人工知能学会論文誌
Online ISSN : 1346-8030
Print ISSN : 1346-0714
ISSN-L : 1346-0714
23 巻, 6 号
選択された号の論文の17件中1~17を表示しています
論文特集:システム開発論文
論文
  • Kazuhisa Miwa
    2008 年 23 巻 6 号 p. 374-383
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    Problem solving is understood as a process through which states of problem solving are transferred from the initial state to the goal state by applying adequate operators. Within this framework, knowledge and strategies are given as operators for the search. One of the most important points of researchers' interest in the domain of problem solving is to explain the performance of problem solving behavior based on the knowledge and strategies that the problem solver has. We call the interplay between problem solvers' knowledge/strategies and their behavior the causal relation between mental operations and behavior. It is crucially important, we believe, for novice learners in this domain to understand the causal relation between mental operations and behavior. Based on this insight, we have constructed a learning system in which learners can control mental operations of a computational agent that solves a task, such as knowledge, heuristics, and cognitive capacity, and can observe its behavior. We also introduce this system to a university class, and discuss which findings were discovered by the participants.
  • 竹内 広宜, 那須川 哲哉, 渡辺 日出雄
    2008 年 23 巻 6 号 p. 384-391
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    Recently it has become feasible to transcribe textual records from telephone conversations at call centers by using automatic speech recognition. In this research, we extended a text mining system for call summary records and constructed a conversation mining system for the business-oriented conversations at the call center. To acquire useful business insights from the conversational data through the text mining system, it is critical to identify appropriate textual segments and expressions as the viewpoints to focus on. In the analysis of call summary data using a text mining system, some experts defined the viewpoints for the analysis by looking at some sample records and by preparing the dictionaries based on frequent keywords in the sample dataset. However with conversations it is difficult to identify such viewpoints manually and in advance because the target data consists of complete transcripts that are often lengthy and redundant. In this research, we defined a model of the business-oriented conversations and proposed a mining method to identify segments that have impacts on the outcomes of the conversations and can then extract useful expressions in each of these identified segments. In the experiment, we processed the real datasets from a car rental service center and constructed a mining system. With this system, we show the effectiveness of the method based on the defined conversation model.
  • 砂山 渡, 川口 俊明
    2008 年 23 巻 6 号 p. 392-401
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    In recent years, as the Internet has grown, electronic reports have come to be used in educational organizations such as universities. Though reports written by hand must be evaluated by hand except for stereotype descriptions or numerical answers, electronic reports can be rated by computer.
    There are two major criteria in rating reports, correctness and distinctiveness. Correctness is rated by absolute criteria and distinctiveness is rated by relative criteria. Relative evaluation is difficult because raters should memorize all contents of submitted reports to provide objective rates. In addition, electronic data are easily copied or exchanged by students.
    This paper presents a report evaluation support system with which raters can compare each report and give objective rates for distinctiveness. This system evaluates each report by objective similarity criteria and visualizes them in a two-dimensional interface as the calculated distinctiveness order. Experimental results show the system is valid and effective for estimating associations between reports.
  • 糠野 亜紀, 新谷 公朗, 勝木 琢也, 木原 真哉, 上田 真梨, 金田 重郎, 芳賀 博英
    2008 年 23 巻 6 号 p. 402-411
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    Children learn to fit into society through living in a group, and it's greatly influenced by their friend relations. Although preschool teachers need to observe them to assist in the growth of children's social progress and support the development each child's personality, only experienced teachers can watch over children while providing high-quality guidance. To resolve the problem, this paper proposes a mathematical and objective method that assists teachers with observation. It uses numerical data of activity level recorded by pedometers, and we make tree diagram called dendrogram based on hierarchical clustering with recorded activity level. Also, we calculate children's ``breadth'' and ``depth'' of friend relations by using more than one dendrogram. When we record children's activity level in a certain kindergarten for two months and evaluated the proposed method, the results usually coincide with remarks of teachers about the children.
  • 田中 克明, 堀 浩一, 山本 真人
    2008 年 23 巻 6 号 p. 412-423
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    The flood of information on the Internet makes a person who surf it without some strong intention strayed into it. One of the ways to control the balance between a person and the flood is a recommender system by computer, and many web sites use it. These systems work on a web site for the same kind of items. However the field of personal activity is not limited to handle the same kind of thing and a web site, but also offline area in the real world. To handle personal offline activities, LifeLog is proposed as method to record it, but the main purpose of LifeLog is recording a personal history. How to use a history has still been studied stage.
