Search

Sentiment and emotion, which correspond to long-term and short-lived human feelings, are closely linked to each other, leading to the fact that sentiment analysis and emotion recognition are also two interdependent tasks in natural language processing (NLP). One task often leverages the shared knowledge from another task and performs better when solved in a joint learning paradigm. The conversational context dependency, multi-modal interaction and multi-task correlation are three key factors that contribute to such joint paradigm. However, none of the recent approaches have considered them in a unified framework. To fill this gap, we propose a multi-modal multi-task interactive graph attention network, termed M3GAT, to simultaneously solve the three problems. At the heart of the model is a proposed interactive conversation graph layer containing three core sub-modules, which are: (1) local-global context connection for modeling both local and global conversational context; (2) cross-modal connection for learning multi-modal complementary and (3) cross-task connection for capturing the correlation across two tasks. Comprehensive experiments on three benchmarking datasets, MELD, MEISD and MSED, show the effectiveness of M3GAT over state-of-the-art baselines with the margin of 1.88%, 5.37% and 0.19% for sentiment analysis, 1.99%, 3.65% and 0.13% for emotion recognition respectively. In addition, we also show the superiority of multi-task learning over single-task framework.

Web search has become a popular way for people to obtain information. Due to the complexity of search task, one retrieval cannot serve all of user’s information needs. Multiple interactions with the search engine are required, and session search comes into being. Currently proposed session search methods still cannot take advantage of interactive information to capture the user’s implicit information needs. Recently, quantum theory has been successively applied into information retrieval task. We find the similarities between quantum measurements and session interactions. This paper thus develops a Quantum Interactive Retrieval Model (QIRM), which involves both quantum standard measurement (QSM) and quantum weak measurement (QWM) to re-characterize the user’s cognition shifts during a session interaction. Particularly, the user’s cognition states in strong interaction and weak interaction are quantified into quantum-like representations formalized by QSM and QWM, respectively. The representations are then viewed as the user’s information needs for computing ranking score of an evaluated document. We conduct experiments on TREC 2013 and 2014 Session Track datasets. The empirical evaluation demonstrates the effectiveness of our proposed methods, which obtains very comparable performance in terms of nDCG@10 and ERR@10.

In order to support smart construction, digital twin has been a well-recognized concept for virtually representing the physical facility. It is equally important to recognize human actions and the movement of construction equipment in virtual construction scenes. Compared to the extensive research on human action recognition (HAR) that can be applied to identify construction workers, research in the field of construction equipment action recognition (CEAR) is very limited, mainly due to the lack of available datasets with videos showing the actions of construction equipment. The contributions of this research are as follows: (1) the development of a comprehensive video dataset of 2,064 clips with five action types for excavators and dump trucks; (2) a new deep learning-based CEAR approach (known as a simplified temporal convolutional network or STCN) that combines a convolutional neural network (CNN) with long short-term memory (LSTM, an artificial recurrent neural network), where CNN is used to extract image features and LSTM is used to extract temporal features from video frame sequences; and (3) the comparison between this proposed new approach and a similar CEAR method and two of the best-performing HAR approaches, namely, three-dimensional (3D) convolutional networks (ConvNets) and two-stream ConvNets, to evaluate the performance of STCN and investigate the possibility of directly transferring HAR approaches to the field of CEAR.

Learning a classifier from positive and unlabeled data may occur in various applications. It differs from the standard classification problems by the absence of labeled negative examples in the training set. So far, two main strategies have typically been used for this issue: the likely negative examplesbased strategy and the class prior-based strategy, in which the likely negative examples or the class prior is required to be obtained in a preprocessing step. In this paper, a new strategy based on the Bhattacharyya coefficient is put forward, which formalizes this learning problem as an optimization problem and does not need a preprocessing step. We first show that with the given positive class conditional probability density function (PDF) and the mixture PDF of both the positive class and the negative class, the class prior can be estimated by minimizing the Bhattacharyya coefficient of the positive class with respect to the negative class. We then show how to use this result in an implicit mixture model of restricted Boltzmann machines to estimate the positive class conditional PDF and the negative class conditional PDF directly to obtain a classifier without the explicit estimation of the class prior. Many experiments on real and synthetic datasets illustrated the superiority of the proposed approach.

