Misc - KAZAMA Kazuhiro

Development and operations of the distributed system for the Ja-Jakarta Project

KOYAMA Hiroshi, KAZAMA Kazuhiro, OKAMOTO Takashi, YOKOTA Takehiko

Technical report of IEICE. TM ( The Institute of Electronics, Information and Communication Engineers )  104 ( 35 ) 17 - 22   2004.05

　View Summary

The Ja-Jakarta project is a volunteer project to promote the japanization of open source products, and to translate their documents into Japanese. We have developed a distributed system constructed mainly with open source products and operates them to offer a cooperating work environment to the contributors scattered all over the country. We also offer a translation support service based on XML. We evaluated our system design and operation management by the analysis of the information in project mailing lists, system logs, and the questionnaire to our contributors. As the results, we confirmed that our services such as information exchange, information dis- closure, and translation support have been effectively utilized. This paper thus reports that the open source model is effective for translation works.

A Method for Visualizing Web Search Histories

佐藤進也, 原田 昌紀, 風間 一洋

情報処理学会研究報告情報学基礎（FI）   2003 ( 51 ) 119 - 125   2003.05

　View Summary

Webサーチエンジンのログに記録されている検索行為の相互関係を，行為者，検索語，検索結果中から選択・閲覧したWebページという属性データの相互関係を使い，Query Networkというグラフ構造によって表現する方法を提案する．サーチエンジンODINの検索ログから得られたQuery Networkの実例を示しながら，その構造的特徴について述べるとともに，このネットワークを視覚化するツールを紹介する．さらに，Query Networkは複数ユーザのインプリシットな協調作業の結果として捉えられることを述べ，その情報探索手段としての可能性について議論する．We propose a method for interpreting Web search histories into the graph structure called {\em Query Network}, which can be obtained by compiling relationships between users, queries and web pages in the histories. We describe structural characteristics of Query Networks based on some examples derived from logs of the ODIN search engine. We also introduce a tool for visualizing these networks. Furthermore, we claim that the Query Network can be thought of an outcome of implicit collaborations among users. Its potential for collaborative information discoveries is explored from this point of view.

Finding Key People and Visualizing their Relationships on the Web

原田 昌紀, 佐藤進也, 風間 一洋

情報処理学会研究報告情報学基礎（FI）   2003 ( 51 ) 17 - 24   2003.05

　View Summary

我々は文書検索と固有表現抽出を組み合わせた情報検索手法NEXAS(Named Entity eXtraction and Association Search)を提案する．NEXASは，検索質問と適合する文書だけでなく，それらと関連する実世界のエンティティを発見する．これにより利用者は文書集合内の情報に加え，実世界に関する知識を得ることができる．本稿では提案手法の実際的な適用例として，与えられたトピックに関するキーパーソンをWeb上から発見するシステムについて述べる．また，キーパーソン間の関係を無向グラフとして可視化する方法を説明する．We propose NEXAS(Named Entity eXtraction and Association Search), an information retrieval method that combines document retrieval with named entity extraction. NEXAS finds not only documents relevant to a query but also real world entities related to them. This helps users to get real world knowledge as well as information in the document collection. As a practical application of the proposed method, this paper presents a system to find key people on the Web for a given topic. We also describe a visualization method of relationships of key people as undirected graphs.

Measuring Similarity among Information Sources by Comparing String Frequency Distributions

SATO Shin-ya, HARADA Masanori, KAZAMA Kazuhiro

IPSJ SIG Notes ( Information Processing Society of Japan (IPSJ) )  2002 ( 28 ) 119 - 126   2002.03

　View Summary

We propose a novel method for measuring similarity among information sources, such as web servers, by comparing distributions of string occurrence frequency in their resources. This approach is an analogue of the literary detective, which is to identify an author by comparing statistical characteristics of documents (e.g., word frequency distributions) that reflect authrs' writing styles. In this paper, we show how we have developed and validated the method. Similarity measured with this method is compared with that of derived from a Web directory service where information sources are classified and hierarchically arranged. We also describe a way to apply the similarity measuring method to selecting feature terms of information sources.

