Search

With the growing availability and popularity of sentiment-rich resources like blogs and online reviews, new opportunities and challenges have emerged regarding the identification, extraction, and organization of sentiments from user-generated documents or sentences. Recently, many studies have exploited lexicon-based methods or supervised learning algorithms to separately conduct sentiment analysis tasks; however, the former approaches ignore contextual information of sentences and the latter ones do not take sentiment information embedded in sentiment words into consideration. To tackle these limitations, we propose a new model named Sentiment Convolutional Neural Network (SentiCNN) to analyze the sentiments of sentences with both contextual and sentiment information of sentiment words, in which, contextual information is captured from word embeddings and sentiment information is identified using existing lexicons. We incorporate a Highway Network into our model to adaptively combine sentiment and contextual information from sentences by strengthening the connection between features of both sentences and their sentiment words. Furthermore, we propose three lexicon-based attention mechanisms (LBAMs) for our SentiCNN model to find the most important indicators of sentiments and make predictions more effectively. Experiments over two well-known datasets indicate that sentiment words, the Highway Network, and LBAMs contribute to sentiment analysis.

The parallel Hierarchical Dirichlet Process (pHDP) is an efficient topic model which explores the equivalence of the generation process between Hierarchical Dirichlet Process (HDP) and Gamma-Gamma-Poisson Process (G2PP), in order to achieve parallelism at the topic level. Unfortunately, pHDP loses the non-parametric feature of HDP, i.e., the number of topics in pHDP is predetermined and fixed. Furthermore, under the bootstrap structure of pHDP, the topic-indiscriminate words are of high probabilities to be assigned to different topics, resulting in poor qualities of the extracted topics. To achieve parallelism without sacrificing the non-parametric feature of HDP, in addition to improve the quality of extracted topics, we propose a parallel dynamic topic model by developing an adjustment mechanism of evolving topics and reducing the sampling probabilities of topic-indiscriminate words. Both supervised and unsupervised experiments on benchmark datasets show the competitive performance of our model.

Topic models have been widely used for learning the latent explainable representation of documents, but most of the existing approaches discover topics in a flat structure. In this study, we propose an effective hierarchical neural topic model with strong interpretability. Unlike the previous neural topic models, we explicitly model the dependency between layers of a network, and then combine latent variables of different layers to reconstruct documents. Utilizing this network structure, our model can extract a tree-shaped topic hierarchy with low redundancy and good explainability by exploiting dependency matrices. Furthermore, we introduce manifold regularization into the proposed method to improve the robustness of topic modeling. Experiments on real-world datasets validate that our model outperforms other topic models in several widely used metrics with much fewer computation costs.

Prediction of protein–protein interaction sites plays an important role for understanding the protein interactions and functions. However, in the protein–protein interaction site prediction problem, the number of binding-site residues is usually much less than that of other amino acid residues in a protein chain, which would lead to the performance drop of standard machine learning methods on minority class, i.e., the binding-site residues. Therefore, to improve the prediction performance on binding-site residues, we propose in this paper a new boosting algorithm (SSTBoost) that consists of stochastic sensitivity measure-based undersampling method and AdaBoost algorithm. Stochastic sensitivity measure-based undersampling method aims to re-balance the dataset by selecting those samples with the highest probability to be incorrectly labeled, and AdaBoost algorithm aims to improve the performance of base hypotheses by making them to be complementary and be conjunction with each other. Twenty UCI datasets are first used to evaluate the robustness and effectiveness of the SSTBoost. After that, the SSTBoost is tested on twenty-two practical protein–protein interaction sites prediction problems. Experimental results show that the SSTBoost significantly improves the performances against state-of-the-art methods by $\text{57.3}\%$ , $\text{88.2}\%$ , and $\text{78.2}\%$ out of 110 cases in terms of Recall, F-score, and G-mean, respectively. This shows its potential to handle other bioinformatic applications in near future.

Chinese word embeddings have recently garnered considerable attention. Chinese characters and their sub-character components, which contain rich semantic information, are incorporated to learn Chinese word embeddings. Chinese characters can represent a combination of meaning, structure, and pronunciation. However, existing embedding learning methods focus on the structure and meaning of Chinese characters. In this study, we aim to develop an embedding learning method that can make complete use of the information represented by Chinese characters, including phonology, morphology, and semantics. Specifically, we propose a pronunciation-enhanced Chinese word embedding learning method, where the pronunciations of context characters and target characters are simultaneously encoded into the embeddings. Evaluation of word similarity, word analogy reasoning, text classification, and sentiment analysis validate the effectiveness of our proposed method.