Social influence is essential to social recommendation. Current influence-based social recommendation focuses on the explicit influence on observed social links. However, in real cases, implicit social influence can also impact users' preference in an unobserved way. In this work, we concern two kinds of implicit influence: Local Implicit Influence of persons on unobserved interpersonal relations, and Global Implicit Influence of items broadcasted to users. We improve the state-of-the-art GNN-based social recommendation methods by modeling two kinds of implicit influences separately. Local implicit influence is involved by predicting unobserved social relationships. Global implicit influence is involved by defining global popularity of each item and personalize the impact of the popularity on each user. In a GCN network, explicit and implicit influence are integrated to learn the social embedding of users and items in social recommendation. Experimental results on Yelp initially demonstrate the effectiveness of proposed model.

Real-valued word vectors (RWV) often perform poorly when showing semantic relations such as the ambiguity of the word itself (“spring” can refer to a kind of season or fountain) and the polarity between words (words “love” and “hate” have a high degree of similarity due to their similar distribution). Although dynamic word vectors alleviate these problems, such methods are still lacking interpretability. Inspired by quantum theory, we believe that basic reasons are insufficient modeling ability of RWV and weak abstract ability of language models. To solve the above problems, we propose to use complex-valued word vectors (CWV) for the representation, which can naturally model semantic relations of words. We also design an explicable lexicon-extended quantum probability neural network (LQPNN) based on the theory of quantum measurement to training CWV. Experiments on three datasets of classification demonstrate the effectiveness of our model, which achieves superior performances.

Sentiment analysis aims to judge the sentiment polarity of various types of text at the document and sentence level. It has important theoretical and practical significance and is a hot topic in natural language processing. Although existing sentiment analysis methods based on sentiment dictionaries and machine learning consider contextual semantic information, they still have the problem of not being able to effectively encode the mixing of semantic subspaces, and therefore ignore the feature interaction between sentiment words. To solve this problem, this article combines quantum mechanics, deep learning and natural language processing technology, and introduces the concept of density matrix into the convolutional neural network. The density matrix can encode more semantic dependencies and can be integrated into the neural network architecture. We proposed a Quantum-inspired Model based on Convolutional Neural Network for Sentiment Analysis (QI-CNN). Experiments on IMDB English dataset and Weibo Chinese dataset verify the effectiveness of our proposed model.

Recently, the mathematical framework of quantum mechanics has boosted interpretability and performance in a variety of NLP tasks. Since different linguistic units are explicitly weighted by this framework in density matrix, it is also often believed that the models inspired by quantum mechanics afford transparency. However, the relationship between weights of linguistic units and model outputs is uncertain. In this work we test whether this assumption holds by performing extensive experiments in question answering (QA) model to manipulating semantic weights and assessing the resulting differences in its output. Although we observe in some ways that higher weights do correlate with greater effect on model, but most of the data does not support it. i.e., the weight is independent of the prediction, and different weight distributions can produce equivalent result. Our results illustrate that standard quantum-inspired model has interpretable structure but does not accurately model feature importance and requires improvement in semantic representation and model interpretability.

Conversational Sentiment analysis (CSA) is an import task in the text sentiment analysis filed, which aims to predict the sentiment sequence corresponding to the sentence sequence in the conversation. Recently, many researchers have proposed deep learning based models to learn both speaker-context and sentence-context information interactions. However, the existing models generally learn the sentiment-context interactions by simple non-linear function, which does not focus on affective reasoning and is lack of interpretability. To address those issues, we introduce probabilistic graph model (PGM). Inspired by the interaction matrix of the Coupled HMM, we propose a novel interaction matrix based improvement for output gating mechanism, which contains local 2-order sentiment-context interaction. Then we propose the interaction matrix Based Long Short-Term Memory Network (ConvLSTM), which captures local sentiment-context interactions. Next, inspired by the reasoning of PGM, we develop a novel PGM layer to model global high-order sentiment-context interaction for sentiment label sequence. Finally, we propose PGM Based Hierarchical Affective Reasoning Network (PHARN), which can learn both local and global sentiment-context interaction. Experiments on MELD and IEMOCAP datasets demonstrate the effectiveness of our model, which achieves excellent results on multiple benchmarks.

