人工知能学会論文誌
検索
OR
閲覧
検索
17 巻 , 6 号
Technical Papers
選択された号の論文の7件中1~7を表示しています
    • |<
    • <
    • 1
    • >
    • >|
一般論文
論文
  • 屋野 武秀, 笹島 宗彦, 河野 恭之
    原稿種別: 研究論文
    専門分野: パターン理解
    17 巻 (2002) 6 号 p. 658-666
    公開日: 2002/08/01
    ジャーナル フリー
    In this paper, we propose BTH, a parsing algorithm which is able to efficiently parse keyword lattice that contains large number of false candidates. In BTH, the grammar is written in template form, and then, it is compiled into a hash table. BTH analyzes the lattice without unfolding to keyword sequences, by propagating acceptable templates among the linked keywords and filtering through the hash table in each keywords. It has a time bound proportional to n2 (where n is the number of keywords in the lattice), although the number of false candidates increases exponentially. Simulation results shows that BTH can parse lattice which contains over 100 billion false candidates within 0.35 sec, with grammar which is corresponding to 2 million of templates, on a notebook-PC(PentiumII 266MHz).
    抄録全体を表示
  • 土屋 誠司, 小島 一秀, 渡部 広一, 河岡 司
    原稿種別: 研究論文
    専門分野: 自然言語
    17 巻 (2002) 6 号 p. 667-675
    公開日: 2002/08/01
    ジャーナル フリー
          When we humans receive uncertain information, we interpret it properly, so we can expand the conversation, and take the proper actions. This is possible because we have “common sense” concerning the basic word concept, which is built up from long time experience storing knowledge of our language. Of the common sense we use in our every day lives we think that there are; common sense concerning quantity such as size, weight, speed, time, or place; common sense concerning sense or feeling such as hot, beautiful, or loud; and moreover common sense concerning emotion such as happy or sad. In order to make computers closer to human beings, we think that the construction of a “Common Sense Judgment System” which deals with these kinds of common sense is necessary.
          When aiming to realize this “Common Sense Judgment System” and trying to make a computer have the same common sense knowledge and judgment ability as human beings, a very important factor is the handling of unknown words. Judgment concerning words which were given to the computer as knowledge before hand, it can refer to that knowledge, and the process will have no problem at all. But when an unknown word, which is not registered as knowledge, is inputted, how to process that word is a very difficult problem.
          In this paper, by using a concept base, which is made from several electric dictionaries; the degree of association, which is done based on the concept base; neural network, putting the closeness of meaning in consideration, we propose a method of unknown word processing, which connects an inputted unknown word to a representing word that is registered in the judgment knowledge base, and we will verify its effectiveness by experiment applied to the emotional judgment subsystem.
    抄録全体を表示
  • 森山 甲一, 沼尾 正行
    原稿種別: 研究論文
    専門分野: 創発システム
    17 巻 (2002) 6 号 p. 676-683
    公開日: 2002/08/15
    ジャーナル フリー
    The authors aim at constructing an agent which learns appropriate actions in a Multi-Agent environment with and without social dilemmas. For this aim, the agent must have nonrationality that makes it give up its own profit when it should do that. Since there are many studies on rational learning that brings more and more profit, it is desirable to utilize them for constructing the agent. Therefore, we use a reward-handling manner that makes internal evaluation from the agent's rewards, and then the agent learns actions by a rational learning method with the internal evaluation. If the agent has only a fixed manner, however, it does not act well in the environment with and without dilemmas. Thus, the authors equip the agent with several reward-handling manners and criteria for selecting an effective one for the environmental situation. In the case of humans, what generates the internal evaluation is usually called emotion. Hence, this study also aims at throwing light on emotional activities of humans from a constructive view.

    In this paper, we divide a Multi-Agent environment into three situations and construct an agent having the reward-handling manners and the criteria. We observe that the agent acts well in all the three Multi-Agent situations composed of homogeneous agents.