The application of artificial intelligence (AI) technologies in assisting human electroencephalogram (EEG) analysis has become an active scientific field. This study aims to present a comprehensive review of the research field of AI-enhanced human EEG analysis. Using bibliometrics and topic modeling, research articles concerning AI-enhanced human EEG analysis collected from the Web of Science database during the period 2009–2018 were analyzed. After examining 2053 research articles published around the world, it was found that the annual number of articles had significantly grown from 78 to 468, with the USA and China being the most influential and prolific. The results of the keyword analysis showed that “electroencephalogram,” “brain–computer interface,” “classification,” “support vector machine,” “electroencephalography,” and “signal” were the most frequently used. The results of topic modeling and evolution analyses highlighted several important issues, including epileptic seizure detection, brain–machine interface, EEG classification, mental disorders, emotion, and alcoholism and anesthesia. The findings suggest that such visualization and analysis of the research articles could provide a comprehensive overview of the field for communities of practice and inquiry worldwide.

With the advances and development of imaging and computer technologies, the application of artificial intelligence (AI) in the processing of magnetic resonance imaging (MRI) data has become a significant research field. Based on 2572 research articles concerning AI-enhanced brain MRI processing, this study provides a latent Dirichlet allocation based bibliometric analysis for the exploration of the status, trends, major research issues, and potential future directions of the research field. The trend analyses of articles and citations demonstrate a flourishing and increasing impact of the research. Neuroimage is the most prolific and influential journal. The USA and University College London have contributed the most to the research. The collaboration between European countries is very close. Essential research issues such as Image segmentation, Mental disorder, Functional network connectivity, and Alzheimer’s disease have been uncovered. Potential inter-topic research directions such as Functional network connectivity and Mental disorder, Image segmentation and Image classification, Cognitive impairment and Diffusion imaging, as well as Sense and memory and Emotion and feedback, have been highlighted.

Knowledge graphs, as linked data, can be extracted from texts in triple form that illustrate the structure of “entity–relation–entity.” Knowledge graph embedding (KGE) models are used to map entities and relations into a continuous vector space with semantic constraints so as to learn a knowledge graph with fact triples. In the KGE model training process, both positive and negative triples are necessarily provided. Thus, negative sampling methods are meaningful in generating negative samples based on the representations of entities and relations. This paper proposes an innovative neighborhood knowledge selective adversarial network (NKSGAN), which leverages the representation of aggregating neighborhood information to generate high-quality negative samples for enhancing the performances of the discriminator. Experiments are conducted on widely used standard datasets such as FB15k, FB15k-237, WN18 and WN18RR to evaluate our model for link prediction task. The results present the superiority of our proposed NKSGAN than other baseline methods, indicating that the negative sampling process in NKSGAN is effective in generating high-quality negative samples for boosting KGE models.

Sentiment strength detection is an essential task in sentiment analysis, wherein the sentiment strength of subjective text is automatically determined. Sentiment analysis has numerous applications in different sectors, including business and social domains. In this study, we present a model to effectively extract the features and strength of sentiment from words and text using a context-dependent, lexicon-based convolutional neural network. To build this convolutional neural network, the model is trained using the sentiment polarity for each word from a co-occurrence pattern of words and labels. Then, a context-dependent lexicon is generated from the corpus, which is used to generate positive and negative sentiment word embeddings. Positive sentiment word embeddings, negative sentiment word embeddings, and the pre-trained word embeddings are input to a 3-channel convolutional neural network (CNN) to predict the strength of the sentiments. Moreover, with the trained convolutional neural network model, we can obtain a learned sentiment strength-specific word embedding, which generates a sentiment strength-specific lexicon (SSS-Lex) that contains word associations and sentiment intensity scores. To validate the effectiveness of sentiment strength detection in the proposed model, we evaluate the model using six real-world datasets. Furthermore, to evaluate the sentiment strength-specific lexicon, we compare it with seven existing lexicons in three evaluation tasks from the SemEval-2015 and SemEval-2016 competitions. Experimental results indicate that the proposed model can predict the sentiment strength of documents more effectively than the baseline methods, and that the SSS-Lex is of higher quality than the existing lexicons.