Deep learning model is popular and widely used. Its generalization performance is superior to traditional machine learning methods. Traditional statistical learning theory cannot explain its success. Most of network architectures are designed artificially and empirically. There are few methods to design new structures and mathematical tools to evaluate feature representation capabilities of networks. To address this issue, based on information theory, we propose an interpretable regularization method named Minimize Mutual Information Method (MMIM), which reduce the generalization error by minimizing the mutual information between hidden layer neurons. We derive a tighter generalization error boundary and propose a two-step method to reduce the neural network generalization error boundary. The experimental results verify the effectiveness of our proposed MMIM on Fashion-MNIST dataset and CIFAR datasets. Inspired by ensemble learning, we discuss why our method is effective and suggest how to design new networks.

Multi-head attention is applied in Transformer to jointly attend to information from different representation sub-spaces at different positions. In this work, we are inspired by ensemble learning propose a new disagreement term regularization to ensure that the multi-head mechanism works effectively. Specifically, we treat each head as a base learner of emsemble model and make multi-head more diverse by the bias-variance-covariance decomposition. The test results show that our method is quite effective in Multi-30K Enlish⟹German and IWSLT16 Enlish⟹German translation tasks.

User preference modeling is an essential task for online recommender systems. Recently, methods have been applied to model short-term user preferences within a short-term period. These approaches use recent user behavior as the context to determine the current short-term preferences. However, we argue that short-term user preferences are related to more complex contexts, e.g., the seasons or the time of the day. Furthermore, we make the hypothesis that short-term preferences of a user is actually a joint effect of his/her stable long-term preferences and the context-aware impact. Therefore, we propose LoCo-VAE, a unified model of this joint effect with Variational Auto-Encoder (VAE) based strategies. First, we utilize a Multilayer Perceptron(MLP) to capture long-term user preferences. Second, we improve the traditional VAE by distributing user interactions with respect to different contexts to introduce the context-aware impact. Finally, the long-term preferences and context-aware impact are combined with a joint generative training process to generate the embedding of short-term user preferences. Experiments on real-world datasets of Amazon consumption and music selection demonstrate the superiority of our model compare with state-of-the-art methods in recommendation system.

In order to improve the efficiency of construction and reduce or avoid accidents, it is necessary to accurately identify the activities of construction equpment. However, research on construction equipment action recognition (CEAR) is still very limited, mainly due to the lack of video datasets for algorithm development. This research developed a video dataset of 2064 clips with five action categories of excavators and dump trucks. Using this dataset, a simplified temporal convolutional network (STCN) was developed by combining convolutional neural networks (CNN) with long-short-term memory networks (LSTM), where CNN is used to extract image features from video clips and LSTM is used to extract time-series features from the frame sequences of videos. By comparing with a similar CEAR method (CNN-DLSTM) and two advanced human action recognition methods (C3D and two-stream ConvNets), the STCN method has comparable performance and requires less training time. This study also preliminarily proved that the human action recognition (HAR) methods have the potential for action recongnition of construction equipment.

Safety accidents frequently occur in the construction industry, which leads to heavy work in the reports and investigations of safety accidents. Thus, under the background of accelerating the establishment of informationization of construction safety, it is of particular importance to deeply study the informationization of the reports on construction safety accidents and investigations. This paper explored a C-BiLSTM-based approach that combines with deep learning approaches to automatically classify the reports on construction safety accidents and is capable of extracting semantic features from the construction accident reports and automatically classifies them into pre-defined category. The results showed that the proposed C-BiLSTM-based classification performed better in the classification of construction accident reports with less manual intervention and complex engineering. In addition, the reasons and limitations of classification errors were discussed, and suggestions were proposed for improvement. The results of this study can help building safety managers to develop risk management systems.

Click-Through Rate (CTR) prediction is an important task in the online advertisement recommended filed, which aims to predict the probabilities that an ad is clicked by a user in the display of the ads. Recently, many researchers have proposed deep learning based models to learn both low-order and high-order feature interactions. However, the existing deep learning based CTR models generally learn the feature interactions by inner product, which is simple and lack of interpretability. To address this issue, we introduce a concept of quantum probability theory: density matrix. Inspired by the feature of the density matrix, we propose a novel density matrix based representation for input instances, which can contain global 2-order feature interaction information. Then we combine the advantages of both density matrix and Convolutional Neural Network and propose the Density Matrix Based Convolutional Neural Network (DMCNN), which can capture more feature interactions than other models. Experiments on Criteo and Avazu datasets demonstrate the effectiveness of our model, which achieves excellent results on multiple benchmarks.

