We present AliMe Avatar, a Vtuber designed for live-streaming sales in the E-commerce field. To support the emerging live shopping mode, the core of our digitial avatar is to enable customers to understand products and encourage customers to purchase in a virtual broadcasting room. Based on computer graphics & vision, natural language processing, and speech recognition & synthesis, our AI avatar is able to offer three kinds of key capabilities: custom appearance, product broadcasting, and multi-modal interaction. Currently, it has been launched online in the Taobao app, broadcasts 700+ hours and serves hundreds of thousands of customers per day. In this paper, we mainly focus on the product broadcasting part, demonstrate the system, present the underlying techniques, and share our experience in dealing with live-streaming E-commerce.

Abstract Machine learning methods, especially deep learning models, have achieved impressive performance in various natural language processing tasks including sentiment analysis. However, deep learning models are more demanding for training data. Data augmentation techniques are widely used to generate new instances based on modifications to existing data or relying on external knowledge bases to address annotated data scarcity, which hinders the full potential of machine learning techniques. This paper presents our work using part‐of‐speech (POS) focused lexical substitution for data augmentation (PLSDA) to enhance the performance of machine learning algorithms in sentiment analysis. We exploit POS information to identify words to be replaced and investigate different augmentation strategies to find semantically related substitutions when generating new instances. The choice of POS tags as well as a variety of strategies such as semantic‐based substitution methods and sampling methods are discussed in detail. Performance evaluation focuses on the comparison between PLSDA and two previous lexical substitution‐based data augmentation methods, one of which is thesaurus‐based, and the other is lexicon manipulation based. Our approach is tested on five English sentiment analysis benchmarks: SST‐2, MR, IMDB, Twitter, and AirRecord. Hyperparameters such as the candidate similarity threshold and number of newly generated instances are optimized. Results show that six classifiers (SVM, LSTM, BiLSTM‐AT, bidirectional encoder representations from transformers [BERT], XLNet, and RoBERTa) trained with PLSDA achieve accuracy improvement of more than 0.6% comparing to two previous lexical substitution methods averaged on five benchmarks. Introducing POS constraint and well‐designed augmentation strategies can improve the reliability of lexical data augmentation methods. Consequently, PLSDA significantly improves the performance of sentiment analysis algorithms.

Attention models are proposed in sentiment analysis and other classification tasks because some words are more important than others to train the attention models. However, most existing methods either use local context based information, affective lexicons, or user preference information. In this work, we propose a novel attention model trained by cognition grounded eye-tracking data. First,a reading prediction model is built using eye-tracking data as dependent data and other features in the context as independent data. The predicted reading time is then used to build a cognition grounded attention layer for neural sentiment analysis. Our model can capture attentions in context both in terms of words at sentence level as well as sentences at document level. Other attention mechanisms can also be incorporated together to capture other aspects of attentions, such as local attention, and affective lexicons. Results of our work include two parts. The first part compares our proposed cognition ground attention model with other state-of-the-art sentiment analysis models. The second part compares our model with an attention model based on other lexicon based sentiment resources. Evaluations show that sentiment analysis using cognition grounded attention model outperforms the state-of-the-art sentiment analysis methods significantly. Comparisons to affective lexicons also indicate that using cognition grounded eye-tracking data has advantages over other sentiment resources by considering both word information and context information. This work brings insight to how cognition grounded data can be integrated into natural language processing (NLP) tasks.

Automatic Chinese irony detection is a challenging task, and it has a strong impact on linguistic research. However, Chinese irony detection often lacks labeled benchmark datasets. In this paper, we introduce Ciron, the first Chinese benchmark dataset available for irony detection for machine learning models. Ciron includes more than 8.7K posts, collected from Weibo, a micro blogging platform. Most importantly, Ciron is collected with no pre-conditions to ensure a much wider coverage. Evaluation on seven different machine learning classifiers proves the usefulness of Ciron as an important resource for Chinese irony detection.

Deep neural network models have played a critical role in sentiment analysis with promising results in the recent decade. One of the essential challenges, however, is how external sentiment knowledge can be effectively utilized. In this work, we propose a novel affection-driven approach to incorporating affective knowledge into neural network models. The affective knowledge is obtained in the form of a lexicon under the Affect Control Theory (ACT), which is represented by vectors of three-dimensional attributes in Evaluation, Potency, and Activity (EPA). The EPA vectors are mapped to an affective influence value and then integrated into Long Short-term Memory (LSTM) models to highlight affective terms. Experimental results show a consistent improvement of our approach over conventional LSTM models by 1.0% to 1.5% in accuracy on three large benchmark datasets. Evaluations across a variety of algorithms have also proven the effectiveness of leveraging affective terms for deep model enhancement.

