Catálogo de publicaciones - libros

We describe the research on fusion of Web and TV broadcasting contents conducted by Kyoto University 21 COE program and NICT communication/broadcasting contents fusion project. Despite much talk about the fusion of broadcasting and the Internet, no technology has been established for fusing web and TV program content. We proposed several ways to acquire information from diverse information sources of different media types, especially from Web and TV broadcasting. A notable difference between Web contents and TV program contents is that the former is a document-based information media and the latter is a time-based continuous information media, which leads to the difference of information accessing methods. Conventional “Web browsing” is an active manner of accessing information. On the other hand, conventional “TV watching” is a passive way of accessing information. In order to search, integrate and view the information of Web and TV, we explored (1) media conversion between Web and TV contents, (2) watching TV with live chats, (3) dynamic TV-content augmentation by Web, and (4) searching for TV contents with Web.

Mining from graph-structured data has its root in concept formation. Recent advancement of data mining techniques has broadened its applicability. Graph mining faces with subgraph isomorphism which is known to be NP-complete. Two contrasting approaches of our work on extracting frequent subgraphs are revisited, one using complete search (AGM) and the other using heuristic search (GBI). Both use canonical labelling to deal with subgraph isomorphism. AGM represents a graph by its adjacency matrix and employs an Apriori-like bottom up search algorithm using anti-monotonicity of frequency. It can handle both connected and dis-connected graphs, and has been extended to handle a tree data and a sequential data by incorporating a different bias to each in joining operators. It has also been extended to incorporate taxonomy in labels to extract generalized subgraphs. GBI employs a notion of chunking, which recursively chunks two adjoining nodes, thus generating fairly large subgraphs at an early stage of search. The recent improved version extends it to employ pseudo-chunking which is called chunkingless chunking, enabling to extract overlapping subgraphs. It can impose two kinds of constraints to accelerate search, one to include one or more of the designated subgraphs and the other to exclude all of the designated sub-graphs. It has been extended to extract paths and trees from a graph data by placing a restriction on pseudo-chunking operations. GBI can further be used as a feature constructor in decision tree building. The paper explains how both GBI and AGM with their extended versions can be applied to solve various data mining problems which are difficult to solve by other methods.

In this paper, the concept of Relative Frequency Ratio (RFR) is presented to evaluate the strength of collocation. Based on RFR, a WSD Model RFR-SUM is put forward to disambiguate polysemous Chinese word sense. It selects 9 frequently used polysemous words as examples, and achieves the average precision up to 92:50% in open test. It has compared the model with Naïve Bayesian Model and Maximum Entropy Model. The results show that the precision by RFR-SUM Model is 5:95% and 4:48% higher than that of Naïve Bayesian Model and Max- imum Entropy Model respectively. It also tries to prune RFR lists. The results reveal that leaving only 5% important collocation information can keep almost the same precision. At the same time, the speed is 20 times higher.

In the automated text classification, is often considered as the default term weighting scheme and has been widely reported in literature. However, does not directly reflect terms’ category membership. Inspired by the analysis of various feature selection methods, we propose a simple probability based term weighting scheme which directly utilizes two critical information ratios, i.e. relevance indicators. These relevance indicators are nicely supported by probability estimates which embody the category membership. Our experimental study based on two data sets, including Reuters-21578, demonstrates that the proposed probability based term weighting scheme outperforms significantly using Bayesian classifier and Support Vector Machines (SVM).

Healthcare information support (HIS) is essential in managing, gathering, and disseminating information for healthcare decision support through the Internet. To support HIS, text classification (TC) is a key kernel. Upon receiving a text of healthcare need (e.g. symptom description from patients) or healthcare information (e.g. information from medical literature and news), a text classifier may determine its corresponding categories (e.g. diseases), and hence subsequent HIS tasks (e.g. online healthcare consultancy and information recommendation) may be conducted. The key challenge lies on TC, which aims to classify most texts into suitable categories (i.e. recall is very high), while at the same time, avoid misclassifications of most texts (precision is very high). High-quality TC is particularly essential, since healthcare is a domain where an error may incur higher cost and/or serious problems. Unfortunately, high-quality TC was seldom achieved in previous studies. In the paper, we present a case study in which a high-quality classifier is built to support HIS in Chinese disease-related information, including the cause, symptom, curing, side-effect, and prevention of cancer. The results show that, without relying on domain knowledge and complicated processing, cancer information may be classified into suitable categories, with a controlled amount of confirmations.