It has long been recognized as a difficult problem to determine whether the observed statistical correlations between two classical variables arise from causality or from common causes. Recent research has shown that, in a quantum theoretical framework, the mechanisms of entanglement and quantum coherence provide advantages in tackling this problem. In some particular cases, quantum common causes and quantum causality can be effectively distinguished by using observations alone. However, these solutions do not apply to all cases. There still exist a large class of cases in which quantum common causes and quantum causality cannot be distinguished. In this paper, along the line of considering unitary transformation as causality in the quantum world, we formally show that quantum common causes and quantum causality are universally separable. Based on the analysis, we further provide a general method to discriminate the two.

In information retrieval (IR), the improvement of the effectiveness often sacrifices the stability of an IR system. To evaluate the stability, many risk-sensitive metrics have been proposed. Since the theoretical limitations, the current works study the effectiveness and stability separately, and have not explored the effectiveness–stability tradeoff. In this paper, we propose a Bias–Variance Tradeoff Evaluation (BV-Test) framework, based on the bias–variance decomposition of the mean squared error, to measure the overall performance (considering both effectiveness and stability) and the tradeoff between effectiveness and stability of a system. In this framework, we define generalized bias–variance metrics, based on the Cranfield-style experiment set-up where the document collection is fixed (across topics) or the set-up where document collection is a sample (per-topic). Compared with risk-sensitive evaluation methods, our work not only measures the effectiveness–stability tradeoff of a system, but also effectively tracks the source of system instability. Experiments on TREC Ad-hoc track (1993–1999) and Web track (2010–2014) show a clear effectiveness–stability tradeoff across topics and per-topic, and topic grouping and max–min normalization can effectively reduce the bias–variance tradeoff. Experimental results on TREC Session track (2010–2012) also show that the query reformulation and increase of user data are beneficial to both effectiveness and stability simultaneously.

In deep learning models, most of network architectures are designed artificially and empirically. Although adding new structures such as convolution kernels in CNN is widely used, there are few methods to design new structures and mathematical tools to evaluate feature representation capabilities of new structures. Inspired by ensemble learning, we propose an interpretable regularization method named Minimize Mutual Information Method(MMIM), which minimize the generalization error by minimizing the mutual information of hidden neurons. The experimental results also verify the effectiveness of our proposed MMIM.

Recently, a Quantum Probability Drive Network (QPDN) is proposed to model different levels of semantic units by extending word embedding to complex-valued representation (CR). The extended complex-valued embeddings are still insensitive to polarity causing that they generalize badly in sentiment analysis (SA). To solve it, we propose a method of encoding sentiment information into sentiment words for SA. Attention mechanism and an auxiliary task are introduced to help learn the CR of sentiment words with the help of the sentiment lexicon. We use the amplitude part to represent the distributional information and the phase part to represent the sentiment information of the language. Experiments on three popular SA datasets show that our method is effective.

As one of the most popular social media platforms today, Twitter provides people with an effective way to communicate and interact with each other. Through these interactions, influence among users gradually emerges and changes people's opinions. Although previous work has studied interpersonal influence as the probability of activating others during information diffusion, they ignore an important fact that information diffusion is the result of influence, while dynamic interactions among users produce influence. In this article, the authors propose a novel temporal influence model to learn users' opinion behaviors regarding a specific topic by exploring how influence emerges during communications. The experiments show that their model performs better than other influence models with different influence assumptions when predicting users' future opinions, especially for the users with high opinion diversity.

