related work and the baseline methods.

In recent years, microblog has become one of the most widely used social media for people to exchange ideas and express emotions. As information propagates fast in social network, it's crucial for governments and public agencies to effectively monitor public sentiment implied in user-generated content. Most previous work of public sentiment analysis takes tweets of different users as a whole without considering the diverse word use of people. Thus, some sentiment words may be neglected in the process of analysis because they are only used by people of specific groups. Inspired by previous psychological findings that personality influences the ways people write and talk, we propose a personality based sentiment classification method. In order to capture more useful but not widely used sentiment words, our approach extracts textual features for people of different personality traits based on the Big Five model. Moreover, we adopt an ensemble learning strategy to utilize both personality related and commonly used textual features. Experimental study shows the effectiveness of our method.

Among the most important and distinctive actionable knowledge are actionable behavioral rules that can directly and explicitly suggest specific actions to take to influence (restrain or encourage) the behavior in the users' best interest. The problem of mining such rules is a search problem in a framework of support and expected utility. The previous definition of a rule's support assumes that each instance which supports a rule has the uniform contribution to the support. However, this assumption is usually violated in practice to some extent, and thus will hinder the performance of algorithms for mining such rules. In this paper, to handle this problem, a power-function-based observation-weighting model for support and corresponding mining algorithm are proposed. The experimental results strongly suggest the validity and the superiority of our approach.

In security-related applications, it is of great need to construct topic hierarchies from the data automatically. It can help us better understand the contents and structure of information and benefit many applications in security informatics. The existing topic hierarchy construction methods either need to specify the structure manually, or are not robust enough for sparse social media data such as microblog. In this paper, we propose an approach to automatically construct topic hierarchies from microblog data in a bottom up manner. We first detect the fine grained topics and then build the topic structure based on a multi-branch hierarchical construction method. We conduct a preliminary empirical study based on Weibo data. The experimental results show that the hierarchical topic structure generated by our method can provide meaningful results.

Action knowledge is an important type of behavioral knowledge and of vital importance to many applications in social computing, especially in behavior modeling, analysis and prediction. In this paper, we present a computational method to action knowledge extraction from online media. Our approach is based on mutual bootstrapping and combined with knowledge reasoning. Compared with the related work, our approach can acquire more types of action knowledge, and needs much less human labor. We evaluate the performance of our method using the Web textual data from security informatics domain. The experimental results show the effectiveness of our proposed method.

Extracting emotions from online reviews is crucial to many security-related applications as well as applications in other domains. Traditional approaches to emotion extraction have mainly focused on mining the polarities of opinions or using annotated data to extract emotion types. Emotion theories, which identify the underlying cognitive structure and emotional dimensions that are key to generate emotions, have almost been totally ignored in previous work. To facilitate the automatic extraction of emotions from textual data, in this paper, we propose an emotion model based approach to emotion extraction from online reviews. Informed by the widely used OCC emotion model, we employ a statistical method to extract emotion words with their dimension values from texts, and implement OCC model to obtain emotions based on the emotion-dimension dictionary. We conduct an empirical study using security-related news reviews. The experimental results demonstrate the effectiveness of our proposed approach.

Social networking sites provide a convenient way for users to participate in discussion groups and communicate with others. While users situate in and enjoy such a social environment, it is important for various security related applications to understand, model and analyze participating users' behavior. In this paper, we make an attempt to model and predict user participation behavior in discussion groups of social networking sites. Our work employs a feature-based approach, which considers four types of features: thread features, content similarity, user behavior and social features. We conduct an empirical study on a popular social networking site in China, Douban.com. The experimental results show the effectiveness of our approach.

Actions are the primary way an entity interacts with other entities and acts on the external world. Action knowledge is of vital importance for behavior modeling, analysis and prediction in security informatics. In this paper, we present our approach to action knowledge extraction from Web textual data. Our approach is based on mutual bootstrapping with knowledge reasoning, which can acquire more action knowledge types and require less human participation compared with the related work. We evaluate the performance of our method and demonstrate its effectiveness through experiment.

Opinion mining has gained increasing attention and shown great practical value in recent years. Existing research on opinion mining mainly focuses on the extraction of lexicon orientation and opinion targets. The explanations of opinions, which are potentially valuable for many applications, are totally ignored. To address this specific research challenge, in this paper, we propose an approach to extract the explanation of reason and/or consequence behind an opinion via learning word pairs and using causal indicators from Chinese online reviews. We also improve our word pair based method by constructing clusters of word paris. Experiments on a Chinese business review corpus show that our method is feasible and effective.