Nurse scheduling is a complex scheduling problem and involves generating a schedule for each nurse that consists of shift duties and days off within a short-term planning period. The problem involves multiple conflicting objectives such as satisfying demand coverage requirements and maximizing nurses’ preferences subject to a variety of constraints imposed by legal regulations, personnel policies and many other hospital-specific requirements. The inherent nature of the nurse scheduling problem (NSP) bears vagueness of information on target values of hospital objectives and on personal preferences. Also, the ambiguity of the constraints is some source of uncertainty that needs to be treated in providing a high quality schedule. Taking these facts into account, this paper presents the application of Fuzzy Set Theory (FST) within the context of NSP and proposes a fuzzy goal programming model. To explore the viability of the proposed model, computational experiments are presented on a real world case problem.

Sensor networks are often deployed in unattended environments, thus leaving these networks vulnerable to false data injection attacks. False data injection attacks will not only cause false alarms that waste real world response efforts, but also drain the finite amount of energy in a battery powered network. The key inheritance based filtering scheme can detect a false report at the very next node of the compromised node that injected the false report before it consumes a significant amount of energy. The choice of a security threshold value in this scheme represents a trade off between security and overhead. In this paper, we propose a fuzzy adaptive threshold determining method for the key inheritance based filtering scheme. The fuzzy rule based system is exploited to determine the security threshold value by considering the average energy level of all the nodes along the path from the base station to a cluster, the number of nodes in that cluster, and the number of compromised nodes. We also introduce a modified version of this scheme to reduce the overhead for changing the threshold value. The proposed method can conserve energy, while it provides sufficient resilience.

In this paper, we present a new approach for evaluating students’ answerscripts based on the similarity measure between interval-valued fuzzy sets. The marks awarded to the answers in the students’ answerscripts are represented by interval-valued fuzzy sets, where each element in the universe of discourse belonging to an interval-valued fuzzy set is represented by an interval between zero and one. An index of optimism determined by the evaluator is used to indicate the degree of optimism of the evaluator, where ∈ [0, 1]. The proposed approach using interval-valued fuzzy sets for evaluating students’ answerscripts can evaluate students’ answerscripts in a more flexible and more intelligent manner.

The proposed system relates to an interactive scoring system for learning a language, in which a means such as a web camera is used to capture the learners lip movements and then a score is given by making a comparison with images stored in the database. The images stored in the database are those previously recorded by a teacher. By means of the scoring system, the learner can identify and rectify pronunciation problems concerning the lips and tongue. The system also records sounds as well as images from the student. The proposed system processes this data with multimedia processing techniques. With regard to the interactive perspective, a user-friendly visual interface was constructed to help learners use the system. The learners can choose the words they want to practice by capturing their lip image sequences and speech. The lip region image sequences are extracted automatically as visual feature parameters. Combining the visual and voice parameters, the proposed system calculates the similarity between a learners and a teachers pronunciation. An evaluation score is suggested by the proposed system through the previous similarity computation. By this learning process, learners can see the corresponding lip movement of both themselves and a teacher, and correct their pronunciation accordingly. The learners can use the proposed system to practice their pronunciation as many times as they like, without troubling the human teacher, and thus they are able to take more control of improving their pronunciation.

The problem of matching feature points in multiple images is difficult to solve when their appearance changes due to illumination variance, either by lighting or object motion. In this paper we tackle this ill-posed problem by using the difference of local phase which is known to be stable to a certain extent even under illumination variances. In order to realize a precise matching, we basically compute the local phase by convolutions with Gabor filters which we design in multi scales. We then evaluate the stability of local phase against lighting changes. Through experiments using both CG and real images that are with illumination variance, we show the relevancy of our theoretical investigations.