In many research and application areas, such as information retrieval and machine learning, we often encounter dealing with a probability distribution that is mixed by one distribution that is relevant to our task in hand and the other that is irrelevant and that we want to get rid of. Thus, it is an essential problem to separate the irrelevant distribution from the mixture distribution. This article is focused on the application in Information Retrieval, where relevance feedback is a widely used technique to build a refined query model based on a set of feedback documents. However, in practice, the relevance feedback set, even provided by users explicitly or implicitly, is often a mixture of relevant and irrelevant documents. Consequently, the resultant query model (typically a term distribution) is often a mixture rather than a true relevance term distribution, leading to a negative impact on the retrieval performance. To tackle this problem, we recently proposed a Distribution Separation Method (DSM), which aims to approximate the true relevance distribution by separating a seed irrelevance distribution from the mixture one. While it achieved a promising performance in an empirical evaluation with simulated explicit irrelevance feedback data, it has not been deployed in the scenario where one should automatically obtain the irrelevance feedback data. In this article, we propose a substantial extension of the basic DSM from two perspectives: developing a further regularization framework and deploying DSM in the automatic irrelevance feedback scenario. Specifically, in order to avoid the output distribution of DSM drifting away from the true relevance distribution when the quality of seed irrelevant distribution (as the input to DSM) is not guaranteed, we propose a DSM regularization framework to constrain the estimation for the relevance distribution. This regularization framework includes three algorithms, each corresponding to a regularization strategy incorporated in the objective function of DSM. In addition, we exploit DSM in automatic (i.e., pseudo) irrelevance feedback, by automatically detecting the seed irrelevant documents via three different document reranking methods. We have carried out extensive experiments based on various TREC datasets, in order to systematically evaluate the proposed methods. The experimental results demonstrate the effectiveness of our proposed approaches in comparison with various strong baselines.

Despite their widespread applications, deep neural networks often tend to overfit the training data. Here, we propose a measure called VECA (Variance of Eigenvalues of Covariance matrix of Activation matrix) and demonstrate that VECA is a good predictor of networks' generalization performance during the training process. Experiments performed on fully-connected networks and convolutional neural networks trained on benchmark image datasets show a strong correlation between test loss and VECA, which suggest that we can calculate the VECA to estimate generalization performance without sacrificing training data to be used as a validation set.

Deep Neural Networks (DNNs) with a large number of parameters are very powerful machine learning systems. However, overfitting is a serious problem in such networks. Till now, many regularizers such as dropout, data augmentation have been proposed to prevent overfitting. Motivated by ensemble learning, we treat each hidden layer in neural networks as an ensemble of some base learners by dividing hidden units into some non-overlapping groups and each group is considered as a base learner. Based on the theoretical analysis of generalization error of ensemble estimators (bias-variance-covariance decomposition), we find the variance of each base learner plays an important role in preventing overfitting and propose a novel regularizer---\emphEnsemble-based Low-Level Sample-Variances Method (ELSM) to encourage each base learner of hidden layers to have a low-level sample-variance. Experiments across a number of datasets and network architectures show that ELSM can effectively reduce overfitting and improve the generalization ability of DNNs.

Recently, recurrent neural networks (RNNs) based methods have achieved profitable performance on mining temporal characteristics in user behavior. However, user preferences are changing over time and have not been fully exploited in e-commerce scenarios. To fill in the gap, we propose an approach, called quantum inspired preference interactive networks (QPIN), which leverages the mathematical formalism of quantum theory (QT) and the long short term memory (LSTM) network, to interactively learn user preferences. Specifically, the tensor product operation is used to model the interaction among a single user's own preferences, i.e. individual preferences. A quantum many-body wave function (QMWF) is employed to model interaction among all users' preferences, i.e. group preferences. Further, we bridge them by deriving a rigorous projection, and thus take the interplay between them into account. Experiments on an Amazon dataset as well as a real-world e-commerce dataset demonstrate the effectiveness of QPIN, which achieves superior performances compared with the state-of-the-art methods in terms of AUC and F1-score.

Conversational sentiment analysis (CSA) is emergent research field in natural language processing (NLP). This brings a lot of new issues worth studying and directions worth exploring. At the same time, there are many difficulties that need to be overcome. There are lot of challenges, such as lacking of effective deep learning models and being short of appropriate datasets. Inspired by the concept of density matrix in quantum mechanics, we propose a novel attention mechanism called DMATT and apply it to conversational sentiment analysis tasks. In the experiment, we find that deep learning model combined with DMATT has a great improvement in test results compared to the model with traditional attention mechanism. Recurrent neural networks (RNN) and their variants LSTM and GRU are very effective choices in solving time series problem such as conversational sentiment analysis tasks. In this paper, we propose a new model combining GRU and DMATT called DMATT-BiGRU. We experiment in multiple datasets, one of which is called ScenarioSA collected by ourselves.

