Search

In this paper, we focus on named entity boundary detection , which is to detect the start and end boundaries of an entity mention in text, without predicting its type. The detected entities are input to entity linking or fine-grained typing systems for semantic enrichment. We propose BdryBot , a recurrent neural network encoder-decoder framework with a pointer network to detect entity boundaries from a given sentence. The encoder considers both character-level representations and word-level embeddings to represent the input words. In this way, BdryBot does not require any hand-crafted features. Because of the pointer network, BdryBot overcomes the problem of variable size output vocabulary and the issue of sparse boundary tags. We conduct two sets of experiments, in-domain detection and cross-domain detection, on six datasets. Our results show that BdryBot achieves state-of-the-art performance against five baselines. In addition, our proposed approach can be further enhanced when incorporating contextualized language embeddings into token representations.

Face presentation attack detection (PAD) has become a key component in face-based application systems. Typical face de-spoofing algorithms estimate the noise pattern of a spoof image to detect presentation attacks. These algorithms are device-independent and have good generalization ability. However, the noise modeling is not very effective because there is no ground truth (GT) with identity information for training the noise modeling network. To address this issue, we propose using the bona fide image of the corresponding subject in the training set as a type of GT called appr-GT with the identity information of the spoof image. A metric learning module is proposed to constrain the generated bona fide images from the spoof images so that they are near the appr-GT and far from the input images. This can reduce the influence of imaging environment differences between the appr-GT and GT of a spoof image. Extensive experimental results demonstrate that the reconstructed bona fide image and noise with high discriminative quality can be clearly separated from a spoof image. The proposed algorithm achieves competitive performance.

Dialogue systems have achieved growing success in many areas thanks to the rapid advances of machine learning techniques. In the quest for generating more human-like conversations, one of the major challenges is to learn to generate responses in a more empathetic manner. In this review article, we focus on the literature of empathetic dialogue systems, whose goal is to enhance the perception and expression of emotional states, personal preference, and knowledge. Accordingly, we identify three key features that underpin such systems: emotion-awareness, personality-awareness, and knowledge-accessibility. The main goal of this review is to serve as a comprehensive guide to research and development on empathetic dialogue systems and to suggest future directions in this domain. Agency for Science, Technology and Research (A*STAR) This research is supported by the Agency for Science, Technology and Research (A∗STAR) under its AME Programmatic Funding Scheme (Project #A18A2b0046).

Studying drug relationships can provide deeper information for the construction and maintenance of biomedical databases and provide more important references for disease treatment and drug development. The research model has expanded from the previous focus on a certain drug to the systematic analysis of the pharmaceutical network formed between drugs. Network model is suitable for the study of the nonlinear relationship of the pharmaceutical relationship by modeling the data learning. Association rule mining is used to find the potential correlations between the various sets of massive data. Therefore, based on the network model, this research proposed an algorithm for drug interaction under improved association rules, which achieved accurate analysis and decision-making of drug relationship. Meanwhile, this research applied the established association rule algorithm to discuss the relationship between Chinese medicine and mental illness medicine and conducted the algorithm research and simulation analysis of the association relationship. The results showed the association rule algorithm based on the network model constructed was better than other association algorithms. It had reliability and superiority in decision-making in improving the drug-drug relationship. It also promoted the rational use of medicines and played a guiding role in pharmaceutical research. This provides scientific research personnel with research basis and research ideas for disease-related diagnosis.

In the personal house renting scenario, customers usually make quick assessments based on previous customers' reviews, which makes such reviews essential for the business. If the house is assessed as popular, a Matthew effect will be observed as more people will be willing to book it. Due to the lack of definition and quantity assessment measures, however, it is difficult to make a popularity evaluation and prediction. To solve this problem, the concept of house popularity is well defined in this paper. Specifically, the house popularity is decided by inter-event timeand rating score at the same time. To make a more effective prediction over these two correlated variables, a dual-gated recurrent unit (DGRU) is employed. Furthermore, an encoder-decoder framework with DGRU is proposed to perform popularity prediction. Empirical results show the effectiveness of the proposed DGRU and the encoder-decoder framework in two-correlated sequences prediction and popularity prediction, respectively. Agency for Science, Technology and Research (A*STAR) This research is supported by the Agency for Science, Technology and Research (A*STAR) under its AME Programmatic Funding Scheme (Project #A18A2b0046).

