人工知能学会論文誌
検索
OR
閲覧
検索
17 巻 , 3 号
Technical Papers / Special Issue: Web Intelligence
選択された号の論文の19件中1~19を表示しています
    • |<
    • <
    • 1
    • >
    • >|
一般論文
論文
  • 目良 和也, 市村 匠, 相沢 輝昭, 山下 利之
    原稿種別: 研究論文
    専門分野: ヒューマンインタフェース・知的学習支援システム
    17 巻 (2002) 3 号 p. 186-195
    公開日: 2002/04/04
    ジャーナル フリー
    There have been some studies about spoken natural language dialog,and most of them have successfully been developed within the speci ed task domains. However,current human-computer interfaces only get the data to process their programs.If the dialog processing has emotion comprehensive faculties, it should lead us to more human-like performance.In this paper,we present a method for constructing an emotion-handling dialog system in order to facilitate more confortable interaction with the users. We describe how to calculate emotions from the utterances,focusing on the similarities between the grammar structures and the semantic structures within the case frame.We made emotion generating calculations(EGC)to generate pleasure/displeasure emotion from an event.We also calculate the degree of the pleasure/displeasure from an opposite angle's length of the rectangular parallelepiped consisting of the all the terms in the EGC.EGC uses 8 type calculations for 12 event classi ed type by Okada. Word impressions about like/dislike are used for their calculations.Furthermore,we apply these calculations to the negatives and the noun phrases.To verify the e ectiveness of the proposed method,we tested some conversations using WWW-based health service system for elderly. We applied our method to 80 event in the conversations and calculated emotions almost corresponded to human-generating emotions.
    抄録全体を表示
  • 古崎 晃司, 來村 徳信, 池田 満, 溝口 理一郎
    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) 3 号 p. 196-208
    公開日: 2002/04/04
    ジャーナル フリー
    Recently, an ontology is expected to contribute to knowledge sharing and euse. It is,however, difficult to develop well-organized ontologies because the principles of ontology design are not clear enough. Therefore, a methodology for ontology design and a computer system supporting ontology design are needed. Our research goals include a methodology of ontology design, and development of an environment for building and using ontologies. Building an ontology requires a clear understanding of what can be concepts with what relations to others. An ontology thus focuses on “concepts” themselves rather than “representation” of them. Although several systems for building ontologies have been developed to date, they were not based on enough consideration of an ontological theory. We argue that a fundamental consideration of these ontological theories is needed to develop an environment for developing ontologies. Most of the previous ontologies, which are represented in frame-based languages,don't clearly deal with such concepts that need deep ontological investigation. Therefore we begin with a fundamental consideration of an ontological theory. We discuss mainly “role concept” and “relationship”, and consider how these ontologically important concepts should be treated in our environment. On the basis of the consideration we have designed and have developed an environment for building and using ontologies, named “Hozo”. This paper presents an outline of the functionality of Hozo. We focus on how it treats the relations and roles on the basis of fundamental consideration.
    抄録全体を表示
  • Yongguang Bao, Xiaoyong Du, Naohiro Ishii
    原稿種別: scientific monograph
    専門分野: Learning & Discovery
    17 巻 (2002) 3 号 p. 209-216
    公開日: 2002/04/04
    ジャーナル フリー
    The k-nearest neighbor (k-NN) classification is a simple and effective classification approach. However, it suffers from over-sensitivity problem due to irrelevant and noisy features. There are two ways to relax such sensitivity. One is to assign each feature a weight, and the other way is to select a subset of relevant features. Existing researches showed that both approaches can improve generalization accuracy, but it is impossible to predict which one is better for a specific dataset. In this paper, we propose an algorithm to improve the effectiveness of k-NN by combining these two approaches. Specifically, we select all relevant features firstly, and then assign a weight to each relevant feature. Experiments have been conducted on 14 datasets from the UCI Machine Learning Repository, and the results show that our algorithm achieves the highest accuracy or near to the highest accuracy on all test datasets. It increases generalization accuracy 8.68% on the average. It also achieves higher generalization accuracy compared with well-known machine learning algorithm IB1-4 and C4.5.