In this paper, we propose a multimodal fusion network (termed as MFN) to integrate the text and image data from social media for rumor detection. Given the multimodal features, MFN exploits self-attentive fusion (SAF) mechanism to conduct feature-level fusion by assigning corresponding weights to the complementary modalities. In particular, the textual features are combined with the fused features in a skip-connection manner, as textual features tend to be more distinguishable compared with visual features. Furthermore, MFN introduces latent topic memory (LTM) to store the semantic information about rumor and non-rumor events, benefiting to the identification of the upcoming posts. Extensive experiments conducted on two public datasets show that the proposed MFN outperforms the state-of-the-art approaches.

Network embedding plays an important role in various real-world applications. Most traditional algorithms focus on the topological structure while ignore the information from node attributes. The attributed information is potentially valuable to network embedding. To solve this problem, we propose a deep learning model named Extractive Convolutional Adversarial Network (ECAN) for network embedding. This model aims to extract the latent representations from the topological structure, the attributed information, and labels via three components. In the first part, ECAN extracts features from the topological structure and the attributed information of nodes separately. The second part is a prediction model, which aims to exploit labels of vertices. The third part is a convolutional adversarial model. We train it to distinguish the extractive features which are generated by the hidden layers in the extractive network from either the attributed information or the topological structure. Experiments on six real-world datasets demonstrate the effectiveness of ECAN when compared with state-of-the-art embedding algorithms.

Traditional methods of annotating the sentiment of an unlabeled document are based on sentiment lexicons or machine learning algorithms, which have shown low computational cost or competitive performance. However, these methods ignore the semantic composition problem displaying in several ways such as negative reversing and intensification. In this paper, we propose a new method for sentiment classification using negative and intensive sentiment supplementary information, so as to exploit the linguistic feature of negative and intensive words in conjunction with the context information. Particularly, our method can solve the domain-specific problem without relying on the external sentiment lexicons. Experimental results on two real-world datasets demonstrate the effectiveness of our proposed method.

With the rapid development of the Internet, an increasing number of users enjoy to shop online and express their reviews on the products and services. Analysis of these online reviews can not only help potential users make rational decisions when purchasing but also improves the quality of products and services. Hence, sentiment analysis for online reviews has become an important and meaningful research domain.

As the rapid development of Web 2.0 communities, social media service providers offer users a convenient way to share and create their own contents such as online comments, blogs, microblogs/tweets, etc. Understanding the latent emotions of such short texts from social media via the computational model is an important issue as such a model will help us to identify the social events and make better decisions (e.g., investment in stocking market). However, it is always very challenge to detect emotions from above user-generated contents due to the sparsity problem (e.g., a tweet is a short message). In this article, we propose an universal affective model (UAM) to classify readers’ emotions over unlabeled short texts. Different from conventional text classification model, the UAM structurally consists of topic-level and term-level sub-models, and detects social emotions from the perspective of readers in social media. Through the evaluation on real-world data sets, the experimental results validate the effectiveness of the proposed model in terms of the effectiveness and accuracy.

Automatic and accurate segmentation of breast lesion regions from ultrasonography is an essential step for ultrasound-guided diagnosis and treatment. However, developing a desirable segmentation method is very difficult due to strong imaging artifacts e.g., speckle noise, low contrast and intensity inhomogeneity, in breast ultrasound images. To solve this problem, this paper proposes a novel boundary-guided multiscale network (BGM-Net) to boost the performance of breast lesion segmentation from ultrasound images based on the feature pyramid network (FPN). First, we develop a boundary-guided feature enhancement (BGFE) module to enhance the feature map for each FPN layer by learning a boundary map of breast lesion regions. The BGFE module improves the boundary detection capability of the FPN framework so that weak boundaries in ambiguous regions can be correctly identified. Second, we design a multiscale scheme to leverage the information from different image scales in order to tackle ultrasound artifacts. Specifically, we downsample each testing image into a coarse counterpart, and both the testing image and its coarse counterpart are input into BGM-Net to predict a fine and a coarse segmentation maps, respectively. The segmentation result is then produced by fusing the fine and the coarse segmentation maps so that breast lesion regions are accurately segmented from ultrasound images and false detections are effectively removed attributing to boundary feature enhancement and multiscale image information. We validate the performance of the proposed approach on two challenging breast ultrasound datasets, and experimental results demonstrate that our approach outperforms state-of-the-art methods.