With their growing popularity and widespread applications, face recognition systems are attracting more attention from attackers. Thus, face presentation attack detection has emerged as an important research topic in recent years. Existing methods for face presentation attack detection are affected by different cameras and display devices, and their performance is degraded in cross-database testing. In this paper, we propose a face presentation attack detection scheme that fuses multi-perspective dynamic features. One feature is the globally extracted temporal motion pattern of a face in a video. This involves mapping the local and global motion information of the face in the video into a single image. The motion patterns of genuine and fake faces are different, and these patterns are independent of cameras and display devices. Another feature is the visual rhythm of noise patterns, which differs significantly between single and secondary imaging. The proposed scheme fuses these two features at the decision level. Cross-database tests were conducted among the CASIA-FASD, MSU-MFSD and Replay-Attack databases. The experimental results show that the proposed scheme outperforms state-of-the-art algorithms.

Due to the existence of unfavorable factors such as turbid water quality and target occlusion, it is difficult to obtain valid data of target features. Due to the repeated calculation of similar data, the real-time performance of the algorithm is poor. In view of the above problems, this paper proposes a multi-AUV collaborative target recognition method based on transfer-reinforcement learning. The features of the target information which is collected by multi-AUV are fused based on wavelet transformation and affine invariance. The similarity of features is calculated by Mahalanobis distance and the learning model is selected autonomously based on the similarity threshold. Based on the Q-learning reinforcement learning model, the target information under the interference environment is trained intensively, and the effective features are extracted and stored in the source domain, which can reduce the impact of the environmental interference on the target recognition. The feature transfer learning model based on deep confidence network transfers the feature data of the source domain to the target domain, reducing the repeated calculation of similar data, and then ensuring the real-time performance of the algorithm. Simulation experiments are conducted in the SUN dataset under five underwater environments (turbid water, target occlusion, insufficient light, complex background, and overlapping targets), and the results demonstrate that the proposed model achieves better performance.

Stock market prediction is one of the most challenging problems which has been distressing both researchers and financial analysts for more than half a century. To tackle this problem, two completely opposite approaches, namely technical and fundamental analysis, emerged. Technical analysis bases its predictions on mathematical indicators constructed on the stocks price, while fundamental analysis exploits the information retrieved from news, profitability, and macroeconomic factors. The competition between these schools of thought has led to many interesting achievements, however, to date, no satisfactory solution has been found. Our work aims to combine both technical and fundamental analysis through the application of data science and machine learning techniques. In this paper, the stock market prediction problem is mapped in a classification task of time series data. Indicators of technical analysis and the sentiment of news articles are both exploited as input. The outcome is a robust predictive model able to forecast the trend of a portfolio composed by the twenty most capitalized companies listed in the NASDAQ100 index. As a proof of real effectiveness of our approach, we exploit the predictions to run a high frequency trading simulation reaching more than 80% of annualized return. This project represents a step forward to combine technical and fundamental analysis and provides a starting point for developing new trading strategies.

Understanding the entire urban diffuse pollutant transport process in a Low Impact Development (LID) pilot catchment is critical for developing efficient and reliable LID strategies to remove stormwater pollutants. Previous studies focused on the LID performance at a system scale and modeling work. However, the influence of LID practices on the entire urban diffuse pollutant transport process at a catchment scale through field studies remains unclear. The present study analyzed particle loads, particle size distribution and heavy metal concentrations during the accumulation-release-transport process of urban diffuse pollutants. It was found that 13.11% particles can be washed-off from the road surfaces, and a vast majority (12.40%) of suspended solids (SS) in the road surface runoff was retained by LID while only a small fraction (0.71%) can overflow through the LID system and finally 0.42% SS was discharged from the drain outlet. This result implies that LID practices can significantly reduce the contribution of road deposited sediments pollution to urban receiving water. Outcomes also found that most heavy metals in stormwater are particulate, and furthermore, finer particles contain more heavy metals and are easily removed by the stormwater runoff. Therefore, it is essential to consider the influence of particle size on SS removal performance of LID and apply enhanced LID techniques which can remove the fine particles, especially the particles 10–105 μm, in the stormwater.