    抄録全体を表示
  • 松尾 豊, 石塚 満
    原稿種別: 研究論文
    専門分野: 自然言語
    17 巻 (2002) 3 号 p. 217-223
    公開日: 2002/04/04
    ジャーナル フリー
    We present a new keyword extraction algorithm that applies to a single document without using a large corpus. Frequent terms are extracted first, then a set of co-occurrence between each term and the frequent terms, i.e., occurrences in the same sentences, is generated. The distribution of co-occurrence shows the importance of a term in the document as follows. If the probability distribution of co-occurrence between term a and the frequent terms is biased to a particular subset of the frequent terms, then term a is likely to be a keyword. The degree of the biases of the distribution is measured by χ²-measure. We show our algorithm performs well for indexing technical papers.
    抄録全体を表示
ショートノート
論文
  • 市瀬 龍太郎, 武田 英明, 本位田 真一
    原稿種別: 研究論文
    専門分野: その他
    17 巻 (2002) 3 号 p. 230-238
    公開日: 2002/04/25
    ジャーナル フリー
    With the rapid advances of information technology, we are acquiring much information than ever before. As a result, we need tools for organizing this data. Concept hierarchies such as ontologies and information categorizations are powerful and convenient methods for accomplishing this goal, which have gained wide spread acceptance. Although each concept hierarchy is useful, it is difficult to employ multiple concept hierarchies at the same time because it is hard to align their conceptual structures. This paper proposes a rule learning method that inputs information from a source concept hierarchy and finds suitable location for them in a target hierarchy. The key idea is to find the most similar categories in each hierarchy, where similarity is measured by the κ(kappa) statistic that counts instances belonging to both categories. In order to evaluate our method, we conducted experiments using two internet directories: Yahoo! and LYCOS. We map information instances from the source directory into the target directory, and show that our learned rules agree with a human-generated assignment 76% of the time.
    抄録全体を表示
  • 池田 心, 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) 3 号 p. 239-246
    公開日: 2002/04/25
    ジャーナル フリー
    Genetic Algorithms(GAs) are effective approximation algorithms which focus on “hopeful area” in the searching process. However, in harder problems, it is often very difficult to maintain a favorable trade-off between exploitation and exploration. All individuals leave the big-valley including the global optimum, and concentrate on another big-valley including a local optimum often. In this paper, we define such a situation on conventional GAs as the “UV-phenomenon”, and suggest UV-structures as hard landscape structures that will cause the UV-phenomenon. We introduce a test function which has explicit UV-structures, and show UV-phenomenon caused by them. Next we analyze Fletcher and Powell function to confirm our hypothesis. Finally we propose a novel framework of GAs which can cope with UV-structures.
    抄録全体を表示
特集論文
論文
  • 伊藤 孝行, 服部 宏充, 新谷 虎松
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 247-258
    公開日: 2002/04/25
    ジャーナル フリー
    Online auctions are becoming an increasingly important channel for electronic commerce. There are a number of online auction sites on the Internet. It is hard for users to attend, monitor, and make bids at simultaneous multiple auctions. In this paper, we implement BiddingBot which is a multiagent system that supports users in attending, monitoring, and bidding in multiple auctions. Further, we propose a new cooperative bidding mechanism among agents. Our current experiment demonstrates that our new cooperative bidding mechanism can effectively support users to monitor, attend, and make bids on multiple simultaneous auctions.
    抄録全体を表示
  • 松村 真宏, 大澤 幸生, 石塚 満
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 259-267
    公開日: 2002/04/25
    ジャーナル フリー
    Business people, especially marketing researchers, are keen to understand peoples' potential sense of value to create fascinating topics stimulating peoples' interest. In this paper, we aim at finding influential people, comments, and terms contributing the discovery of such topics. For this purpose, we propose an Influence Diffusion Model in text-based communication, where the influence of people, comments, and terms are defined as the degree of text-based relevance of messages. We apply this model to Bulletin Board Service(BBS) on the Internet, and present our discoveries on experimental evaluations.
    抄録全体を表示
  • 平 博司, 福島 伸一, 大澤 幸生, 伊庭 斉志, 石塚} 満
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 268-275
    公開日: 2002/04/25
    ジャーナル フリー
    Information on the World Wide Web(WWW) is increasing day by day because of its open characteristics. It becomes difficult for users to find useful information in this huge WWW information space. Even if a user can fortunately find useful pages, it is difficult for him/her to acquire all the aspects or a structured knowledge view regarding his/her query.       In this paper, we describe a system called “AreaView2001”, which presents an overall structured view of the queried area together with a set of useful Web pages explaining the area and its subareas. The style of the presentation is similar to book style, consisiting of chapters and sections. When extracting important keywords of the area from collected Web pages, the system employs the method of KeyGraph which can extract keywords of the main topics and underlying basis knowledge of a text document.       AreaView2001 is particularly useful for those users that want to know unfamiliar areas, such as academic areas, since the area knowledge available in the WWW will be presented as a collection of useful Web pages sorted out according to the overall structure of the area. Although the area knowledge to be presented by the system is not so well structured as book chapters because of the full-automatic structuring, it can provide a variety of vivid knowledge not available in books. Some experimental evaluations are given to illustrate the effectiveness of the sysytem.