Most past works for DNA-binding residue prediction did not consider the relationships between residues. In this paper, we propose a novel approach for DNA-binding residue prediction, referred to as EL_LSTM, which includes two main components. The first component is the Long Short-Term Memory (LSTM), which learns pairwise relationships between residues through a bi-gram model and then learns feature vectors for all residues. The second component is an ensemble learning based classifier introduced to tackle the data imbalance problem in binding residue predictions. We use a variant of the bagging strategy in ensemble learning to achieve balanced samples. Evaluations on PDNA-224 and DBP-123 show that adding feature relationships performs better than classifiers without feature relationships by at least 0.028 on MCC, 1.18 percent on ST and 0.012 on AUC. This indicates the usefulness of feature relationships for DNA-binding residue predictions. Evaluation on using ensemble learning indicates that the improvement can reach at least 0.021 on MCC, 1.32 percent on ST, and 0.018 on AUC compared to the use of a single LSTM classifier. Comparisons with the state-of-the-art predictors show that our proposed EL_LSTM outperforms them significantly. Further feature analysis validates the effectiveness of LSTM for the prediction of DNA-binding residues.

Most proposed methods for TF-binding site (TFBS) predictions only use low order dependencies for predictions due to the lack of efficient methods to extract higher order dependencies. In this work, we first propose a novel method to extract higher order dependencies by applying CNN on histone modification features. We then propose a novel TFBS prediction method, referred to as CNN_TF, by incorporating low order and higher order dependencies. CNN_TF is first evaluated on 13 TFs in the mES cell. Results show that using higher order dependencies outperforms low order dependencies significantly on 11 TFs. This indicates that higher order dependencies are indeed more effective for TFBS predictions than low order dependencies. Further experiments show that using both low order dependencies and higher order dependencies improves performance significantly on 12 TFs, indicating the two dependency types are complementary. To evaluate the influence of cell-types on prediction performances, CNN_TF was applied to five TFs in five cell-types of humans. Even though low order dependencies and higher order dependencies show different contributions in different cell-types, they are always complementary in predictions. When comparing to several state-of-the-art methods, CNN_TF outperforms them by at least 5.3 percent in AUPR.

MOTIVATION: The prediction of transcription factor binding sites (TFBSs) is crucial for gene expression analysis. Supervised learning approaches for TFBS predictions require large amounts of labeled data. However, many TFs of certain cell types either do not have sufficient labeled data or do not have any labeled data. RESULTS: In this paper, a multi-task learning framework (called MTTFsite) is proposed to address the lack of labeled data problem by leveraging on labeled data available in cross-cell types. The proposed MTTFsite contains a shared CNN to learn common features for all cell types and a private CNN for each cell type to learn private features. The common features are aimed to help predicting TFBSs for all cell types especially those cell types that lack labeled data. MTTFsite is evaluated on 241 cell type TF pairs and compared with a baseline method without using any multi-task learning model and a fully shared multi-task model that uses only a shared CNN and do not use private CNNs. For cell types with insufficient labeled data, results show that MTTFsite performs better than the baseline method and the fully shared model on more than 89% pairs. For cell types without any labeled data, MTTFsite outperforms the baseline method and the fully shared model by more than 80 and 93% pairs, respectively. A novel gene expression prediction method (called TFChrome) using both MTTFsite and histone modification features is also presented. Results show that TFBSs predicted by MTTFsite alone can achieve good performance. When MTTFsite is combined with histone modification features, a significant 5.7% performance improvement is obtained. AVAILABILITY AND IMPLEMENTATION: The resource and executable code are freely available at http://hlt.hitsz.edu.cn/MTTFsite/ and http://www.hitsz-hlt.com:8080/MTTFsite/. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

BACKGROUND: Protein secondary structure is the three dimensional form of local segments of proteins and its prediction is an important problem in protein tertiary structure prediction. Developing computational approaches for protein secondary structure prediction is becoming increasingly urgent. RESULTS: We present a novel deep learning based model, referred to as CNNH_PSS, by using multi-scale CNN with highway. In CNNH_PSS, any two neighbor convolutional layers have a highway to deliver information from current layer to the output of the next one to keep local contexts. As lower layers extract local context while higher layers extract long-range interdependencies, the highways between neighbor layers allow CNNH_PSS to have ability to extract both local contexts and long-range interdependencies. We evaluate CNNH_PSS on two commonly used datasets: CB6133 and CB513. CNNH_PSS outperforms the multi-scale CNN without highway by at least 0.010 Q8 accuracy and also performs better than CNF, DeepCNF and SSpro8, which cannot extract long-range interdependencies, by at least 0.020 Q8 accuracy, demonstrating that both local contexts and long-range interdependencies are indeed useful for prediction. Furthermore, CNNH_PSS also performs better than GSM and DCRNN which need extra complex model to extract long-range interdependencies. It demonstrates that CNNH_PSS not only cost less computer resource, but also achieves better predicting performance. CONCLUSION: CNNH_PSS have ability to extracts both local contexts and long-range interdependencies by combing multi-scale CNN and highway network. The evaluations on common datasets and comparisons with state-of-the-art methods indicate that CNNH_PSS is an useful and efficient tool for protein secondary structure prediction.

202101 bcrc

Emotion cause extraction aims to identify the reasons behind a certain emotion expressed in text.It is a much more difficult task compared to emotion classification.Inspired by recent advances in using deep memory networks for question answering (QA), we propose a new approach which considers emotion cause identification as a reading comprehension task in QA.Inspired by convolutional neural networks, we propose a new mechanism to store relevant context in different memory slots to model context information.Our proposed approach can extract both word level sequence features and lexical features.Performance evaluation shows that our method achieves the state-of-the-art performance on a recently released emotion cause dataset, outperforming a number of competitive baselines by at least 3.01% in F-measure.