Cortical structures in the central nervous system exhibit an ordered laminar organization. Defined cell layers are significant to our understanding of brain structure and function. In this work, multiphoton microscopy (MPM), based on second harmonic generation and two-photon excited fluorescence, is applied for visualizing the mouse and human cerebral cortex from fresh, unfixed, and unstained specimen. We show that label-free MPM is able to effectively visualize microscopic scale of cortical neurons, six-layer cortical lamination, and functional cortex that correlates with the corresponding stained histological images. Moreover, we further demonstrate that MPM is feasible for identifying focal cortical dysplasia (FCD) in the human cerebral cortex. The border between gray and white matter, abnormal radial cortical lamination, immature small diameter neurons, pia mater, and blood vessels are also clearly visualized as the typical histopathologic hallmark of FCD. In addition, to diagnose FCD more accurately and rapidly, an image processing method is employed to extract MPM image features for auxiliary identification of FCD specimen. These results offer additional metrics to pathologists and surgeons, which hold the translational potential to be used as a diagnostic tool into routine clinical use.

The Chinese pronunciation system offers two characteristics that distinguish it from other languages: deep phonemic orthography and intonation variations. In this paper, we hypothesize that these two important properties can play a major role in Chinese sentiment analysis. In particular, we propose two effective features to encode phonetic information and, hence, fuse it with textual information. With this hypothesis, we propose Disambiguate Intonation for Sentiment Analysis (DISA), a network that we develop based on the principles of reinforcement learning. DISA disambiguates intonations for each Chinese character (pinyin) and, hence, learns precise phonetic representations. We also fuse phonetic features with textual and visual features to further improve performance. Experimental results on five different Chinese sentiment analysis datasets show that the inclusion of phonetic features significantly and consistently improves the performance of textual and visual representations and surpasses the state-of-the-art Chinese character-level representations.

The complex underwater environment leads to a small number of samples and a serious lack of target information. To solve the above problems, this paper proposes an underwater target detection method based on third party information transfer learning (TITL). Firstly, the method obtains the targets feature description from the knowledge base as third-party information. Using the third-party information and target information fusion, constitute a new target domain. Solved the problem of low detection accuracy caused by insufficient target data. Secondly, according to the uneven distribution of underwater image features, this method introduces the feature distribution difference adaptive principle into integrated migration learning. This method reduces the time of feature mapping and ensures the accuracy of migration learning. Experimental results show that the algorithm is effective in the Underwater Target dataset compared with existing algorithms.

Existing algorithms aiming to learn a binary classifier from positive (P) and unlabeled (U) data generally require estimating the class prior or label noises ahead of building a classification model. However, the estimation and classifier learning are normally conducted in a pipeline instead of being jointly optimized. In this paper, we propose to alternatively train the two steps using reinforcement learning. Our proposal adopts a policy network to adaptively make assumptions on the labels of unlabeled data, while a classifier is built upon the output of the policy network and provides rewards to learn a better strategy. The dynamic and interactive training between the policy maker and the classifier can exploit the unlabeled data in a more effective manner and yield a significant improvement on the classification performance. Furthermore, we present two different approaches to represent the actions sampled from the policy. The first approach considers continuous actions as soft labels, while the other uses discrete actions as hard assignment of labels for unlabeled examples.We validate the effectiveness of the proposed method on two benchmark datasets as well as one e-commerce dataset. The result shows the proposed method is able to consistently outperform state-of-the-art methods in various settings., Proceedings of the IEEE International Conference on Data Mining, pp. 309-408, Beijing, 2019

The names of software artifacts, e.g., method names, are important for software understanding and maintenance, as good names can help developers easily understand others’ code. However, the existing naming guidelines are difficult for developers, especially novices, to come up with meaningful, concise and compact names for the variables, methods, classes and files. With the popularity of open source, an enormous amount of project source code can be accessed, and the exhaustiveness and instability of manually naming methods could now be relieved by automatically learning a naming model from a large code repository. Nevertheless, building a comprehensive naming system is still challenging, due to the gap between natural language functional descriptions and method names. Specifically, there are three challenges: how to model the relationship between the functional descriptions and formal method names, how to handle the explosion of vocabulary when dealing with large repositories, and how to leverage the knowledge learned from large repositories to a specific project. To answer these questions, we propose a neural network to directly generate readable method names from natural language description. The proposed method is built upon the encoder-decoder framework with the attention and copying mechanisms. Our experiments show that our method can generate meaningful and accurate method names and achieve significant improvement over the state-of-the-art baseline models. We also address the cold-start problem using a training trick to utilize big data in Github for specific projects.