    抄録全体を表示
ショートノート
  • 藤井 啓彰, 小島 一秀, 渡部 広一, 河岡 司
    原稿種別: 技術報告
    専門分野: その他
    17 巻 (2002) 6 号 p. 684-689
    公開日: 2002/08/15
    ジャーナル フリー
    In this information-oriented society that we live in, a method of retrieving necessary information is needed. Obtaining necessary information, by information retrieval using the Boolean method, which uses keywords given by the user, narrows corresponding documents down to the thousands or ten thousands, which results in the user not knowing which document to look at first. Therefore, by transferring the user's retrieval query and documents into quantitative values and ranking the requested information in order of relation to the user's demand, it is needed to give the user information that fits the user's demand. As a calculation method to transfer the retrieval query and the documents into quantitative values, the vector space model exists, but this paper shows that the method of ranking information, using the concept-base and the degree of association, is effective.
    抄録全体を表示
論文
  • 梅原 雅之, 岩沼 宏治, 鍋島 英知
    原稿種別: 研究論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 6 号 p. 690-698
    公開日: 2002/08/29
    ジャーナル フリー
    The recognition and extraction of semantic/logical structures in HTML documents are substantially important and difficult tasks for intelligent document processing. In this paper, we show that the alignment technology is an appropriate tool, within a framework of case-based reasoning, for recognizing semantic structures inherently embedded in a series of HTML documents. That is, given a series of HTML documents and a document example of which semantic structures are explicitly indicated by a user, then the alignment can identify semantic structures in the HTML document series, by matching a text-block sequence in each HTML document with the text-block sequence in the example document. Several important properties in text documents, such as continuity, sequentiality of texts, can be treated by the alignment in a quite natural way.

          The alignment technology can significantly improve the capability of the case-based transformation method which transforms a spatial and/or temporal series of HTML documents into machine-readable XML formats. Moreover, the alignment dramatically eases the construction of transformation exmaples. Throughout experimental evaluation for 47 pages of 8 series of HTML documents, we show that the case-based method using the alignment achieved a highly accurate transformation into XML formats.

    抄録全体を表示
  • 佐久間 淳, 小林 重信
    原稿種別: 研究論文
    専門分野: 創発システム
    17 巻 (2002) 6 号 p. 699-707
    公開日: 2002/09/19
    ジャーナル フリー
    We propose a new Real-coded GA(RCGA) using the combination of two crossovers, UNDX-m and EDX. The search region of UNDX-m is biased to the inside area that the population of the RCGA covers. Because of this search bias, the GA using UNDX-m causes stagnation of its search if the cost function has a kind of structure, so called, a ridge structure or a multiple-peak structure. In order to overcome this stagnation, we propose a new crossover EDX, whose search is biased toward extrapolative one. Experimental results show that RCGA with EDX can deal with both ridge-structure function whose dimension reaches more than hundreds and multiple-peak function whose optimum resides at the corner of the search area.
    抄録全体を表示
  • 稲田 政則, 寺野 隆雄
    原稿種別: 研究論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 6 号 p. 708-715
    公開日: 2002/10/01
    ジャーナル フリー
    As Active Mining is a new concept among data mining and/or knowledge discovery in databases communities, in order to validate the effectiveness, it is important to carry out empirical studies using practical data. Based on the concept of Active User Reaction, this paper develops a causal model from liver function test data in a medical domain. To develop the model, we have set a problem to predict the values of ICG (indocyanine green) test from given observation data and experts' background knowledge. We therefore employ a framework of meta-learning and structural equation modeling. In this paper meta-learning means learning about mined results from multiple data-mining techniques. Structural equation modeling enables us to describe flexible models from background knowledge. The construction of the causal model contains two phases: meta-learning and the model building. The meta-learning phase utilizes both the linear regression and the neural network as data mining techniques, then examines the predictability on the given data set. Mining models are n-folded learned from the training data set. Each of the prediction accuracy of the mining models is compared using with the testing data. On the model building phase, we use structural equation modeling to develop a causal model based on results of meta-learning and background knowledge. We again compare the accuracy of the causal model with each of the mining models. Consequently we have developed the causal model, which is comprehensible and have good predictive performance, via the meta-learning phase. Through the empirical study, we have got the conclusion that the framework of meta-learning is effective in data mining in a difficult medical domain.
    抄録全体を表示
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top