The task of entity linking aims to correctly link mentions in a text fragment to a reference knowledge base. Most existing methods apply single neural network model to learn semantic representations on all granularities in contextual information, which neglecting the trait of different granularities. Also, these solely representation-based methods measure the semantic matching based on the abstract vector representation that frequently miss concrete matching information. To better capture contextual information, this paper proposes a new neural network model called Hybrid Semantic Matching (HSM) for the entity linking task. The model captures two different aspects of semantic information via representation and interaction-based neural semantic matching models. Furthermore, to consider the global consistency of entities, a recurrent random walk is applied to propagate entity linking evidences among related decisions. Evaluation was conducted on three publicly available standard datasets. Results show that our proposed HSM model is more effective compared with a list of baseline models.

Massive Open Online Courses (MOOCs), which are open for anyone without limitations on time or location, have attracted millions of registered online students. The large number of online courses available raises the question of how appropriate courses can be effectively recommended to interested learners. The recommendation system, widely used in various online applications, is a good solution for reducing decision complexity. In this paper, we propose the method of using attention-based convolutional neural networks (CNN) to obtain a user's profile, predict the user ratings, and recommend the top-n courses. First, we represent the learner behaviors and learning histories into feature vectors. The attention mechanism is then used to improve relevance estimation according to the differences between the estimation scores and the actual scores given by users to train the neural network. Finally, the trained model will recommend courses to learners. At the end of the paper, we introduce the framework of our system.

Combining topological information and attributed information of nodes in networks effectively is a valuable task in network embedding. Nevertheless, many prior network embedding methods regarded attributed information of nodes as simple attribute sets or ignored them totally. In some scenarios, the hidden information contained in vertex attributes are essential to network embedding. For instance, networks that contain vertexes with text information play an increasingly important role in our life, including citation networks, social networks, and entry networks. In these textual networks, the latent topic relevance information of different vertexes contained in textual attributes information are valuable in the network analysis process. Shared latent topics of nodes in networks may influence the interaction between them, which is critical to network embedding. However, much prior work for textual network embedding only regarded the text information as simple word sets while ignored the embedded topic information. In this paper, we develop a model named Topical Adversarial Capsule Network (TACN) for textual network embedding, which extracts a low-dimensional latent space of the original network from node structures, vertex attributes, and topic information contained in text of nodes. The proposed TACN contains three parts. The first part is an embedding model, which extracts the embedding representation from the topological structure, vertex attributes, and document-topic distributions. To ensure a consistent training process by back-propagation, we generate document-topic distributions by the neural topic model with Gaussian Softmax constructions. The second part is a prediction model, which is used to exploit labels of vertices. In the third part, an adversarial capsule model is used to help distinguish the latent representations from node structure domain, vertex attribute domain, or document-topic distribution domain. The latent representations, which may come from the three domains, are the output of the embedding model. We incorporate the adversarial idea into the adversarial capsule model to combine the information from these three domains, rather than to distinguish the representations conventionally. Experiments on seven real-world datasets validate the effectiveness of our method.

This paper looks at this intriguing question: are single images with their details lost during deraining, reversible to their artifact-free status? We propose an end-to-end detail-recovery image deraining network (termed a DRDNet) to solve the problem. Unlike existing image deraining approaches that attempt to meet the conflicting goal of simultaneously deraining and preserving details in a unified framework, we propose to view rain removal and detail recovery as two seperate tasks, so that each part could specialize rather than trade-off between two conflicting goals. Specifically, we introduce two parallel sub-networks with a comprehensive loss function which synergize to derain and recover the lost details caused by deraining. For complete rain removal, we present a rain residual network with the squeeze-and-excitation (SE) operation to remove rain streaks from the rainy images. For detail recovery, we construct a specialized detail repair network consisting of welldesigned blocks, named structure detail context aggregation block (SDCAB), to encourage the lost details to return for eliminating image degradations. Moreover, the detail recovery branch of our proposed detail repair framework is detachable and can be incorporated into existing deraining methods to boost their performances. DRD-Net has been validated on several well-known benchmark datasets in terms of deraining robustness and detail accuracy. Comparisons show clear visual and numerical improvements of our method over the state-of-the-arts.