The establishment of social links is not only determined by personal interests but also by neighbors’ influences, which may vary across different neighbors. However, the independent influence of each neighbor has not been separately considered on semantic level in current approaches. In this work, we predict missing social links by modeling semantic influence of each neighbor separately with an embedding approach. The semantic of influence is fine grained on each neighbor’s specific interest with attention-based method. The proposed model named AIMN (Attention-based semantic Influence of Multiple Neighbors) is integrated with structure information with a uniform framework. Extensive experiments on different real-world networks demonstrate that AIMN outperforms state-of-the-art methods.

The growing amount of fault smart meter raises concern from the State Grid Corporation of China in Tianjin. However, data analysis on smart meter fault is still not enough. This paper investigates the fault causes among fault meters data and tries to separate the effective meter fault causes out. To this end, several steps have been followed: first, it builds a fault causes analysis system to find out the effective causes. Second, it raises a cause separation method to extract effective reasons from massive causes. Then a cyclic iterative algorithm has been proposed by maximum likelihood solution for solving meter fault causes equations built in cause separation method. The cause separation method behaves better in practical applications than in the traditional statistics approach. The methods are applied to and evaluated on the practical data.

In classic cases, Reichenbach's principle implies that discriminating between common causes and causality is unprincipled since the discriminative results essentially depend on the selection of possible conditional variables. For some typical quantum cases, K.Reid $et$ $al$. \href{https://www.nature.com/articles/nphys3266}{[Nat. Phys. 11, 414 (2015)]} presented the statistic $C$ which can effectively discriminate quantum common causes and quantum causality over two quantum random variables (i.e., qubits) and which only uses measurement information about these two variables. In this paper, we formalize general quantum common causes and general quantum causality. Based on the formal representation, we further investigate their decidability via the statistic $C$ in general quantum cases. We demonstrate that (i) $C \in \left[ { - 1,\frac{1}{{27}}} \right]$ if two qubits are influenced by quantum common causes; (ii) $C \in \left[ { - \frac{1}{{27}},1} \right]$ if the relation between two qubits is quantum causality; (iii) a geometric picture can illuminate the geometric interpretation of the probabilistic mixture of quantum common causes and quantum causality. This geometric picture also provides a basic heuristic to develop more complete methods for discriminating the cases corresponding to $C \in \left[ { - \frac{1}{{27}},\frac{1}{{27}}} \right]$. Our results demonstrate that quantum common causes and quantum causality can be discriminated in a considerable scope., Comment: 8 pages,3 figures,this paper has been accepted for publication as a regular article in Physical Review A

Sentence classification has always been a crucial research topic in Natural Language Processing (NLP). Classical bag-of-words based models have a major limitation: the contextual information between words, which is the key to form meaningful semantic entities, is missing. Moreover, the semantic entities are not necessarily limited to syntactically valid phrases or named entities, but can be high-order association (also referred to as high-order dependence) patterns. To address this issue, in this paper, we propose PPD-CNN, a convolutional neural network (CNN) architecture with Pair-wise Pure Dependence (PPD) for sentence classification. Compared with the traditional CNN, our PPD-CNN (1) combines PPD pattern which is a couple of dependence words as strong un-separable high-level semantic entity and (2) extracts multi-granular semantic information, which treats PPD pattern as an input channel to capture the whole features and the original sentence as another input channel through variable-size convolution filters to catch all kinds of local features. With this design, our PPD-CNN can model the contextual information, which is important for grasping the word sense. The experimental results show that our approach significantly outperforms a wide range of baselines and state-of-the-art methods.

Quantum Theory (QT) has been applied in a number of fields outside physics, e.g. Information Retrieval (IR). A series of pioneering works have verified the necessity to employ QT in IR user models. In this paper, we explore the process of relevance judgement from a novel perspective of the two state vector quantum weak measurement (WM) by considering context information in time domain. Experiments are carried out to verify our arguments.

Ontology, as a representation of shared conceptualization for variety of specific domains, is the core of the semantic web. Concept hierarchy is one of the most popular backbones of ontology which organizes the concepts according to hyponymy relationships, and stores massive entities as the instances of the concepts. An open concept hierarchy, e.g., Wikipedia, always needs to be constantly updated by adding new entities and concepts. In this paper, we propose an automatic solution for ontology update by inserting new entities and generating new concepts for concept hierarchy. The method only requires very limited information of new entity, i.e., the attributes of each entity. The solution is based on a hybrid strategy synthesizing the benefits from the structure of the concept tree and the content of the attributes. The content of the attributes is used to measure the similarity between an entity and a concept. The structure of the concept tree is used to determine which concepts need to be measured. During similarity measurement, the solution also synthesizes the statistical and rule-based factors. The effectiveness of the proposed method is verified by the experiments extending the Chinese and English Wikipedia concept hierarchy with new entities and concepts.