Over the last twenty years, researchers and practitioners have attempted in many ways to effectively predict market trends. Till date, however, no satisfactory solution has been found. Many approaches have been applied to predict market trends, from technical analysis to fundamental analysis passing through sentiment analysis. A promising research direction is to exploit market technical indicators together with market sentiments extracted from social media for predicting market directional movements. In this paper, we propose a new approach that leverages technical analysis to predict market directional movements. In particular, we aim to predict the directional movement of the NASDAQ's most capitalized stocks by solving a classification problem. The results on real-world data show that our proposal achieves interesting performance when predicting the market directional movements. This work focuses on forecasting a portfolio of different stocks, instead of concentrating on a single stock which most of the works in this field do. Furthermore, the proposed model is able to solve the issue of skewed classes through the use of appropriate data balancing techniques.

Predicting stock market movements is an interesting and challenging problem: researchers and traders have approached this task with different techniques, from time series prediction to technical and fundamental analysis. Nowadays, a huge amount of textual data can be used to lead a new source of information on this task, well known to be highly stochastic and temporal dependent. In this paper, we investigate the problem of analyzing the timing and the impact that news have on the stock market. We focus on two different important aspects: the possible contribution that new information can have on the stock price and its possible relation with the aggregate news recently published. Our approach involves using the information available now to predict different prices movements, from the recent past to the far future. Results on US market show that the information contained in news can be used as an accurate predictor for past price movements. When the future is considered, however, the correlation between news and stock market becomes less clear.

Face anti-spoofing based on deep learning achieved good accuracy recently. However, deep learning model has no explicit mathematical presentation. Therefore, it is not clear about how the model works effectively. In this paper, we estimate the regions in face image, which are sensitive in deep learning based anti-spoofing algorithms. We first generate the adversarial examples from two different gradient-based methods. Then we analyze the distribution of the gradient and perturbations on the adversarial examples. And next we obtain the sensitive regions and evaluate the contribution of these regions to classification performance. By analyzing the sensitive regions, it could be observed that the CNN based anti-spoofing algorithms are sensitive to rich detailed regions and illumination. These observations are helpful to design an effective face anti-spoofing algorithm.

Discriminative language models (DLMs) have been widely used for reranking competing hypotheses produced by an Automatic Speech Recognition (ASR) system. While existing DLMs suffer from limited generalization power, we propose a novel DLM based on a discriminatively trained Restricted Boltzmann Machine (RBM). The hidden layer of the RBM improves generalization and allows for employing additional prior knowledge, including pre-trained parameters and entity-related prior. Our approach outperforms the single-layer-perceptron (SLP) reranking model, and fusing our approach with SLP achieves up to 1.3% absolute Word Error Rate (WER) reduction and a relative 180% improvement in terms of WER reduction over the SLP reranker. In particular, it shows that proposed prior informed RBM reranker achieves largest ASR error reduction (3.1% absolute WER) on content words.

Sentiment analysis has emerged as one of the most popular natural language processing (NLP) tasks in recent years. A classic setting of the task mainly involves classifying the overall sentiment polarity of the inputs. However, it is based on the assumption that the sentiment expressed in a sentence is unified and consistent, which does not hold in the reality. As a fine-grained alternative of the task, analyzing the sentiment towards a specific target and aspect has drawn much attention from the community for its more practical assumption that sentiment is dependent on a particular set of aspects and entities. Recently, deep neural models have achieved great successes on sentiment analysis. As a functional simulation of the behavior of human brains and one of the most successful deep neural models for sequential data, long short-term memory (LSTM) networks are excellent in learning implicit knowledge from data. However, it is impossible for LSTM to acquire explicit knowledge such as commonsense facts from the training data for accomplishing their specific tasks. On the other hand, emerging knowledge bases have brought a variety of knowledge resources to our attention, and it has been acknowledged that incorporating the background knowledge is an important add-on for many NLP tasks. In this paper, we propose a knowledge-rich solution to targeted aspect-based sentiment analysis with a specific focus on leveraging commonsense knowledge in the deep neural sequential model. To explicitly model the inference of the dependent sentiment, we augment the LSTM with a stacked attention mechanism consisting of attention models for the target level and sentence level, respectively. In order to explicitly integrate the explicit knowledge with implicit knowledge, we propose an extension of LSTM, termed Sentic LSTM. The extended LSTM cell includes a separate output gate that interpolates the token-level memory and the concept-level input. In addition, we propose an extension of Sentic LSTM by creating a hybrid of the LSTM and a recurrent additive network that simulates sentic patterns. In this paper, we are mainly concerned with a joint task combining the target-dependent aspect detection and targeted aspect-based polarity classification. The performance of proposed methods on this joint task is evaluated on two benchmark datasets. The experiment shows that the combination of proposed attention architecture and knowledge-embedded LSTM could outperform state-of-the-art methods in two targeted aspect sentiment tasks. We present a knowledge-rich solution for the task of targeted aspect-based sentiment analysis. Our model can effectively incorporate the commonsense knowledge into the deep neural network and be trained in an end-to-end manner. We show that the two-step attentive neural architecture as well as the proposed Sentic LSTM and H-Sentic-LSTM can achieve an improved performance on resolving the aspect categories and sentiment polarity for a targeted entity in its context over state-of-the-art systems.