Attention models are proposed in sentiment analysis because some words are more important than others.However, most existing methods either use local context based text information or user preference information.In this work, we propose a novel attention model trained by cognition grounded eye-tracking data.A reading prediction model is first built using eye-tracking data as dependent data and other features in the context as independent data.The predicted reading time is then used to build a cognition based attention (CBA) layer for neural sentiment analysis.As a comprehensive model, We can capture attentions of words in sentences as well as sentences in documents.Different attention mechanisms can also be incorporated to capture other aspects of attentions.Evaluations show the CBA based method outperforms the state-of-the-art local context based attention methods significantly.This brings insight to how cognition grounded data can be brought into NLP tasks.

BACKGROUND: Prediction of DNA-binding residue is important for understanding the protein-DNA recognition mechanism. Many computational methods have been proposed for the prediction, but most of them do not consider the relationships of evolutionary information between residues. RESULTS: In this paper, we first propose a novel residue encoding method, referred to as the Position Specific Score Matrix (PSSM) Relation Transformation (PSSM-RT), to encode residues by utilizing the relationships of evolutionary information between residues. PDNA-62 and PDNA-224 are used to evaluate PSSM-RT and two existing PSSM encoding methods by five-fold cross-validation. Performance evaluations indicate that PSSM-RT is more effective than previous methods. This validates the point that the relationship of evolutionary information between residues is indeed useful in DNA-binding residue prediction. An ensemble learning classifier (EL_PSSM-RT) is also proposed by combining ensemble learning model and PSSM-RT to better handle the imbalance between binding and non-binding residues in datasets. EL_PSSM-RT is evaluated by five-fold cross-validation using PDNA-62 and PDNA-224 as well as two independent datasets TS-72 and TS-61. Performance comparisons with existing predictors on the four datasets demonstrate that EL_PSSM-RT is the best-performing method among all the predicting methods with improvement between 0.02-0.07 for MCC, 4.18-21.47% for ST and 0.013-0.131 for AUC. Furthermore, we analyze the importance of the pair-relationships extracted by PSSM-RT and the results validates the usefulness of PSSM-RT for encoding DNA-binding residues. CONCLUSIONS: We propose a novel prediction method for the prediction of DNA-binding residue with the inclusion of relationship of evolutionary information and ensemble learning. Performance evaluation shows that the relationship of evolutionary information between residues is indeed useful in DNA-binding residue prediction and ensemble learning can be used to address the data imbalance issue between binding and non-binding residues. A web service of EL_PSSM-RT ( http://hlt.hitsz.edu.cn:8080/PSSM-RT_SVM/ ) is provided for free access to the biological research community.

Affective lexicon is one of the most important resource in affective computing for text. Manually constructed affective lexicons have limited scale and thus only have limited use in practical systems. In this work, we propose a regression-based method to automatically infer multi-dimensional affective representation of words via their word embedding based on a set of seed words. This method can make use of the rich semantic meanings obtained from word embedding to extract meanings in some specific semantic space. This is based on the assumption that different features in word embedding contribute differently to a particular affective dimension and a particular feature in word embedding contributes differently to different affective dimensions. Evaluation on various affective lexicons shows that our method outperforms the state-of-the-art methods on all the lexicons under different evaluation metrics with large margins. We also explore different regression models and conclude that the Ridge regression model, the Bayesian Ridge regression model and Support Vector Regression with linear kernel are the most suitable models. Comparing to other state-of-the-art methods, our method also has computation advantage. Experiments on a sentiment analysis task show that the lexicons extended by our method achieve better results than publicly available sentiment lexicons on eight sentiment corpora. The extended lexicons are publicly available for access.

In this paper, we claim that vector cosine – which is generally considered among the most efficient unsupervised measures for identifying word similarity in Vector Space Models – can be outperformed by an unsupervised measure that calculates the extent of the intersection among the most mutually dependent contexts of the target words. To prove it, we describe and evaluate APSyn, a variant of the Average Precision that, without any optimization, outperforms the vector cosine and the co-occurrence on the standard ESL test set, with an improvement ranging between +9.00% and +17.98%, depending on the number of chosen top contexts.

In this paper, we describe ROOT13, a supervised system for the classification of hypernyms, co-hyponyms and random words. The system relies on a Random Forest algorithm and 13 unsupervised corpus-based features. We evaluate it with a 10-fold cross validation on 9,600 pairs, equally distributed among the three classes and involving several Parts-Of-Speech (i.e. adjectives, nouns and verbs). When all the classes are present, ROOT13 achieves an F1 score of 88.3%, against a baseline of 57.6% (vector cosine). When the classification is binary, ROOT13 achieves the following results: hypernyms-co-hyponyms (93.4% vs. 60.2%), hypernyms-random (92.3% vs. 65.5%) and co-hyponyms-random (97.3% vs. 81.5%). Our results are competitive with state-of-the-art models.