    抄録全体を表示
  • 濱崎 雅弘, 武田 英明, 松塚 健, 谷口 雄一郎, 河野 恭之, 木戸出 正継
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 276-284
    公開日: 2002/04/25
    ジャーナル フリー
    In this paper, we propose shared topic networks as a model of human network to organize Internet Information, and developed a system called kMedia that can generate shared topic networks by using WWW bookmark files. We also evaluate the system with experiments to know how shared topics network can help uesrs especially to know each other. A shared topic network is formed by linking topics of participants, and used to know interests of others and to exchange information with others. kMedia can generate shared topics networks by using structures of WWW bookmarks, i.e., folders of bookmarks are regarded as topics of their owners. Relations among topics of different users are estimated by aggregating similarity among pages in these topics. The experiments were performed to clarify two points; one is whether topics is a better way to exchange information among people and the other is how we can measure human relationship. The first point is examined that topic recommendation is more acceptable than page recommendation. For the second point, we propose category resemblance as measurement of human relationship. Since we compare results between cases with subjects belonging to the same community and cases without communities, we noticed similarity of topics structure is affective. The category resemblance is to estimate this similarity of topic structure and it is proved that it is better than any other parameters with respect to measurement for human relationship.
    抄録全体を表示
  • 後藤 将志, 大囿 忠親, 新谷 虎松
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 285-292
    公開日: 2002/04/25
    ジャーナル フリー
    In this paper, we propose a new method for selecting search engines on WWW for selective meta-search engine. In selective meta-search engine, a method is needed that would enable selecting appropriate search engines for users' queries. Most existing methods use statistical data such as document frequency. These methods may select inappropriate search engines if a query contains polysemous words. In this paper, we describe an search engine selection method based on thesaurus. In our method, a thesaurus is constructed from documents in a search engine and is used as a source description of the search engine. The form of a particular thesaurus depends on the documents used for its construction. Our method enables search engine selection by considering relationship between terms and overcomes the problems caused by polysemous words. Further, our method does not have a centralized broker maintaining data, such as document frequency for all search engines. As a result, it is easy to add a new search engine, and meta-search engines become more scalable with our method compared to other existing methods.
    抄録全体を表示
  • 岡本 昌之, 山中 信敏
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 293-300
    公開日: 2002/04/25
    ジャーナル フリー
    We propose a framework to construct conversational Web agents, which guide the visitors of Web sites, from actual dialogues incrementally. We apply the Wizard of Oz method enhanced by adding the functionality of learning dialogue models to the process. In this method, a developer collects annotated dialogues by chatting with users. At first, the developer has to input almost all replies. As the learning proceeds, the system infers proper utterances and the load of developer is reduced. Finally, a conversational agent is constructed. We developed a system to construct such kind of agents in the Web environment. The features of this system are: (a) FSM-based dialogue models and incremental algorithms of learning probabilistic DFAs; (b) using annotations for the meaning of each utterance and for contents associated with Web pages; and (c) a character interface with speech functionality. We also examined how the developer's cost is reduced as the growth of the dialogue models, applying it to Kyoto tour guide task.
    抄録全体を表示
  • 伊藤 禎宣, 角 康之, 間瀬 健二, 國藤 進
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 301-312
    公開日: 2002/04/25
    ジャーナル フリー
    SmartCourier is a Web-based system for browsing electronic documents shared in a group of people. It allows users to annotate on the documents with a pen interface, and share the annotations among the users having similar interests on the documents. Also, the system recommends a user unread documents inferred interesting to the user by using a collaborative filtering method. We prototyped two versions of SmartCourier. The first version used a method to matchmake users according to agreement of annotated area on the same documents without any semantic processing of the annotations and the base documents. The second version employed a method to extract keywords from the annotated area of the base documents and find other related documents sharing the keywords. The experimented evaluation shows the second version of SmartCourier was accepted by the users as more efficient system for facilitating knowledge interaction among them.