In this paper, we propose a normal estimation method for unstructured point cloud. We observe that geometric estimators commonly focus more on feature preservation but are hard to tune parameters and sensitive to noise, while learning-based approaches pursue an overall normal estimation accuracy but cannot well handle challenging regions such as surface edges. This paper presents a novel normal estimation method, under the co-support of geometric estimator and deep learning. To lowering the learning difficulty, we first propose to compute a suboptimal initial normal at each point by searching for a best fitting patch. Based on the computed normal field, we design a normal-based height map network (NH-Net) to fine-tune the suboptimal normals. Qualitative and quantitative evaluations demonstrate the clear improvements of our results over both traditional methods and learning-based methods, in terms of estimation accuracy and feature recovery.

Natural language processing (NLP) is an effective tool for generating structured information from unstructured data, the one that is commonly found in clinical trial texts. Such interdisciplinary research has gradually grown into a flourishing research field with accumulated scientific outputs available. In this study, bibliographical data collected from Web of Science, PubMed, and Scopus databases from 2001 to 2018 had been investigated with the use of three prominent methods, including performance analysis, science mapping, and, particularly, an automatic text analysis approach named structural topic modeling. Topical trend visualization and test analysis were further employed to quantify the effects of the year of publication on topic proportions. Topical diverse distributions across prolific countries/regions and institutions were also visualized and compared. In addition, scientific collaborations between countries/regions, institutions, and authors were also explored using social network analysis. The findings obtained were essential for facilitating the development of the NLP-enhanced clinical trial texts processing, boosting scientific and technological NLP-enhanced clinical trial research, and facilitating inter-country/region and inter-institution collaborations.

Text categorization can solve the problem of information clutter to a large extent, and it also provides a more efficient search strategy and more effective search results for information retrieval. In recent years, Convolutional Neural Networks have been widely applied to this task. However, most existing CNN models are difficult to extract longer n-grams features for the reason as follow: the parameters of the standard CNN model will increase with the increase of the length of n-grams features because it extracts n-grams features through convolution filters of fixed window size. Meanwhile, the term weighting schemes assigning reasonable weight values to words have exhibited excellent performance in traditional bag-of-words models. Intuitively, considering the weight value of each word in n-grams features may be beneficial in text classification. In this paper, we proposed a model called weighted n-grams CNN model. It is a variant of CNN introducing a weighted n-grams layer. The parameters of the weighted n-grams layer are initialized by term weighting schemes. Only by adding fixed parameters can the model generate any length of weighted n-grams features. We compare our proposed model with other popular and latest CNN models on five datasets in text classification. The experimental results show that our proposed model exhibits comparable or even superior performance.

Collaborative Web services QoS prediction (CQoSP) has been proved to be an effective tool to predict unknown QoS values of services. Recently a number of efforts have been made in this area, focusing on improving the accuracy of prediction. In this paper, we consider a novel kind of CQoSP, shared CQoSP, where multiple parties share their data with each other in order to provide more accurate prediction than a single party could do. To encourage data sharing, we propose a privacy-preserving framework which enables shared collaborative QoS prediction without leaking the private information of the involved party. Our framework is based on differential privacy, a rigorous and provable privacy model. We conduct extensive experiments on a real Web services QoS dataset. Experimental results show the proposed framework increases prediction accuracy while ensuring the privacy of data owners.

Advanced 3D optical and laser scanners bring new challenges to computer graphics. We present a novel nonrigid surface registration algorithm based on Iterative Closest Point (ICP) method with multiple correspondences. Our method, called the Nonrigid Iterative Closest Points (NICPs), can be applied to surfaces of arbitrary topology. It does not impose any restrictions on the deformation, e.g. rigidity or articulation. Finally, it does not require parametrization of input meshes. Our method is based on an objective function that combines distance and regularization terms. Unlike the standard ICP, the distance term is determined based on multiple two-way correspondences rather than single one-way correspondences between surfaces. A Laplacian-based regularization term is proposed to take full advantage of multiple two-way correspondences. This term regularizes the surface movement by enforcing vertices to move coherently with their 1-ring neighbors. The proposed method achieves good performances when no global pose differences or significant amount of bending exists in the models, for example, families of similar shapes, like human femur and vertebrae models.

