The objective of this paper is to give overviews of text mining or textual data mining in Japan from the practical aspects. Text mining is the technology utilized for analyzing large volumes of textual data applying various parameters for purpose of withdrawing useful knowledge and information. The essence of “Mining” is "the discovery of knowledge or information." And target of text mining is to objectively discover and extract knowledge, facts, and meaningful relationships from the text documents. This paper summarizes the related disciplines and application fields which are applied in text mining, and introduces features and application examples of text mining tools.

抄録全体を表示

Because a protein expresses its function through interaction with other substrates, it is vital to create a database of protein interaction. Since the total volume of information on protein interaction is described in terms of thousands of literatures, it is nearly impossible to extract all this information manually. Although extraction systems for interaction information based on the template matching method have already been developed, it is not possible to match all the sentences with interaction information due to the extent of sentence complexity.
We propose a method of extracting sentences with interaction information independent of sentence structure. In a protein-compound complex structure, the interacting residue is near to its partner. The distance between them can be calculated by using the structure data in the PDB database, with a short distance indicating that the sentences associated with them might describe the interaction information. In a free-protein structure, the distance cannot be calculated because the coordinates of the protein's partner are not registered in the structure data. Hence, we use the homology protein structure data, which is complexed with the protein's parter.
The proposed method was applied to seven literatures written about protein-compound complexes and four literatures written about free proteins, obtaining F-measures of 71% and 72%, respectively.

抄録全体を表示

In order to meet the demand to acquire necessary information efficiently from large electronic text, the Question and Answering (QA) technology to show a clear reply automatically to a question asked in the user's natural language has widely attracted attention in recent years. Although the research of QA system in China is later than that in western countries and Japan, it has attracted more and more attention recently. In this paper, we propose a Question-Answering construction, which synthesizes the answer retrieval to the questions asked most frequently based on common knowledge, and the document retrieval concerning sightseeing information. In order to improve reply accuracy, one must consider the synthetic model based on statistic VSM and the shallow semantic analysis, and the domain is limited to sightseeing information. A Chinese QA system about sightseeing based on the proposed method has been built. The result is obtained by evaluation experiments, where high accuracy can be achieved when the results of retrieval were regarded as correct, if the correct answer appeared among those of the top three resemblance degree. The experiments proved the efficiency of our method and it is feasible to develop Question-Answering technology based on this method.

抄録全体を表示

Information extraction (IE) is essential owing to increase of protein related literature contains active site and interaction information. It is important to identify correctly named entities for the accurate information extraction. Named entities in molecular biology fields is described variously and cannot be identified easily by terminology dictionary. This paper proposes a new method for automatic identification of named entities in literatures using GenomeNet. GenomeNet has text search system for a lot of protein related databases which are updated every day. Our method utilizes the search result of GenomeNet as knowledge base. The search result contains the clues for identifying named entities. Focusing on the clues in the search results, three parameters are defined. And we can identify the named entities to figure out if the three parameters meets the identification condition. We applied proposed method for some protein related literatures to compare with other method. The evaluated value is improved substantially and efffectness of proposed method is clear.

抄録全体を表示

This paper proposes the method for characteristic analysis on research institutes. Basic premise under this research is that each research institute has its web pages on its research themes. Our concerns are how to find out from their text expression that 1) what are major research themes among them? 2) what are current trends on research focus? 3) are there distinguished research themes? 4) Are there similar research institutes? and 5) are there complementary research institutes? Regarding research theme as term and research institute as document, this paper derives TF, DF and TF/IDF values for the method. Then TF implies major research themes and TF/IDF proposes the focused research themes which are not popular in other research institutes. It is also shown that the inner products between TF/IDF values vectors imply the similarity and complementary between research institutes. Experimental discussion for industrial engineering of Japanese Universities is done for validating method.

抄録全体を表示

Information technology is increasingly important in recent years for the development of our society. IT has brought many changes to everything in our society with incredible speed. Hence, when we investigate R & D themes or plan business strategies in IT, we must understand overall situation around the target technology area besides technology itself. Especially it is crucial to understand overall situation as time series to know what will happen in the near future in the target area. For this purpose, we developed a method to generate Multiple-phased trend maps automatically based on the Internet content. Furthermore, we introduced quantitative indicators to analyze near future possible changes. According to the evaluation of this method we got successful and interesting results.