Advanced Web Search Engine Technologies (<Special Issue>"Supporting Technologies for Web-Based Information Acquisition System")

KAZAMA Kazuhiro, HARADA Masanori, Kazuhiro Kazama, Masanori Harada, NTT Network Innovation Laboratories, NTT Network Innovation Laboratories

Journal of Japanese Society for Artificial Intelligence ( 人工知能学会 )  16 ( 4 ) 503 - 508   2001.07

Automated Web Directory Expansion using Co - citation Analysis of Hyperlinks

HARADA Masanori, KAZAMA Kazuhiro, SATO Shin-ya

IPSJ SIG Notes ( Information Processing Society of Japan (IPSJ) )  2001 ( 20 ) 45 - 52   2001.03

　View Summary

With the rapid growth of the web, it is getting harder to build and maintain web directories and there are significant delays in registering or updating information. To meet the situation, we propose a new method to automatically expand a web directory using related web finding algorithms based on co-citation analysis of hyperlinks. We apply the proposed method with four algorithms respectively to a web directory in real use which consists of hundreds of categories to show effectiveness of the method and differences among the algorithms.

An N - gram indexing method for Unicode based search engine

HARADA Masanori, KAZAMA Kazuhiro, SATO Shin-ya

IPSJ SIG Notes ( Information Processing Society of Japan (IPSJ) )  2000 ( 29 ) 127 - 134   2000.03

　View Summary

We investigate some indexing methods in order to implement an Unicode based full-text search engine and propose an N-gram indexing method that varies N-gram length per character block. With the proposed indexing method applied to our WWW search engine, we estimate parameters suitable for Japanese. We also describe some possible language dependent extensions.

Some Results from an Analysis of Queries to a Search Engine

SATO Shin-ya, HARADA Masanori, KAZAMA Kazuhiro

IPSJ SIG Notes ( Information Processing Society of Japan (IPSJ) )  2000 ( 29 ) 135 - 142   2000.03

　View Summary

We present some results from an analysis of queries to a WWW search engine. We analized relationships between users and keywords of queries, and found that they resemble relationships between documents and words in them. We also show that a conventional method for text analysis can be applied to query analysis.

ハイパーリンクとアンカーテキストを利用した情報検索とランキングの一手法

風間一洋

情報処理学会研究会報告 ( 情報処理学会 )  59   17 - 24   2000

Estimation of Hierarchical Structure of Web for Search Engine

原田 昌紀, 佐藤進也, 風間 一洋

情報処理学会研究報告情報学基礎（FI）   1999 ( 39 ) 105 - 112   1999.05

　View Summary

WWWの急速な普及に伴い，サーチエンジンの重要性はますます高まりつつある。しかし，今日のサーチエンジンの出力する膨大な検索結果は未整理で，わかりにくいものとなっている。一因として既存のサーチエンジンは，Webページを独立に扱っており，WWW空間の持つ構造をなんら利用していないことが挙げられる。本稿では，Webページをまとまりを持ったグループに分ける方法と，サーチエンジンにおけるグループ単位での検索機能について述べる。ロボットで収集したWebページを用いて実験を行ない，本手法の有効性を確認した。The rapid spread of World Wide Web lets search engines be more and more important tools. However, their output are chaotic and hard to read. This is because existing search engines treat pages as individuals and do not exploit any structure of the WWW. In this paper, we describe a method to group WWW pages and the search-by-group feature of our new search engine. Experiments using actual web pages collected by a spider indicate its effectiveness.