Emotion distribution learning aims to annotate unlabeled instances with a set of emotion categories and their strengths. Non-negative Matrix Tri-Factorization (NMTF) introduces an association matrix between document clusters and word clusters to help the domain adaptation task in emotion distribution learning. Nevertheless, many prior cross-domain emotion distribution learning methods had two major deficiencies. First, they hypothesize that there is a one-to-one correspondence between document clusters and emotion labels. In their experiments, the number of document clusters depends on the number of labels. Second, the prior work does not endow models with adequate constraints. In the real scenario of cross-domain emotion distribution learning, there are potential constraints that may improve the performance of such models. In order to address these problems, we propose a constrained optimization approach based on NMTF for cross-domain emotion distribution learning. In our model, the relationship between document clusters and emotion labels is not always one-to-one. A novel content-based constraint is also endowed based on the hypothesis that documents belonging to the same clusters must have similar content. We solve the optimization problem by an alternately iterative algorithm and show the proof of convergence. Experiments on 12 real-world cross-domain emotion distribution learning tasks validate the effectiveness of our method.

The aim of automatic text summarization (ATS) is to extract representative texts from documents and keep major points of the extracted texts consistent with the original documents. However, most existing studies ignore sentimental information loss in the summarization process, which leads to sentiment loss summarization. To address the sentiment loss issue during summarization, we introduce a sentiment compensation mechanism into document summarization and propose a graph-based extractive summarization approach named Sentiment Lossless Summarization (SLS). SLS first creates a graph representation for a document to obtain the importance score (i.e., literal indicator) of each sentence. Second, sentiment dictionaries are leveraged to analyze the sentence sentiments. Third, during each summarization iteration, the sentences with the lowest scores are iteratively removed, and the sentiment compensation weights of the remaining sentences are updated. With the help of sentiment compensation during the summarization process, sentiment consistencies between candidate summaries and the original documents are maintained. Intrinsic evaluations conducted on the DUC2001, DUC2002, DUC2004, and Multi-News datasets demonstrate that our approach outperforms baselines and state-of-the-art summarization methods in terms of Recall-Oriented Understudy for Gisting Evaluation (ROUGE) scores. Additionally, to further evaluate SLS performance in sentiment retention, extrinsic evaluations are introduced, and summary quality in terms of sentiment loss is evaluated by measuring the prediction accuracy for sentiment polarities of either movie (IMDb dataset) or product (Amazon dataset) review summaries. The experimental results demonstrate that our approach can improve prediction accuracy by at most 6% compared to the baseline.

Architecture selection is a fundamental problem in artificial neural networks, which could be treated as a decision making process that evaluates, ranks, and makes choices from a set of network structures. Traditional methods evaluate a network structure by designing a criterion based on a validation model or an error bound model. On one hand, the time complexity of a validation model is usually high; on the other hand, different validation models or error bound models may lead to different (even conflicting) results, which post challenges to the traditional single criterion-based architecture selection methods. In the area of decision making, many problems employed multiple criteria since the performance is better than using a single criterion. In this paper, we propose a multi-criteria decision making based architecture selection algorithm for single-hidden layer feedforward neural networks trained by extreme learning machine. Two criteria are incorporated into the selection process, i.e., training accuracy and the Q-value estimated by the localized generalization error model. The training accuracy reflects the capability of the model on correctly categorizing the known samples, and the Q-value estimated by localized generalization error model reflects the size of the neighbourhood of training samples in which the model can predict unseen samples with confidence. By achieving a trade-off between these two criteria, a new architecture selection algorithm is proposed. Experimental comparisons demonstrate the feasibility and effectiveness of the proposed method.