抄録全体を表示

N-gram indexing is the most popular algorithm for the full text search system where each index consists of serial N characters. Especially the full text search for Japanese text usually has the 2-gram characters index as base. The additional higher-gram index is expected to improve the performance. This paper presents the entropy-based method for mining additional indexing terms from DB in order to reduce the waste of AND operation for 2gram.

抄録全体を表示

This paper discusses the architecture for high speed learning of Neural Network (NN) using Genetic Algorithm (GA). The proposed architecture prevents local minimum by using the GA characteristic of holding several individual populations for a population-based search and achieves high speed processing adopting dedicated hardware. To keep general purpose equal software processing, the proposed architecture can be flexible genetic operations on GA and is introduced both Sigmoid function and Heaviside function on NN. Furthermore, the proposed architecture is not optimized only the pipeline at evaluation phase on NN, but also optimized hierarchic pipelines on the whole at evolutionary phase. We have done the simulation, verification and logic synthesis using library of 0.35μm CMOS standard cell. Simulation results evaluating the proposed architecture show to achieve 22 times speed on average compared with software processing.

抄録全体を表示

We demonstrated a carrier-envelope phase (CEP) stabilized chirped-pulse amplifier system. This amplifier system is composed of grating based pulse-stretcher and compressor, a regenerative amplifier and a multi-pass amplifier. We employed a new pulse-pick-up method to select CEP stabilized seed pulses. This pulse selection method is different from established practice which is based on pulse train timing, but is based on CEP of seed pulse. We measured amplitude-to-phase noise conversion coefficient of microstructure fiber and evaluated the additional out-of-loop error of carrier-envelope offset (CEO) control. We also investigated the effect of beam pointing on the measured fringe shift in self-referencing spectral interference method.

抄録全体を表示

We propose a parallel transmission to improve transmission speed using IEEE 802.11b wireless LAN standard. Parallel transmission is a method for transmitting data over two lines (frequency channel) simultaneously using one wireless station. Furthermore, we suppressed the frequency bandwidth by narrowing the frequency channel interval of each line. We applied a synchronous transmission, which controls the packet transmission timing of each line to reduce adjacent channel interference. Computer simulation results clarified the bit error rate with a synchronization gap quantity between lines for various numbers of lines and frequency channel intervals. We also examined SNR-BER performance with the number of lines and the frequency channel interval on synchronous transmission. Then, we confirmed that a parallel transmission improves transmission speed by selecting the number of lines and the frequency channel interval that are suitable for a required transmission speed and communication environment.

抄録全体を表示

Analysis of skin inner tissue in three-dimensions has great value in evaluating cosmetics. In this paper we developed a method of automatically extracting the interface of skin inner tissue from volumetric confocal microscopic images with image segmentation algorithm, as well as reconstructing a 3D visual model of skin inner tissues interface. 3D Active contour model is an image segmentation algorithm and is used extensively in computer vision and image processing applications. We improved 3D active contour model by modifying the energy function of 3D active contour model and proposed an method for setting intial surface of it automatically. Furthermore we use nonlinear programming for finding numerical solution of the energy function. The proposed technique has successfully been applied to extract dermo-epidermal surface and upper surface of epidermis.

抄録全体を表示

Recently, biologically inspired robots have been developed to acquire the capacity for directing visual attention to salient stimulus generated from the audiovisual environment. On purpose to realize this behavior, a general method is to calculate saliency maps to represent how much the external information attracts the robot's visual attention, where the audiovisual information and robot's motion status should be involved. In this paper, we represent a visual attention model where three modalities, that is, audio information, visual information and robot's motor status are considered, while the previous researches have not considered all of them. Firstly, we introduce a 2-D density map, on which the value denotes how much the robot pays attention to each spatial location. Then we model the attention density using a Bayesian network where the robot's motion statuses are involved. Secondly, the information from both of audio and visual modalities is integrated with the attention density map in integrate-fire neurons. The robot can direct its attention to the locations where the integrate-fire neurons are fired. Finally, the visual attention model is applied to make the robot select the visual information from the environment, and react to the content selected. Experimental results show that it is possible for robots to acquire the visual information related to their behaviors by using the attention model considering motion statuses. The robot can select its behaviors to adapt to the dynamic environment as well as to switch to another task according to the recognition results of visual attention.