    The authors have developed a recommender system that captures personal context from history of personal online and offline activities, treats information on web sites as a large set of context, and finds out and extends overlap of them, then recommends information located there. The aim of the system is that a person can enjoy waves of information again.
    The system worked as a part of My-life Assist Service. It was a service for mobile phones provided by NTT DoCoMo, Inc. as a field experiment from Dec. 2007 to Feb. 2008.
  • 人工物の機能的知識の外化・共有・共創支援基盤の実現
    高藤 淳, 來村 徳信, 溝口 理一郎
    2008 年 23 巻 6 号 p. 424-436
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    We propose new knowledge management system, which is based on ontology engineering and XML technology, for manufacturing field. The purpose of this system is to support to externalize implicit functional knowledge of a product that a designer creates, to share the knowledge among those who are involved in the product development, and utilize the knowledge for any designer to collaboratively create or invent new products. It is true, however, there are not a few superior design tool such as CAD/CAE software, simulation tools, etc. Though such tools are indispensable to complete a product, they are insufficient to reveal design rational of a designer that indicates the reason why he/she adopted the function, the structure, materials, etc. Ontology engineering contributes to clarify design rational of any designer by describing a function decomposition tree (hereinafter FDT), which has the specific feature of separating function and way, that is a kind of tree to show functional structure based on device ontology and functional ontology. While the functional modeling framework is independent to any data representation infrastructure, XML has a synergy with the framework because of being able to handle semantics of the framework such as subject, object, function, etc. In addition to the semantic view point, XML also has flexibility to compound a document from multiple information fragments, so that it allows FDT to adapt to any application of manufacturing tasks. From those view point, we created OntoGear which is the foundation system that supports a designer or developer to author, share, and use FDT by using xfy technology that JustSystems, Corp. has developed as an XML application development framework.
  • 船越 孝太郎, 中野 幹生, 長谷川 雄二, 辻野 広司
    2008 年 23 巻 6 号 p. 437-446
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    Natural language interfaces are expected to come into practical use in many situations. It is, however, not practical to expect to achieve a universal interface because language use is so diverse. To that end, not only advancements in speech and language technologies but also well-designed development frameworks are required so that developers can build domain-specific interfaces rapidly and easily. This paper proposes KNOLU, a framework for building natural language interfaces of a broad range of applications. Developers using this framework can easily build an interface capable of understanding subsets of natural language expressions just by providing an ontology (a concept hierarchy with semantic frames and a lexicon), an onomasticon (a set of instances and their names) and API functions that provide procedural knowledge required to connect the interface to a target application. To develop an interface using KNOLU, first developers define a concept hierarchy for a target domain. Then they provide other declarative and procedural knowledge components with these knowledge components asscicated to the hierarchy. This developmental flow affords an unobstructed view both for development and maintanance. KNOLU uses an existing general-purpose parser and requires neither grammar rules nor expression patterns. It does not require rules to generate semantic interpretations from parsing results, either. Therefore, developers can build an interface without deep knowledge and experience of natural language processing. We applied KNOLU to two applications and confirmed the effectiveness.
  • 谷岡 稔真, 江頭 広幸, 高田 真由美, 岡崎 泰久, 渡辺 健次, 近藤 弘樹
    2008 年 23 巻 6 号 p. 447-456
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    We have designed and implemented a PC operation support system for a physically disabled person with a speech impediment via voice. Voice operation is an effective method for a physically disabled person with involuntary movement of the limbs and the head. We have applied a commercial speech recognition engine to develop our system for practical purposes. Adoption of a commercial engine reduces development cost and will contribute to make our system useful to another speech impediment people. We have customized commercial speech recognition engine so that it can recognize the utterance of a person with a speech impediment. We have restricted the words that the recognition engine recognizes and separated a target words from similar words in pronunciation to avoid misrecognition. Huge number of words registered in commercial speech recognition engines cause frequent misrecognition for speech impediments' utterance, because their utterance is not clear and unstable. We have solved this problem by narrowing the choice of input down in a small number and also by registering their ambiguous pronunciations in addition to the original ones. To realize all character inputs and all PC operation with a small number of words, we have designed multiple input modes with categorized dictionaries and have introduced two-step input in each mode except numeral input to enable correct operation with small number of words. The system we have developed is in practical level. The first author of this paper is physically disabled with a speech impediment. He has been able not only character input into PC but also to operate Windows system smoothly by using this system. He uses this system in his daily life. This paper is written by him with this system. At present, the speech recognition is customized to him. It is, however, possible to customize for other users by changing words and registering new pronunciation according to each user's utterance.