Protein-DNA complexes play crucial roles in gene regulation. The prediction of the residues involved in protein-DNA interactions is critical for understanding gene regulation. Although many methods have been proposed, most of them overlooked motif features. Motif features are sub sequences and are important for the recognition between a protein and DNA. In order to efficiently use motif features for the prediction of DNA-binding residues, we first apply the Convolutional Neural Network (CNN) method to capture the motif features from the sequences around the target residues. CNN modeling consists of a set of learnable motif detectors that can capture the important motif features by scanning the sequences around the target residues. Then we use a neural network classifier, referred to as CNNsite, by combining the captured motif features, sequence features and evolutionary features to predict binding residues from sequences. The datasets PDNA-62 and PDNA-224 are used to evaluate the performance of CNNsite by five-fold cross-validation. Performance evaluation shows that the motif features performs better than sequence features and evolutionary features with at least 6.73% on ST, 0.097 on MCC and 0.069 on AUC. When comparing with previously published methods, CNNsite performs better with at least 0.019 on MCC, 4.37% on ST and 0.040 on AUC. CNNsite is also evaluated on an independent dataset TS-72 and CNNsite outperforms the previous methods by at least 0.012 on AUC. The discriminant powers of the motif features of size from 2 to 6 residues show that many motif features with large discriminant power are composed by the residues that play important roles in the DNA-protein interactions. The standalone version of the CNNsite is available at http://hlt.hitsz.edu.cn:8080/CNNsite/.

Inferring latent user preferences using both structured and unstructured data is an important social computing task. In this paper, we propose a user preference representation based on user activities embedded in unstructured data to better encode the homophily theory. The representation of an individual user is learned using a embedding based method to integrate latent user preferences in social media. The method has the ability to integrate a variety of user activities based cues from user comments, user social network (i.e; follower/followee connections) and user interested topics which are indicated by the topics a user has participated in. Experiments are conducted to evaluate the prediction of each user's favorite team as a part of user preferences in a dataset collected from the Hu-pu basketball discussion forum.1Results clearly indicate that our proposed user representation outperforms other user representation baselines. Integrating user social network and user interested topics with user comments can improve the overall performance of user preference prediction.

In this paper, we present our work in emotion cause extraction.Since there is no open dataset available, the lack of annotated resources has limited the research in this area.Thus, we first present a dataset we built using SINA city news.The annotation is based on the scheme of the W3C Emotion Markup Language.Second, we propose a 7-tuple definition to describe emotion cause events.Based on this general definition, we propose a new event-driven emotion cause extraction method using multi-kernel SVMs where a syntactical tree based approach is used to represent events in text.A convolution kernel based multikernel SVM are used to extract emotion causes.Because traditional convolution kernels do not use lexical information at the terminal nodes of syntactic trees, we modify the kernel function with a synonym based improvement.Even with very limited training data, we can still extract sufficient features for the task.Evaluations show that our approach achieves 11.6% higher F-measure compared to referenced methods.The contributions of our work include resource construction, concept definition and algorithm development.

Information ordering is a nontrivial task in multi‐document summarization (MDS), which typically relies on the traditional vector space model (VSM) notorious for semantic deficiency. In this article, we propose a novel event‐enriched VSM to alleviate the problem by building event semantics into sentence representations. The mediation of event information between sentence and term, especially in the news domain, has an intuitive appeal as well as technical advantage in common sentence‐level operations such as sentence similarity computation. Inspired by the block‐style writing by humans, we base the sentence ordering algorithm on sentence clustering. To accommodate the complexity introduced by event information, we adopt a soft‐to‐hard clustering strategy on the event and sentence levels, using expectation–maximization clustering and K ‐means, respectively. For the purpose of cluster‐based sentence ordering, the event‐enriched VSM enables us to design an ordering algorithm to enhance event coherence computed between sentence and sentence–context pairs. Drawing on the findings of earlier research, we also incorporate topic continuity measures and time information into the scheme. We evaluate the performance of the model and its variants automatically and manually, with experimental results showing clear advantage of the event‐based model over baseline and non‐event‐based models in information ordering for multi‐document news summarization. We are confident that the event‐enriched VSM has even greater potential in summarization and beyond, which awaits further research. © 2014 Wiley Periodicals, Inc.

Noise in class labels of any training set can lead to poor classification results no matter what machine learning method is used. In this paper, we first present the problem of binary classification in the presence of random noise on the class labels, which we call class noise. To model class noise, a class noise rate is normally defined as a small independent probability of the class labels being inverted on the whole set of training data. In this paper, we propose a method to estimate class noise rate at the level of individual samples in real data. Based on the estimation result, we propose two approaches to handle class noise. The first technique is based on modifying a given surrogate loss function. The second technique eliminates class noise by sampling. Furthermore, we prove that the optimal hypothesis on the noisy distribution can approximate the optimal hypothesis on the clean distribution using both approaches. Our methods achieve over 87% accuracy on a synthetic non-separable dataset even when 40% of the labels are inverted. Comparisons to other algorithms show that our methods outperform state-of-the-art approaches on several benchmark datasets in different domains with different noise rates.

Transfer learning has been used as a machine learning method to make good use of available language resources for other resource-scarce languages. However, the cumulative class noise during iterations of transfer learning can lead to negative transfer which can adversely affect performance when more training data is used. In this paper, we propose a novel transfer learning method which can detect negative transfers. This approach detects high quality samples after certain iterations to identify class noise in new transferred training samples and remove them to reduce misclassifications. With the ability to detect bad training samples and remove them, our method can make full use of large unlabeled training data available in the target language. Furthermore, the most important contribution in this paper is the theory of class noise detection. Our new class noise detection method overcame the theoretic flaw of a previous method based on Gaussian distribution. We applied this transfer learning method with negative transfer detection to cross lingual opinion analysis. Evaluation on the NLP&CC 2013 cross-lingual opinion analysis dataset shows that the proposed approach outperforms the state-of-the-art systems.