Aspect-level sentiment analysis is an essential subtask of sentiment classification. It aims at classifying the sentiment polarity of given aspect in its context. Recently, a variety of deep learning models have been proposed to solve this task, such as Long Short-Term Memory Networks (LSTM), Convolutional Neural Networks (CNN). In particular, great improvement has been achieved by using attention mechanism. At the same time, the adoption of linguistic resources to improve sentiment classification has also drawn researchers’ attention, and achieved state-of-the-art performance on traditional sentiment classification. Hence in this paper, we explore to combine linguistically constraints with attention mechanism to achieve comparable performance on aspect-level sentiment analysis. Experimental results on SemEval 2014 Datasets showed that the proposed model achieves good performance and verifies the effectiveness of linguistic resources on this task. To our knowledge, there are no work combining the attention mechanism and linguistic resources on this task before. This work gives inspirations to further research.

Attention-based neural network models recently proposed have achieved great success in question answering task. They focus on introducing the interaction information in sentence modeling rather than representing the question and the answer individually. However, there are some limitations of the previous work. First, in the interaction layer, most attention mechanisms do not make full use of the diverse semantic information of the question. Second, they have limited capability to construct interaction from multiple aspects. In this paper, to address these two limitations, we propose a two-step multi-factor attention neural network model. The two-step strategy encodes the question into different representations according to separate words in the answer, and these representations are employed to build dynamic-question-aware attention. Additionally, a multi-factor mechanism is introduced to extract various interaction information, which aims at aggregating meaningful facts distributed in different matching results. The experimental results on three traditional QA datasets show that our model outperforms various state-of-the-art systems.

Though current researches of online collaboration often study the social relationship from an objective view, individuals’ subjective attitudes on their interrelationships are more important for collaboration. Inspired by sociolinguistic theories, the latest work indicates that individuals’ different attitudes on interrelationships can be measured by interactive language. However, it is still an open problem that what kind of factors influences our different attitudes on interrelationships. In this work, we investigate how individuals’ position i.e., the topological factors in social network influence the differences in our bidirectional attitudes on interrelationships. Measuring the attitudes with interactive language on Enron email dataset, we analyze the correlation between attitudes and the topological factors of email network. The results indicate that individuals’ differences in attitudes on interrelationships are related to some typical topological factors. These results inspire us to measure individuals’ attitude in online collaboration with their topological factors in social network.

Concept in Quantum theory (QT) has been successfully inspired analogous concepts in the field of Information Retrieval (IR). Many IR researchers have employed the QT to investigate cognitive phenomena within user behaviors, and also have verified the existence of quantum-like phenomena in real web search. However, for some complex search task, in which user’s information need (IN) is dynamic and hard to be captured, QT currently adopted still can not explain some more complex cognitive phenomena. In this paper, a user experiment is conducted to investigate the variance of relevance judgement, and its results demonstrate that quantum Weak Measurement (WM) is more appropriate than the standard quantum measurement to model relevance judgement. Further, a WM-based session search model (WSM) is presented to model user’s dynamic evolving IN. The extensive experiments are tested on the session track of TREC 2013 & 2014 and verify the effectiveness of WSM.

Tang poetry, as one of the most typical ways for ancient Chinese to express their emotions, has been continuously inherited for millenniums in China. Nowadays, though our way of life has changed dramatically, the traditional culture about Tang poetry is still affecting us deeply. However, the psychological effect of Tang poetry remains unclear currently. Motivated by this, we aim to investigate the impact of Tang poetry on human emotional state through designing effective experimental paradigm. We gathered 16 Tang poetry video clips with different genres, and played these to 18 subjects, meanwhile recorded their brain neural responses via Electroencephalogram (EEG). A questionnaire was designed to record subjects' subjective ratings of the emotional experience when watching the Tang poetry videos. Through analyzing the questionnaires, we found that Tang poetry can indeed induce the subjects' specific emotions. Finally, we performed a pilot analysis of the recorded EEG signals and the subjective ratings, for exploring the correlation of the brain activity with the emotional states. The results indicate that the Tang poetry can be used as a kind of emotive stimuli in affective computing research, and the EEG induced by Tang poetry can be utilized to probe the human's internal emotions.