Aspect-based sentiment analysis aims at identifying sentiment polarity towards aspect targets in a sentence. Previously, the task was modeled as a sentence-level sentiment classification problem that treated aspect targets as a hint. Such approaches oversimplify the problem by averaging word embeddings when the aspect target is a multi-word sequence. In this paper, we formalize the problem from a different perspective, i.e., that sentiment at aspect target level should be the main focus. Due to the fact that written Chinese is very rich and complex, Chinese aspect targets can be studied at three different levels of granularity: radical, character and word. Thus, we propose to explicitly model the aspect target and conduct sentiment classification directly at the aspect target level via three granularities. Moreover, we study two fusion methods for such granularities in the task of Chinese aspect-level sentiment analysis. Experimental results on a multi-word aspect target subset from SemEval2014 and four Chinese review datasets validate our claims and show the improved performance of our model over the state of the art.

This paper leverages heterogeneous auxiliary information to address the data sparsity problem of recommender systems. We propose a model that learns a shared feature space from heterogeneous data, such as item descriptions, product tags and online purchase history, to obtain better predictions. Our model consists of autoencoders, not only for numerical and categorical data, but also for sequential data, which enables capturing user tastes, item characteristics and the recent dynamics of user preference. We learn the autoencoder architecture for each data source independently in order to better model their statistical properties. Our evaluation on two MovieLens datasets and an e-commerce dataset shows that mean average precision and recall improve over state-of-the-art methods., Comment: Proceedings of the IEEE International Conference on Data Mining, pp. 1164-1169, Singapore, 2018

Referencing URLs of external web resources (e.g., official language references and API documents) is an effective mechanism for knowledge sharing in Q&A websites like Stack Overflow. We show that reference frequencies of URLs follow power law distribution, meaning that web resources that have been referenced frequently will likely to be referenced again. However, there lack of effective methods to manage and reuse already-shared web resources relevant to entities (e.g., APIs or programming concepts) that are mentioned in Q&A discussions. As URL references are done in an ad-hoc manner, large amounts of entity mentions have not been linked to relevant web resources. To enhance management and reuse of alreadyshared web resources in Stack Overflow, we build a knowledge base of official documentation of languages and APIs that have been shared in Stack Overflow, and develop an automatic web resources linking technique to linkify entity mentions to relevant official documentation in the knowledge base. A challenge in automatic web resources linking is that entity mentions often have ambiguity, for example, same programming concepts across different languages, same name APIs in different libraries. To disambiguate the right web resource to link among several URL candidates for an entity mention, our technique examines both the global popularity of the URL candidates for the entity mention and the local context relatedness of the URL candidates with the discussion thread in which the entity is mentioned. We conduct large scale evaluation of the built knowledge base and the performance of our automatic web resource linking technique.

With the rapid development of face recognition systems in various practical applications, numerous face spoofing attacks under different environment and devices have emerged. The countermeasure of face spoofing attacks in cross-database have caused increasing attention. This paper proposes a face spoofing detection method with motion analysis based cross-database voting. We employ the consistency motion information of different databases like eye-blink, mouth movements and facial expression etc. Then the motion information maps of a video is classified to real or fake by CNN model. Furthermore, cross-database voting strategy is constructed to transfer motion characteristics from a database to another for face spoofing inference. Experimental results demonstrate that the proposed method outperforms its comparisons taking benefits of motion analysis based CNN classification and cross-database voting.

Named entity recognition (NER) from open-domain conversation is challenging due to the informality of spoken language. Instead of increasing the size of labeled data, which is expensive and time-consuming, word embeddings learned from unlabeled data have been used by NER models to handle data sparsity. We propose a novel method for training the word embeddings specifically for the NER task. We show that our task-specific word embeddings outperform task-independent word embeddings when used as features of NER method.