Enrico Santus, Alessandro Lenci, Qin Lu, Sabine Schulte im Walde. Proceedings of the 14th Conference of the European Chapter of the Association for Computational Linguistics, volume 2: Short Papers. 2014.

Training speed and accuracy are two major concerns of large-scale natural language processing systems. Typically, we need to make a tradeoff between speed and accuracy. It is trivial to improve the training speed via sacrificing accuracy or to improve the accuracy via sacrificing speed. Nevertheless, it is nontrivial to improve the training speed and the accuracy at the same time, which is the target of this work. To reach this target, we present a new training method, feature-frequency–adaptive on-line training, for fast and accurate training of natural language processing systems. It is based on the core idea that higher frequency features should have a learning rate that decays faster. Theoretical analysis shows that the proposed method is convergent with a fast convergence rate. Experiments are conducted based on well-known benchmark tasks, including named entity recognition, word segmentation, phrase chunking, and sentiment analysis. These tasks consist of three structured classification tasks and one non-structured classification task, with binary features and real-valued features, respectively. Experimental results demonstrate that the proposed method is faster and at the same time more accurate than existing methods, achieving state-of-the-art scores on the tasks with different characteristics.

This paper proposes a novel approach using a coarse-to-fine analysis strategy for sentence-level emotion classification which takes into consideration of similarities to sentences in training set as well as adjacent sentences in the context. First, we use intra-sentence based features to determine the emotion label set of a target sentence coarsely through the statistical information gained from the label sets of the k most similar sentences in the training data. Then, we use the emotion transfer probabilities between neighboring sentences to refine the emotion labels of the target sentences. Such iterative refinements terminate when the emotion classification converges. The proposed algorithm is evaluated on Ren-CECps, a Chinese blog emotion corpus. Experimental results show that the coarse-to-fine emotion classification algorithm improves the sentence-level emotion classification by 19.11% on the average precision metric, which outperforms the baseline methods.

2011-2012 > Academic research: refereed > Refereed conference paper

The key issue of person name disambiguation is to discover different namesakes in massive web documents rather than simply cluster documents by using textual features. In this paper, we describe a novel person name disambiguation method based on social networks to effectively identify namesakes. The social network snippets in each document are extracted. Then, the namesakes are identified via splicing the social networks of each namesake by using the snippets as a bipartite graph. Experimental results show that our method achieves better result than the top performance of WePS-2 in identifying different namesakes.

Abstract This article addresses a two‐step approach for term extraction. In the first step on term candidate extraction, a new delimiter‐based approach is proposed to identify features of the delimiters of term candidates rather than those of the term candidates themselves. This delimiter‐based method is much more stable and domain independent than the previous approaches. In the second step on term verification, an algorithm using link analysis is applied to calculate the relevance between term candidates and the sentences from which the terms are extracted. All information is obtained from the working domain corpus without the need for prior domain knowledge. The approach is not targeted at any specific domain and there is no need for extensive training when applying it to new domains. In other words, the method is not domain dependent and it is especially useful for resource‐limited domains. Evaluations of Chinese text in two different domains show quite significant improvements over existing techniques and also verify its efficiency and its relatively domain‐independent nature. The proposed method is also very effective for extracting new terms so that it can serve as an efficient tool for updating domain knowledge, especially for expanding lexicons.

Abstract In this article, the authors address the problem of sentence ranking in summarization. Although most existing summarization approaches are concerned with the information embodied in a particular topic (including a set of documents and an associated query) for sentence ranking, they propose a novel ranking approach that incorporates intertopic information mining. Intertopic information, in contrast to intratopic information, is able to reveal pairwise topic relationships and thus can be considered as the bridge across different topics. In this article, the intertopic information is used for transferring word importance learned from known topics to unknown topics under a learning‐based summarization framework. To mine this information, the authors model the topic relationship by clustering all the words in both known and unknown topics according to various kinds of word conceptual labels, which indicate the roles of the words in the topic. Based on the mined relationships, we develop a probabilistic model using manually generated summaries provided for known topics to predict ranking scores for sentences in unknown topics. A series of experiments have been conducted on the Document Understanding Conference (DUC) 2006 data set. The evaluation results show that intertopic information is indeed effective for sentence ranking and the resultant summarization system performs comparably well to the best‐performing DUC participating systems on the same data set.

Abstract Sentence ranking is the issue of most concern in document summarization today. While traditional feature‐based approaches evaluate sentence significance and rank the sentences relying on the features that are particularly designed to characterize the different aspects of the individual sentences, the newly emerging graph‐based ranking algorithms (such as the PageRank‐like algorithms) recursively compute sentence significance using the global information in a text graph that links sentences together. In general, the existing PageRank‐like algorithms can model well the phenomena that a sentence is important if it is linked by many other important sentences. Or they are capable of modeling the mutual reinforcement among the sentences in the text graph. However, when dealing with multidocument summarization these algorithms often assemble a set of documents into one large file. The document dimension is totally ignored. In this article we present a framework to model the two‐level mutual reinforcement among sentences as well as documents. Under this framework we design and develop a novel ranking algorithm such that the document reinforcement is taken into account in the process of sentence ranking. The convergence issue is examined. We also explore an interesting and important property of the proposed algorithm. When evaluated on the DUC 2005 and 2006 query‐oriented multidocument summarization datasets, significant results are achieved.