  • Masaki Murata, Akira Sasaki, Toshiyuki Kanamaru, Tamotsu Shirado, Hito ...
    2008 年 23 巻 6 号 p. 457-462
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    We introduce software that recognizes, extracts, and displays expressions concerning atomic and molecular data from academic papers in the electronic form. This software includes a toolbar application that can be installed in Internet Explorer (IE). This toolbar can be used by scientific readers and researchers to highlight, color-code, and collect important expressions more easily. Those expressions include atomic and molecular symbols (e.g., Xe+ and H2O) and electron configurations(e.g., 4d95s25p) from the atomic and molecular data of a large number of academic papers. We confirmed by experiments that the software could find important expressions with high precision (0.8-1.0). This software is also useful for compiling databases of atomic and molecular data, which is important for plasma simulations, because the simulations critically depend on atomic and molecular data, including the energy levels and collisional and radiative rate coefficients.
  • 牧野 恭子, 早川 ルミ, 寺井 公一, 深津 博
    2008 年 23 巻 6 号 p. 463-472
    発行日: 2008年
    公開日: 2008/07/23
    ジャーナル フリー
    In this paper, we will introduce a system which supports creating diagnostic reports. Diagnostic reports are documents by doctors of radiology describing the existence and nonexistence of abnormalities from the inspection images, such as CT and MRI, and summarize a patient's state and disease. Our system indicates insufficiencies in these reports created by younger doctors, by using knowledge processing based on a medical knowledge dictionary. These indications are not only clerical errors, but the system also analyzes the purpose of the inspection and determines whether a comparison with a former inspection is required, or whether there is any shortage in description. We verified our system by using actual data of 2,233 report pairs, a pair comprised of a report written by a younger doctor and a check result of the report by an experienced doctor. The results of the verification showed that the rules of string analysis for detecting clerical errors and sentence wordiness obtained a recall of over 90% and a precision of over 75%. Moreover, the rules based on a medical knowledge dictionary for detecting the lack of required comparison with a former inspection and the shortage in description for the inspection purpose obtained a recall of over 70%. From these results, we confirmed that our system contributes to the quality improvement of diagnostic reports. We expect that our system can comprehensively support diagnostic documentations by cooperating with the interface which refers to inspection images or past reports.
  • 近藤 恵一, 森田 武史, 和泉 憲明, 橋田 浩一, 山口 高平
    2008 年 23 巻 6 号 p. 473-484
    発行日: 2008年
    公開日: 2008/08/12
    ジャーナル フリー
    In this paper, we propose a supporting method for developments of Web applications in order to support communication between users and developers. Our method combines three-layer architecture modeling of Web application and knowledge sharing technologies based on ontologies. The ontologies, which are used in the method and called enterprise application ontology (EAO), enable users and developers to form common understanding about system specification. EAO covers three aspects (legal aspect, business process, and software design) of business which is a candidate of development of a Web application. EAO includes three ontologies constructed from each of the three aspects. Concepts and structure of EAO are applied on page transition models, which are called Web Process Architecture, of a Web application. As a case study, we applied our method for a development of a ledger accounting system in a local government. And we confirmed that our method could support the development on the business analysis phase.
一般論文
原著論文
  • 内田 誠, 白山 晋
    2008 年 23 巻 6 号 p. 485-493
    発行日: 2008年
    公開日: 2008/08/26
    ジャーナル フリー
    We investigate a factor of the `network effect' that affects on communication service markets by a multi-agent based simulation approach. The network effect is one of a market characteristic, whereby the benefit of a service or a product increase with use. So far, the network effect has been studied in terms of macroscopic metrics, and interaction patterns of consumers in the market were often ignored. To investigate an infulence of structures of the interaction patterns, we propose a multi-agent based model for a communication serivce market, in which embedded complex network structures are considered as an interaction pattern of agents. Using several complex network models as the interaction patterns, we study the dynamics of a market in which two providers are competing. By a series of simulations, we show that the structural properties of the complex networks, such as the clustering coefficient and degree correlations, are the major factors of the network effect. We also discuss an adequate model of the interaction pattern for reproducing the market dynamics in the real world by performing simulations exploiting with a real data of social network.