    抄録全体を表示
  • 久保田 秀和, 山下 耕二, 福原 知宏, 西田 豊明
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 313-321
    公開日: 2002/04/25
    ジャーナル フリー
    We propose a broadcating agent system called {\it POC caster} that generates understandable conversational representation from heterogeneous text-based opinions. POC caster introduces an opinion of a community member by a conversational method in Public Opinion Channel(POC) that is an interactive broadcasting system supporting community knowledge creation. The way to generate conversational representation from an opinion is consist of two processes. The first process is an analysis of an intention of the opinion by referring the last word of a sentence. The second process is applying some rules about intentions and positions of the sentences to make an understandable conversation. The psychological experiments about understandability of generated conversations are described.
    抄録全体を表示
  • 村田 剛志
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 322-329
    公開日: 2002/04/25
    ジャーナル フリー
    Discovery of representative Web pages regarding specific topics is important for assisting users' information retrieval from the Web. Researches on Web structure mining, whose goals are to discover or to rank important Web pages based on the graph structure of hyperlinks, have been very active recently. A complete bipartite of Web graph, which is composed of centers (containing useful information regarding specific topic) and fans (containing hyperlinks to centers), can be regarded as a Web community sharing a common interest. Although Murata's method for discovering Web communities is a simple method for finding related Web pages, it has the following weaknesses: (1) since the number of centers increases monotonously, pages irrelevant to the members of Web communities may be added in the process of discovery, and (2) since the number of fans decreases monotonously according as the number of centers increases, the method may suffer topic drift. This paper describes an improved method for refining Web communities in order to acquire representative Web pages of the topics of input Web communities. The method is based on the assumption that most of the fans contain hyperlinks pointing to representative pages regarding their topic, and that hyperlinks to the pages of the same quality often co-occur. In our new method, both fans and centers are renewed iteratively by the result of the majority vote of the members of previous Web community. Results of our experiments show that the new method has abilities of finding desirable pages for several topics.
    抄録全体を表示
  • 嶋津 恵子, 尾崎 知伸, 古川 康一
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 330-342
    公開日: 2002/04/25
    ジャーナル フリー
    KM (Knowledge Management) systems have recently been adopted within the realm of enterprise management. On the other hand, data mining technology is widely acknowledged within Information systems' R&D Divisions. Specially, acquisition of meaningful information from Web usage data has become one of the most exciting eras. In this paper, we employ a Web based KM system and propose a framework for applying Web Usage Mining technology to KM data. As it turns out, task duration varies according to different user operations such as referencing a table-of-contents page, down-loading a target file, and writing to a bulletin board. This in turn makes it possible to easily predict the purpose of the user's task. By taking these observations into account, we segmented access log data manually. These results were compared with results abstained by applying the constant interval method. Next, we obtained a segmentation rule of Web access logs by applying a machine-learning algorithm to manually segmented access logs as training data. Then, the newly obtained segmentation rule was compared with other known methods including the time interval method by evaluating their segmentation results in terms of recall and precision rates and it was shown that our rule attained the best results in both measures. Furthermore, the segmented data were fed to an association rule miner and the obtained association rules were utilized to modify the Web structure.
    抄録全体を表示
ショートノート
  • 齋藤 邦子, 篠原 章夫, 永田 昌明, 小原 永
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 343-347
    公開日: 2002/04/25
    ジャーナル フリー
    We propose a novel algorithm to transliterate English to Japanese and its application to a voice controlled browser, which enable ordinary Japanese people to browse English Web site by voice. Speech recognition software designed for native English speakers do not work for most Japanese because Japanese can't pronounce English as native English speakers do. Therefore, we combined Japanese speech recognition software with English-to-Japanese transliteration software. The accuracy of our transliteration algorithm is 80% recall for the top candidate, and 92% recall for the top three candidates. The browser using this transliteration algorithm makes it possible for Japanese to navigate English Web pages almost as accurate as Japanese pages by voice commands.
    抄録全体を表示
  • 福田 直樹, 大囿 忠親, 新谷 虎松
    原稿種別: 特集論文
    専門分野: 知識ベース・知識システム
    17 巻 (2002) 3 号 p. 348-353
    公開日: 2002/04/25
    ジャーナル フリー
    In this paper, we present a programming framework, `MiPage', for realizing intelligent WWW applications based on the mobile agent technology. On the framework, an agent is programmed by using hyper text markup language and logic programming language. To realize the framework, we designed a new logic programming environment `MiLog', and an agent program compiler `MiPage Compiler'. The framework enables us to enhance both richness of the services and manageability of the application.
    抄録全体を表示
    • |<
    • <
    • 1
    • >
    • >|
feedback
Top