Sentence ranking is the issue of most concern in document summarization. Early researchers have presented the mutual reinforcement principle (MR) between sentence and term for simultaneous key phrase and salient sentence extraction in generic single-document summarization. In this work, we extend the MR to the mutual reinforcement chain (MRC) of three different text granularities, i.e., document, sentence and terms. The aim is to provide a general reinforcement framework and a formal mathematical modeling for the MRC. Going one step further, we incorporate the query influence into the MRC to cope with the need for query-oriented multi-document summarization. While the previous summarization approaches often calculate the similarity regardless of the query, we develop a query-sensitive similarity to measure the affinity between the pair of texts. When evaluated on the DUC 2005 dataset, the experimental results suggest that the proposed query-sensitive MRC (Qs-MRC) is a promising approach for summarization.

This paper presents an opinion analysis system based on linguistic knowledge which is acquired from small-scale annotated text and raw topic-relevant Web page. Based on the observation on the annotated opinion corpus, some word-, collocation- and sentence-level linguistic features for opinion analysis are discovered. Supervised and unsupervised learning techniques are developed to learn these features from annotated text and raw relevant Web page, respectively. These features are then incorporated into a classifier based on support vector machine (SVM) to identify opinionated sentences and determine their polarities. Evaluations show that the proposed opinion analysis system, namely OA, achieved promising performance, which shows the effectiveness of linguistic knowledge learning from relevant Web page.

The work presented in this paper is motivated by the practical need for content extraction, and the available data source and evaluation benchmark from the ACE program. The Chinese Entity Detection and Recognition (EDR) task is of particular interest to us. This task presents us several language-independent and language-dependent challenges, e.g. rising from the complication of extraction targets and the problem of word segmentation, etc. In this paper, we propose a novel solution to alleviate the problems special in the task. Mention detection takes advantages of machine learning approaches and character-based models. It manipulates different types of entities being mentioned and different constitution units (i.e. extents and heads) separately. Mentions referring to the same entity are linked together by integrating most-specific-first and closest-first rule based pairwise clustering algorithms. Types of mentions and entities are determined by head-driven classification approaches. The implemented system achieves ACE value of 66.1 when evaluated on the EDR 2005 Chinese corpus, which has been one of the top-tier results. Alternative approaches to mention detection and clustering are also discussed and analyzed.

In distance-based location update schemes with a predefined distance threshold d, an object reports its location to the location server, whenever it is located more than a distance of d away from the location expected of by the server. Adopting a small threshold can keep locations maintained in the location server close to exact object locations, but that incurs high location update costs. In this paper, we address the important issue of finding an optimal distance threshold. Our approach exploits a costfunction that takes into account location update and query processing costs, the two key performance costs, based on which an optimal threshold that minimizes the overall cost is derived. In dynamic environments, costs may vary over time, so a threshold good at one moment could become bad at another. To determine an optimal threshold adaptively, we propose two optimization algorithms, namely, conjectural algorithm and progressive algorithm. Conjectural optimization algorithm " guesses" the current system conditions, based on which it directly determines the most probable optimal value. Progressive optimization algorithm starts with a certain threshold value and adjusts it gradually towards the optimal point. To evaluate our proposed algorithms, various simulation studies are conducted and significant performance gain is observed with our algorithms.

Event-based summarization attempts to select and organize the sentences in a summary with respect to the events or the sub-events that the sentences describe. Each event has its own internal structure, and meanwhile often relates to other events semantically, temporally, spatially, causally or conditionally. In this paper, we define an event as one or more event terms along with the named entities associated, and present a novel approach to derive intra- and inter- event relevance using the information of internal association, semantic relatedness, distributional similarity and named entity clustering. We then apply PageRank ranking algorithm to estimate the significance of an event for inclusion in a summary from the event relevance derived. Experiments on the DUC 2001 test data shows that the relevance of the named entities involved in events achieves better result when their relevance is derived from the event terms they associate. It also reveals that the topic-specific relevance from documents themselves outperforms the semantic relevance from a general purpose knowledge base like Word-Net.

Moving object databases (MODs), the core component of location server to support location-related applications, keep track of the locations of moving objects which submit location update reports to the centralized server. In resource-limited wireless environments, the frequency and conditions for generating location update messages exert a strong impact on system performance in terms of update message cost and object location accuracy, hence the query result precision. Conceptually, moving objects are the sources of the location data while the MOD caches recently reported object locations for query processing. Owing to the inherent imprecision of the cached values, we impose a bounded level of inconsistency for the cached values, realized in the form of a "safe range" for a moving object. The cached value needs not be invalidated so long as the deviation of the object's current location from its reported location is within the safe range. A smaller safe range results in a higher accuracy of the cached value and hence more accurate query result at the expense of higher update cost, and vice versa. Since the size of the safe range is the key to system performance, we derive a system cost model to determine its appropriate value. Furthermore, to cater for highly dynamic environments in which object movement, query access pattern and system workload always change, we propose two adaptive safe range adjustment algorithms. Through extensive simulation experiments, the benefits brought about by our algorithms are evidenced