Immunomodulatory drugs (IMiDs) are profoundly active compounds in the treatment of patients with multiple myeloma (MM). However, despite the fact that treatment with IMiDs has dramatically improved survival for patients with MM, the majority of MM patients develop IMiDs resistance over time. We have found that expression of functional cereblon is required for IMiDs′ action. In addition, it has been reported that cells expressing high levels of cereblon are resistant to proteasome inhibitor, implying that patients with high levels of cereblon should be resistant to proteasome inhibitor. If the above conclusions are correct, cereblon could be considered as a biomarker to determine which standard regimens should be used to treat patients with MM. Unfortunately, the conclusions mentioned above have not been clinically confirmed. In order to confirm these conclusions, we have generated three highly specific mouse monoclonal antibodies (mAbs) against full-length human cereblon. These mAbs can be used to do western blot, immunoprecipitation and immunohistochemistry staining. In addition, their epitopes have been precisely determined and the peptides covering their epitopes completely blocked the antibody binding to cereblon in western blot analysis or in immunohistochemistry staining of MM patients′ specimens.

The Restricted Boltzmann Machine (RBM) is an important probabilistic graphical model which often serves as a building block for deep belief networks (DBNs). A critical issue for DBNs is overfitting. The generalization ability of models is affected if an inappropriate model complexity is selected. In order to mitigate this problem, we introduce the Pure Dependency-Restricted Boltzmann Machine (PD-RBM). Compared to the regular RBM, the PD-RBM can adaptively change its structure in the face of different data in order to make its model complexity fit data. The establishment of the structure of PD-RBMs is based on Theta Pure Dependency (TPD), which is a statistical measure of variable dependency, defined under the Information Geometry framework. When specific data is given, we show an algorithm that is used to establish the structure of PD-RBMs. We evaluate the PD-RBM on different datasets. Our experiments show that it significantly improves test performance.

Recently, neural network based methods have made remarkable progresses on various Natural Language Processing (NLP) tasks. However, it is still a challenge to model both short and long texts, e.g. sentences and documents. In this paper, we propose a Hierarchical Attention Bidirectional LSTM (HA-BLSTM) to model both sentences and documents. HA-BLSTM effectively obtains a hierarchy of representations from words to phrases through the hierarchical structure. We design two attention mechanisms: local and global attention mechanisms. The local attention mechanism learns which components of a text are more important for modeling the whole text, while the global attention mechanism learns which representations of the same text are crucial. Thus, HA-BLSTM can model long documents along with short sentences. Experiments on four benchmark datasets show that our model yields a superior classification performance over a number of strong baselines.

Music can convey and evoke powerful emotions. But it is very challenging to recognize the music emotions accurately by computational models. The difficulty of the problem can exponentially increase when the music segments delivery multiple and complex emotions. This paper proposes a novel quantum convolutional neural network (QCNN) to learn music emotions. Inheriting the distinguished abstraction ability from deep learning, QCNN automatically extracts the music features that benefit emotion classification. The main contribution of this paper is that we utilize measurement postulate to simulate the human emotion awareness in music appreciation. Statistical experiments on the standard dataset shows that QCNN outperforms the classical algorithms as well as the state-of-the-art in the task of music emotion classification. Moreover, we provide demonstration experiment to explain the good performance of the proposed technique from the perspective of physics and psychology.

Bi-directional LSTM (BLSTM) often utilizes Attention Mechanism (AM) to improve the ability of modeling sentences. But additional parameters within AM may lead to difficulties of model selection and BLSTM training. To solve the problem, this paper redefines AM from a novel perspective of the quantum cognition and proposes a parameter-free Quantum AM (QAM). Furthermore, we make a quantum interpretation for BLSTM with Two-State Vector Formalism (TSVF) and find the similarity between sentence understanding and quantum Weak Measurement (WM) under TSVF. Weak value derived from WM is employed to represent the attention for words in a sentence. Experiments show that QAM based BLSTM outperforms common AM (CAM) [1] based BLSTM on most classification tasks discussed in this paper.