Food safety events are typical public security events that draw great public concern. In food safety events, millions of netizens pay close attention to the event, express their opinions online and thus influence the decisions of government or food producers. Modeling netizen groups, especially the dynamics of their opinions in these events, can help us understand the mechanism and evolvement of such events and provide valuable insights for social management. However, conventional computational models, such as agent-based models, are usually constructed manually. In this paper, we propose an approach to acquiring netizen group's opinions from online comments to facilitate the modeling of food safety events. We conduct experimental study on typical events happened in China and empirically evaluate the performance of our proposed approach. The results verify the effectiveness of the approach.

Social causality is the inference an entity makes about the social behavior of other entities and self. Besides physical cause and effect, social causality involves reasoning about epistemic states of agents and coercive circumstances. Based on such inference, responsibility judgment is the process whereby one singles out individuals to assign responsibility, credit or blame for multi-agent activities. Social causality and responsibility judgment are a key aspect of social intelligence, and a model for them facilitates the design and development of a variety of multi-agent interactive systems. Based on psychological attribution theory, this paper presents a domain-independent computational model to automate social inference and judgment process according to an agent’s causal knowledge and observations of interaction. We conduct experimental studies to empirically validate the computational model. The experimental results show that our model predicts human judgments of social attributions and makes inferences consistent with what most people do in their judgments. Therefore, the proposed model can be generically incorporated into an intelligent system to augment its social and cognitive functionality.

Forecasting group behavior is critical to national and international security. Various forecasting methods have been developed previously. However, the majority of them are data-driven methods and rely heavily on the structured data which are often hard to obtain. To overcome the limitation of previous methods, we propose a novel plan recognition method for detecting multiple group behavior based on graph search. We further conduct human experiments in security informatics domain to empirically evaluate our proposed method. The experimental results show the effectiveness of our method.

Many security-related applications can benefit from constructing models to predict the behavior of an entity. However, such models do not provide the user with explicit knowledge that can be directly used to influence the behavior for his/her interest. This type of knowledge is called actionable knowledge. Actionability is a very important aspect of the interestingness of mined patterns. In this paper, we formally define a new problem of mining actionable behavioral rules from group data. We also propose an algorithm for solving the problem. Using terrorism group data, our experiment shows the validity of our approach as well as the practical value of our defined problem in security informatics.

Listwise approach is an important method to solve practical Web search problem in learning to rank. In this paper, we first analyze the practical Web search problem and construct the model to solve it. Then we propose an algorithm called DiffRank which can apply boosting technology to learning to rank in listwise. Through theoretical analysis, we prove that the upper bound of training error can be reduced in our proposed algorithm. The experimental results further verify our theoretical analysis and demonstrate that our approach can better perform in practical Web search than other state-of-the-art listwise algorithms.

Terrorism organizations are devising increasingly sophisticated plans to conduct attacks. The ability of emulating or constructing attack plans by potential terrorists can help us understand the intents and motivation behind terrorism activities. A feasible computational method to construct plans is planning technique in AI. Traditionally, AI planning methods rely on a predefined domain theory which is compiled by domain experts manually. To facilitate domain theory construction and plan generation, we propose a method to construct domain theory automatically from free text data. The effectiveness of our proposed approach is evaluated empirically through experimental studies using real world terrorist plans .

Cultural modeling is an emergent and promising research area in social computing. It aims to develop behavioral models of groups and analyze the impact of culture factors on group behavior using computational methods. Classification methods play a critical role in cultural modeling domain. As various cultural-related datasets possess different properties, for group behavior prediction, it is important to gain a computational understanding of the performance of various classification methods. In this paper, we investigate the performance of seven representative classification algorithms using a benchmark cultural modeling dataset and analyze the experimental results.

Cultural modeling is an emergent and promising research area in social computing. It aims at developing behavioral models of groups and analyzing the impact of culture factors on group behavior using computational methods. Machine learning methods in particular classification, play a central role in such applications. In cultural modeling, it is expected that classifiers yield good performance. However, the performance of standard classifiers is often severely hindered in practice due to the imbalanced distribution of class in cultural data. In this paper, we identify class imbalance problem in cultural modeling domain. To handle the problem, we propose a user involved solution employing the receiver operating characteristic (ROC) analysis for classification algorithms with sampling approaches. Finally, we conduct experiment to verify the effectiveness of the proposed solution.