Social emotion classification aims to predict the aggregation of emotional responses embedded in online comments contributed by various users. Such a task is inherently challenging because extracting relevant semantics from free texts is a classical research problem. Moreover, online comments are typically characterized by a sparse feature space, which makes the corresponding emotion classification task very difficult. On the other hand, though deep neural networks have been shown to be effective for speech recognition and image analysis tasks because of their capabilities of transforming sparse low-level features to dense high-level features, their effectiveness on emotion classification requires further investigation. The main contribution of our work reported in this paper is the development of a novel model of semantically rich hybrid neural network (HNN) which leverages unsupervised teaching models to incorporate semantic domain knowledge into the neural network to bootstrap its inference power and interpretability. To our best knowledge, this is the first successful work of incorporating semantics into neural networks to enhance social emotion classification and network interpretability. Through empirical studies based on three real-world social media datasets, our experimental results confirm that the proposed hybrid neural networks outperform other state-of-the-art emotion classification methods.

With the explosion of knowledge and information in the big data era, learning new things efficiently is of crucial significance. Despite recent development of e-learning techniques which have broken the temporal and spatial barriers for learners, it is still very difficult to meet the requirement of efficient learning, as the key issues involve not only searching for learning resources but also identification of learning paths. People from diverse backgrounds, in most cases, also need to work as a group to acquire new knowledge or skills and complete certain tasks. As these tasks are normally assigned with time constraints, employment of e-learning systems may be the optimal approach. In this research, we study the issue of identifying a suitable learning path for a group of learners rather than a single learner in an e-learning environment. Particularly, a profile-based framework for the discovery of group learning paths is proposed by taking various learning-related factors into consideration. We also conduct experiments on real learners to validate the effectiveness of the proposed approach.

First Story Detection (FSD) aims to identify the first story for an emerging event previously unreported, which is essential to practical applications in news analysis, intelligence gathering, and national security. Compared to information retrieval, text clustering, text classification, and other subject-based tasks, FSD is event-based and thus faces the challenging issues of multiple events on the same subject and the evolution of events. To tackle these challenges, several schemes for exploiting temporal information, named entity, and topic modeling, have been proposed for FSD. In this paper, we present a new term weighting scheme called LGT, which jointly models the Local element, Global element, and Topical association of each story. An unsupervised algorithm based on LGT is then devised and applied to FSD. We evaluate 4 feature reduction strategies and test our LGT scheme on an online model. Experiments show that our approach yields better results than existing baseline schemes on both retrospective and online FSD.

Locating network fault problems and filtering trivial alarms from important ones are the two main challenges in Network Operation Centers (NOCs). In this paper, we present an alarm behavior analysis and discovery system, AABD, that establishes flapping and parent–child (P–C) rules to reveal the operation patterns from a large number of alarms in telecom networks. These rules can be exploited to filter out unimportant alarms, conduct multi-dimensional analysis of the alarms and identify potential network problems. We propose two novel and effective algorithms to establish the flapping rules and P-C rules. The proposed system is validated using alarm datasets from five Internet service providers. Specifically, we verify the system and methodology in each of the five network domains, i.e., circuit-switched network (CS), packet-switched network (PS), 2G-radio access network (RAN-2G), 3G-radio access network (RAN-3G) and 4G-radio access network (RAN-4G), as these five domains can, to a great extent, form a complete network environment. More importantly, our system can establish a small number of rules, only dozens of flapping rules and P-C rules, and compress the alarms by approximately 84%, i.e., 84% of alarms will not be sent to the network operator. To summarize, the proposed system can help network operators respond to network faults in a timely fashion, locate the faults accurately and significantly reduce the time spent on these tasks.

Contrary to traditional Web information retrieval methods that can only return a ranked list of Web pages and only allow search terms in the query, we have developed a novel learning framework for retrieving precise information blocks from Web pages given a query, which may contain some search terms and prior information such as the layout format of the data. There are two challenging sub-tasks for this problem. One challenge is information block detection, where a Web page is automatically segmented into blocks. Another challenge is to find the information blocks relevant to the query. Existing page segmentation methods, which make use of only visual layout information or only content information, do not consider the query information, leading to a solution having conflict with the information need expressed by the query. Our framework aims at modeling the query and the block features to capture both keyword information and prior information via a probabilistic graphical model. Fisher Kernel, which can effectively incorporate the graphical model, is then employed to accomplish the two sub-tasks in a unified manner, optimizing the final goal of block retrieval performance. We have conducted experiments on benchmark datasets and read-world data. Comparisons between existing methods have been conducted to evaluate the effectiveness of our framework.