In this paper, we propose a new strategy with time granularity reasoning for utilizing temporal information in topic tracking. Compared with previous ones, our work has four distinguished characteristics. Firstly, we try to determine a set of topic times for a target topic from the given on-topic stories. It helps to avoid the negative influence from other irrelevant times. Secondly, we take into account time granularity variance when deciding whether a coreference relationship exists between two times. Thirdly, both publication time and times presented in texts are considered. Finally, as time is only one attribute of a topic, we increase the similarity between a story and a target topic only when they are related not only temporally but also semantically. Experiments on two TDT corpora show that our method makes good use of temporal information in news stories.

Temporal information is an important attribute of a topic, and a topic usually exists in a limited period. Therefore, many researchers have explored the utilization of temporal information in topic detection and tracking (TDT). They use either a story's publication time or temporal expressions in text to derive temporal relatedness between two stories or a story and a topic. However, past research neglects the fact that people tend to express a time with different granularities as time lapses. Based on a careful investigation of temporal information in news streams, we propose a new strategy with time granularity reasoning for utilizing temporal information in topic tracking. A set of topic times, which as a whole represent the temporal attribute of a topic, are distinguished from others in the given on-topic stories. The temporal relatedness between a story and a topic is then determined by the highest coreference level between each time in the story and each topic time where the coreference level between a test time and a topic time is inferred from the two times themselves, their granularities, and the time distance between the topic time and the publication time of the story where the test time appears. Furthermore, the similarity value between an incoming story and a topic, that is the likelihood that a story is on-topic, can be adjusted only when the new story is both temporally and semantically related to the target topic. Experiments on two different TDT corpora show that our proposed method could make good use of temporal information in news stories, and it consistently outperforms the baseline centroid algorithm and other algorithms which consider temporal relatedness.

In this research, we focus on tracking topics that originate and evolve from a specific event. Intuitively, a few key elements of a target event, such as date, location, and persons involved, would be enough for making a decision on whether a test story is on-topic. Consequently, a profile-based event tracking method is proposed. We attempt to build an event profile from the given on-topic stories by robust information retrieval technologies. A feature selection metric and a recognized event clause are utilized to determine most (if not all) key semantic elements of the target event. Preliminary experiments on the TDT2 mandarin corpus show that this profile-based event tracking method is promising.

k is the most important parameter in a text categorization system based on the k -nearest neighbor algorithm ( k NN). To classify a new document, the k -nearest documents in the training set are determined first. The prediction of categories for this document can then be made according to the category distribution among the k nearest neighbors. Generally speaking, the class distribution in a training set is not even; some classes may have more samples than others. The system's performance is very sensitive to the choice of the parameter k . And it is very likely that a fixed k value will result in a bias for large categories, and will not make full use of the information in the training set. To deal with these problems, an improved kNN strategy, in which different numbers of nearest neighbors for different categories are used instead of a fixed number across all categories, is proposed in this article. More samples (nearest neighbors) will be used to decide whether a test document should be classified in a category that has more samples in the training set. The numbers of nearest neighbors selected for different categories are adaptive to their sample size in the training set. Experiments on two different datasets show that our methods are less sensitive to the parameter k than the traditional ones, and can properly classify documents belonging to smaller classes with a large k . The strategy is especially applicable and promising for cases where estimating the parameter k via cross-validation is not possible and the class distribution of a training set is skewed.

Abstract This paper describes the design and development of a computer interface for blind and visually‐impaired users, who are native speakers of Cantonese (i.e. a Chinese dialect). Apart from enabling the interface to (1) produce Chinese voice output, (2) convert Chinese characters to Braille codes, (3) facilitate Chinese Braille input, and (4) operate in a Microsoft Chinese Windows environment, the significant aspects of this paper include the following: (1) the description of an integrated architecture, which can be used for other languages; (2) a general bilingual Braille input mechanism; (3) a sentence‐based input method that can be used for contracted‐Braille‐to‐text conversion with an error rate of about 6%, operating at about 700 characters/second using a Pentium II 300 MHz PC; (4) a code‐mixed synthesis module for general bilingual and multilingual applications; (5) the potential to directly adopt the system for use with other ideographic languages (like Japanese and Korean), as well as agglutinating languages like Finnish and Turkish, which have no space between words. Copyright © 2003 John Wiley & Sons, Ltd.

Most existing dynamic load distribution (LD) algorithms assume fairly stable task arrival pattern. With this assumption, single task assignments are adequate to provide reasonably good performance. They are, however, inadequate when tasks arrive in bursts. In this paper, we propose a LD algorithm based on batch task assignments. The algorithm is tailored to systems subject to bursty workload. The key of this algorithm is the dynamic negotiation on the amount of workload to be transferred between a sender–receiver pair. Dynamic negotiations ensure the algorithm's adaptive behavior, thus allow task congestions to be resolved quickly. Consequently, CPU utilization can be increased and average task response time reduced substantially. The dynamic negotiations are conducted by the GR Protocol, which also avoids processor thrashing and state waggling – two undesirable phenomena that commonly exist in LD algorithms. © 1999 John Wiley & Sons, Ltd.