  • 松井 正一, 山田 誠二
    2008 年 23 巻 6 号 p. 494-504
    発行日: 2008年
    公開日: 2008/09/17
    ジャーナル フリー
    Hierarchical menus are widely used as a standard user interface in modern applications that use GUIs. The performance of the menu depends on many factors: structure, layout, colors and so on. There has been extensive research on novel menus, but there has been little work on improving performance by optimizing the menu's structure. This paper proposes an algorithm based on the genetic algorithm for optimizing the performance of menus. The algorithm aims to minimize the average selection time of menu items by considering the user's pointer movement and search/decision time. We will show the results on static hierarchical menus of a cellular phone and a PDA as examples where small screen and limited input device are assumed. We will show the effectiveness of the algorithm by using wide variety of usage patterns.
  • 高村 大也, 奥村 学
    2008 年 23 巻 6 号 p. 505-513
    発行日: 2008年
    公開日: 2008/09/17
    ジャーナル フリー
    We discuss text summarization in terms of maximum coverage problem and its variant. To solve the optimization problem, we applied some decoding algorithms including the ones never used in this summarization formulation, such as a greedy algorithm with performance guarantee, a randomized algorithm, and a branch-and-bound method. We conduct comparative experiments. On the basis of the experimental results, we also augment the summarization model so that it takes into account the relevance to the document cluster. Through experiments, we showed that the augmented model is at least comparable to the best-performing method of DUC'04.
  • 尾崎 知伸, 大川 剛直
    2008 年 23 巻 6 号 p. 514-525
    発行日: 2008年
    公開日: 2008/09/17
    ジャーナル フリー
    Recently, pattern mining in structured domain, such as sequences, trees and graphs, is becoming increasingly abundant and several algorithms for especially frequent pattern mining have been developed. On the other hand, the research area of correlation mining in transaction databases, that extracts the underlying dependency among objects, attracts a big attention and extensive studies have been reported. Although we can easily expect to get a more powerful tool for structured data by introducing correlation mining, the most of current research on correlation mining are designed for transaction databases and little attention is paid to mining correlations from structured data. Motivated by these backgrounds, in this paper, we bring the concept of hyperclique pattern in transaction databases into the graph mining and consider the discovery of sets of highly-correlated subgraphs in graph-structured databases. To achieve this objective, a novel algorithm named HSG is proposed. By considering the generality ordering on sets of subgraphs, HSG employs the depth-first/breadth-first search strategy with powerful pruning techniques based on both of the anti-monotone property of support value and the upper bound of h-confidence measure. Experiments with artificial and real world datasets were conducted to assess the effectiveness of the proposed algorithm. The results of experiments show that HSG succeeds in discovering sets of highly-correlated subgraphs within reasonable computation time.
  • 中山 功一, 松井 博和, 井ノ上 直己
    2008 年 23 巻 6 号 p. 526-539
    発行日: 2008年
    公開日: 2008/10/07
    ジャーナル フリー
    MGG (Minimal Generation Gap) is one of popular generation alternation models for Genetic Algorithms (GAs). The conventional MGG is effective for single population GAs, but not for multi-population GAs. This paper proposes ``MGG with global selection (MGGGS)'' that is designed for multi-population GAs. In MGGGS, the selection operation is carried out through the whole population, while the crossover operation is restricted in sub-populations.
    Experiments are carried out to analyze the characteristics of MGGGS with Dynamically Separating Genetic Algorithm (DS-GA). In DS-GA the sub-populations are reconstructed during the evolution, which is suitable for MGGGS. The experimental results show that MGGGS outperforms the conventional MGG especially for multimodal functions, since sub-populations explore various areas by MGGGS.
  • 石川 勉
    2008 年 23 巻 6 号 p. 540-552
    発行日: 2008年
    公開日: 2008/10/15
    ジャーナル フリー
    We propose a knowledge representation scheme (called WPL) and an inference method for the scheme. In WPL, both simple sentence and complex sentence are represented in one atomic formula. Subordinate clauses in a complex sentence are embedded into the formula forming the main clause. WPL is an extended order-sorted logic that can deal with structured sort symbols consisting of multiple ordinary words like noun phrases. Each word in a sort symbol can represent a general concept or a particular object. If it is the latter, each word stands for a variable or constant having itself as a sort symbol. It may also be a proper noun or variable itself. The inference processes for WPL is executed based on the resolution principle, semantically interpreting the sort symbols word by word. We extend the inference rules proposed by Beierle et al. in order to deal with complex sort symbols. This paper also describes an application scheme of the proposed inference rules and an algorithm for judging the subsort relation between complex sort symbols.
feedback
Top