抄録全体を表示

Nonlinear filters have been applied in image restoration. In nonlinear filters, a neural filter was proposed for reducing mixed Gaussian and impulse noises. However, in the case of mixed Gaussian and high probability impulse noises, the performance of neural filter is significantly reduced. In this paper, we propose a filter for removing mixed impulse and Gaussian noises using Trimmed filter and neural filters. Carrying out the simulation, we illustrate the peak signal to noise ratio of the proposed filter, and show that it is effective in removing mixed impulse and Gaussian noises.

抄録全体を表示

This paper is concerned with regulator problem for SISO system using a PID controller. The controlled object is SISO system which is controllable and observable. Our purpose is to analyze stability of PID control system and to determine PID parameters so that the closed system is asymptotically stable. The main idea is based on the minimum phase property and Lyapunov's stability theorem. Quasi pole placement method is applied to stabilize asymptotically zero dynamics whose poles depend on PID parameters. The effectiveness of the proposed method is confirmed by simulation results for various plants.

抄録全体を表示

The spectral distributions of colors are estimated using a sensor system composed of 7 optical filters and a monochromatic CCD camera. The conversion coefficients for the camera outputs are determined by multiple regression analysis. The estimation ability is examined using 20 color patches and 12 natural objects. Then, the multi-band imaging technique is demonstrated to show the highly accurate detection ability. One of the experiments is the detection of human face regions. The detection of human face is sometimes difficult because there are objects that colors are very near to human faces in the image. The spectral distribution data can be used to distinguish human faces and near color objects. The other experiment is the detection of chromaticity vales of electronic displays. The spectral distributions of CRT displays are estimated and the chromaticity values are calculated by the spectral distributions. The errors of chromaticity values were very small compared to conventional RGB camera systems in a wide color space.

抄録全体を表示

This paper proposes a segmentation method of SAR (Synthetic Aperture Radar) images which uses a SOM(Self-Organizing Map). SAR images are obtained by observation using microwave sensor. They are segmented into the drift ice (thick, thin), and sea regions manually, and then features are extracted from partitioned data. However they are not necessarily effective for neural network learning because they can include incorrectly segmented data. Therefore, in particular, a multi-step SOM is used as a learning method to improve reliability of teacher data, and carries out classification. This process enable us to fix all mistook data and segment the SAR data using just data. The validity of this method was demonstrated by computer simulations using the actual SAR images.

抄録全体を表示

Applications of independent component analysis (ICA) to feature extraction have been a topic of research interest. However, the effectiveness of pattern features extracted by conventional ICA algorithms greatly depends on datasets in general. As one of the reasons, we have pointed out that conventional ICA features are obtained by increasing only their independence even if class information is available. In this paper, we propose a supervised learning approach to ICA to extract useful and robust features. The proposed method consists of several modules, each of which is responsible for extracting features for each class and identifying the class labels using the k nearest neighbor classifier. All the module outputs are combined to identify final results based on a majority rule. We evaluate the performance of the proposed method in several recognition tasks. From these results, we confirm the effectiveness of the recognition method using independent components for each class.

抄録全体を表示

Particle Swarm Optimization (PSO) for mixed integer programming problems is proposed. PSO is mainly a method to find a global or quasi-minimum for a nonlinear and non-convex optimization problem, and few researches about optimization problems with discrete decision variables have been studied. In this paper, we present the treatment of discrete decision variables. To treat discrete decision variables as a penalty function, it is possible to treat all decision variables as a continuous decision variable. As a result, the penalty parameter for the penalty function is needed. In this paper, we also present how to determine the penalty parameter for the penalty function. Through mathematical and structural optimization problems, we examine the validity of PSO for the mixed decision variables.

抄録全体を表示

This paper deals with the stability analysis of PSO in order to obtain an understanding about how it searches a globally optimal solution and strategies about how to tune its parameters. While two types of generalized reduced models of PSO are proposed to analyze the dynamics of PSO, the stability analysis is carried out on the basis of both the eigenvalue analysis and the bounded input bounded output stability.

抄録全体を表示

Wrong:(3) Y. Ishida and N. adachi
Right:(3) Y. Ishida and N. Adachi

抄録全体を表示

Wrong:1) Math. And Computer Science Department, Kuwait University
Right:1) Math. and Computer Science Department, Kuwait University

抄録全体を表示