Chinese spell checking is different from its counterparts for Western languages because Chinese words in texts are not separated by spaces. Chinese spell checking in this article refers to how to identify the misuse of characters in text composition. In other words, it is error correction at the word level rather than at the character level. Before Chinese sentences are spell checked, the text is segmented into semantic units. Error detection can then be carried out on the segmented text based on thesaurus and grammar rules. Segmentation is not a trivial process due to ambiguities in the Chinese language and errors in texts. Because it is not practical to define all Chinese words in a dictionary, words not predefined must also be dealt with. The number of word combinations increases exponentially with the length of the sentence. In this article, a Block-of-Combinations (BOC) segmentation method based on frequency of word usage is proposed to reduce the word combinations from exponential growth to linear growth. From experiments carried out on Hong Kong newspapers, BOC can correctly solve 10% more ambiguities than the Maximum Match segmentation method. To make the segmentation more suitable for spell checking, user interaction is also suggested.

We propose a new load distribution (LD) algorithm which is based on anti-tasks and load state vectors. Anti-tasks are composite agents which travel around a distributed system to facilitate the pairing up of task senders and receivers, as well as the collection and dissemination of load information. Time-stamped load information of processing nodes is stored in load state vectors which, when used together with anti-tasks, encourage mutual sharing of load information among processing nodes. Anti-tasks, which make use of load state vectors to decide their travelling paths, are spontaneously directed towards processing nodes having high transient workload, thus allowing their surplus work-load to be relocated quickly.

Polling-based load distribution (LD) algorithms suffer from two weaknesses: (i) load information exchanged during a polling session is confined to the two negotiating nodes only; (ii) as the distributed system grows in size (in terms of the number of constituent nodes), a larger number of polling sessions, and thus a higher amount of network bandwidth consumption and CPU overhead, are needed. We propose a new LD algorithm which is based on anti-tasks and load state vectors. This new algorithm avoids the above weaknesses of polling-based LD algorithms. Anti-tasks are composite agents which travel around a distributed system to facilitate the pairing up of task senders and receivers, as well as the collection and dissemination of load information. Time-stamped load information of processing nodes is stored in load state vectors which, when used together with anti-tasks, encourage mutual sharing of load information among processing nodes. Anti-tasks, which make use of load state vectors to decide their traveling paths, are spontaneously directed towards processing nodes having high transient workload, thus allowing their surplus workload to be relocated quickly. Using simulations, we evaluate the performance of our new algorithm by comparing its performance with a number of well-known polling-based load distribution algorithms. We found that our algorithm provides significant reduction of mean task response time over a large range of system sizes. The cost of achieving this performance gain in terms of CPU overhead and channel bandwidth consumption is generally comparable to the other algorithms we studied. © 1998 John Wiley & Sons, Ltd.

Moment descriptors have been developed as features in pattern recognition since the moment method was first introduced. In this paper, new moment features for Chinese character recognition are proposed. These provide significant improvements in terms of Chinese character recognition, especially for those characters that are very close in shapes.

We propose an adaptive load balancing algorithm for heterogeneous distributed systems. The algorithm intrinsically allows a batch of tasks to be relocated. The key of the algorithm is to transfer a suitable amount of processing demand from senders to receivers. This amount is determined dynamically during sender-receiver negotiations. Factors considered when this amount is determined include processing speeds of different nodes, the current load state of both sender and receiver, and the processing demands of tasks eligible for relocation. Composition of a task batch is modeled as a 0-1 Knapsack problem. We also propose a load state measurement scheme which is designed particularly for heterogeneous systems.

Well designed thesauri can represent semantic/conceptual knowledge so as to reveal relationships among different elements in documents, thus serving as a critical tool in intelligent text retrieval systems and document analysis systems. In this paper, we present a thesaurus system, referred to as TheSys, which can be used as a tool for users to build thesauri according to their own requirements. It is our goal to design a comprehensive thesaurus building tool which can be used in any field of specialty rather than targeting for a particular specialty field. People can use our system to build an electronic thesaurus in any specialty field required for a specific application. We propose a thesaurus model, referred to as the thesaurus frame, which uses weighted links, to represent semantic relationships among concepts and terms. Our approach is to use a set of controlled terms, referred to as semantemes, to build the thesaurus frame. This approach can effectively reduce the size of the thesaurus yet the intelligence of the thesaurus is not compromised.

Abstract Processor thrashing in load distribution refers to the situation when a number of nodes try to negotiate with the same target node simultaneously. The performance of dynamic load‐balancing algorithms can be degraded because processor thrashing can lead to receiver node overdrafting, thus causing congestion at a receiver node and reduction of workload distribution. In the paper we present an adaptive algorithm for resolving processor thrashing in load distribution. The algorithm is based on the integration of three components: (1) a batch task assignment policy, which allows a number of tasks to be transferred as a single batch from a sender to a receiver; (2) a negotiation protocol to obtain mutual agreement between a sender and a receiver on the batch size; and (3) an adaptive symmetrically‐initiated location policy to select a potential transfer partner. Simulations reveal that our algorithm provides a significant performance improvement at high system loads because the algorithm can avoid processor thrashing so that CPU capacity is more fully utilized.