The world has encountered and witnessed the great popularity of various emerging e-learning resources such as massive open online courses (MOOCs), textbooks and videos with the development of the big data era. It is critical to understand the characteristics of users to assist them to find desired and relevant learning resources in such a large volume of resources. For example, understanding the pre-knowledge on vocabulary of learners is very prominent and useful for language learning systems. The language learning effectiveness can be significantly improved if the pre-knowledge levels of learners on vocabulary can be accurately predicted. In this research, the authors model the vocabulary of learners by extracting their history of learning documents and identify the suitable vocabulary knowledge scales (VKS) for pre-knowledge prediction. The experimental results on real participants verify that the optimal VKS and the proposed predicting model are powerful and effective.

With the rapid proliferation of Web 2.0, the identification of emotions embedded in user-contributed comments at the social web is both valuable and essential. By exploiting large volumes of sentimental text, we can extract user preferences to enhance sales, develop marketing strategies, and optimize supply chain for electronic commerce. Pieces of information in the social web are usually short, such as tweets, questions, instant messages, messages, and news headlines. Short text differs from normal text because of its sparse word co-occurrence patterns, which hampers efforts to apply social emotion classification models. Most existing methods focus on either exploiting the social emotions of individual words or the association of social emotions with latent topics learned from normal documents. In this paper, we propose a topic-level maximum entropy (TME) model for social emotion classification over short text. TME generates topic-level features by modeling latent topics, multiple emotion labels, and valence scored by numerous readers jointly. The overfitting problem in the maximum entropy principle is also alleviated by mapping the features to the concept space. An experiment on real-world short documents validates the effectiveness of TME on social emotion classification over sparse words.

With the development of social network platforms, discussion forums, and question answering websites, a huge number of short messages that typically contain a few words for an individual document are posted by online users. In these short messages, emotions are frequently embedded for communicating opinions, expressing friendship, and promoting influence. It is quite valuable to detect emotions from short messages, but the corresponding task suffers from the sparsity of feature space. In this article, we first generate term groups co-occurring in the same context to enrich the number of features. Then, two basic supervised topic models are proposed to associate emotions with topics accurately. To reduce the time cost of parameter estimation, we further propose an accelerated algorithm for our basic models. Extensive evaluations using three short corpora validate the efficiency and effectiveness of the accelerated models for predicting the emotions of unlabeled documents, in addition to generate the topic-level emotion lexicons.

In this chapter, we focus on the detection of sentiment strength values for a given document. A convolution-based model is proposed to encode semantic and syntactic information as feature vectors, which has the following two characteristics: (1) it incorporates shape and morphological knowledge when generating semantic representations of documents; (2) it divides words according to their part-of-speech (POS) tags and learns POS-level representations for a document by convolving grouped word vectors. Experiments using six human-coded datasets indicate that our model can achieve comparable accuracy with that of previous classification systems and outperform baseline methods over correlation metrics.

To detect rumors automatically in social media, methods based on recurrent neural network and convolutional neural network have been proposed. These methods split a stream of posts related to an event into several groups along time, and represent each group using unsupervised methods such as paragraph vector. However, many posts in a group (e.g., retweeted posts) do not contribute much to rumor detection, which deteriorates the performance of rumor detection based on unsupervised group embedding. In this paper, we propose a Supervised Group Embedding based Rumor Detection (SGERD) model that considers both textual and temporal information. Particularly, SGERD exploits post-level textual information to generate group embeddings, and is able to identify salient posts for further analysis. Experimental results on two real-world datasets demonstrate the effectiveness of our proposed model.

Artificial Intelligence (AI) plays an increasingly important role in advancing human brain research, given the continually growing number of academic research articles in the last decade. Meanwhile, human brain research can provide opportunities for the development of innovative AI techniques. Exploring and tracking patterns of the scientific articles of human brain research using AI can provide a comprehensive overview of the interdisciplinary field. Thus, this paper presents a bibliometric analysis to identify research status and development trend of the field between 2009 and 2018. Specifically, we analyze annual distributions of articles and their citations, identify prolific journals and affiliations, and visualize characteristics of scientific collaboration. Furthermore, research topics are analyzed and revealed. The obtained findings benefit scholars in the field, to understand the current status of research as well as monitoring scientific and technological activities.

Learning Analytics (LA) have been widely investigated and applied to understand and optimise the learning process and environment. Among a number of LA tools, Bayesian Knowledge Tracing (BKT) was d...