Design of Japanese Morphological Analysis API Independent upon a Particular Program

KAMBAYASHI Takashi, KAZAMA Kazuhiro

IPSJ SIG Notes ( Information Processing Society of Japan (IPSJ) )  1998 ( 48 ) 47 - 54   1998.05

　View Summary

It is difficult to develop natural language processing programs on multiple platforms at the same time because different Japanese morphological analysis programs are required on different platforms. In this paper, we at first investigate functions of exsiting Japanese morphological analysis programs, in order to clarify issues on platform-independency. Next, we describe a design of a Japanese Morphological Analysis API that is independent of Japanese morphological analysis programs and platforms. Finally, we define several Java classes based on the design.

Java Internationalization Trends

KAZAMA Kazuhiro

IPSJ Magazine ( Information Processing Society of Japan (IPSJ) )  39 ( 4 ) 292 - 295   1998.04

Java最前線(7)Javaの国際化と日本語処理

風間 一洋

Bit ( 共立出版 )  29 ( 12 ) 91 - 101   1997.12

Compound Term Extraction for World - Wide Information Discovery

KAMBAYASHI Takashi, SHIMIZU Susumu, SATO Shin-ya, KAZAMA Kazuhiro

IPSJ SIG Notes ( Information Processing Society of Japan (IPSJ) )  1997 ( 106 ) 19 - 26   1997.11

　View Summary

Search engines are generally used when exploring a very large information space, the Internet. If search engines present users with some adequate related terms, they can easily specify the search for information resources. In this paper, as related terms which aid world-wide information discovery, we focus on compound terms that comprise a sequence of nouns. Moreover, we propose a compound term extraction method. This method organizes an appropriate document set and extracts compound terms according to their frequencies in the document set. Results from the our experimentation indicate that the compound terms are divided into four types and that the documents in a single file-system directory are a reasonable choice for a document set.

Ingrid Newscast : An Autonomous News Distribution System

KAZAMA Kazuhiro, SATO Shin-ya, SHIMIZU Susumu

IEICE technical report. Computer systems ( The Institute of Electronics, Information and Communication Engineers )  97 ( 226 ) 17 - 24   1997.08

　View Summary

We propose an autonomous agent that can gather news from the Internet. In order to assist user's global information discovery, this agent can analyze his behavior and display information that he is interested in.

Post Script.

KAZAMA Kazuhiro

The Journal of the Institute of Electronics,Information and Communication Engineers ( The Institute of Electronics, Information and Communication Engineers )  79 ( 12 ) pp.1248 - 1251   1996.12

Recognition of Parallel Program's Behavior using MIDI instruments

KAZAMA Kazuhiro, SATO Koji

IPSJ SIG Notes ( Information Processing Society of Japan (IPSJ) )  1995 ( 19 ) 21 - 26   1995.02

　View Summary

It is difficult to recognize the behavior of parallel programs such as multi-thread systems or multi-agent systems. In such a case, auralization of parallel program's behavior is as useful as visualization. In this paper, we consider how to recognize parallel program's behavior, and mapping methods of parallel program's infomation to sound.

Real-Time Multi-Agent Model for Integrated Recognition of Scene and Sound

Kazama Kazuhiro

IEICE technical report. Computer systems ( The Institute of Electronics, Information and Communication Engineers )  94 ( 163 ) 57 - 64   1994.07

　View Summary

We propose a real-time multi-agent model for integrated recognition of scene and sound.We classify all agents into three groups:a hard-realtime group,a soft-realtime group and a planner. To satisfy time constraints,all agents can adjust quality of computation and each agent of a soft-realtime group is controlled by a planner.

An Extensible Server for Distributed Systems

風間 一洋

全国大会講演論文集   42   5 - 6   1991.02

　View Summary

近年、コンピュータをネットワークで接続した分散環境は一般的に使用されるようになってきた。半面、分散システムの修正や機能追加は大規模になるほど大変な作業になり、異機種が混在するネットワークでは移植の必要性があるなど、分散システム開発は非常に困難である。従来のサーバシステムは、サーバ・クライアント間の通信はプロトコルという形で厳密に定められていて、サーバはあらかじめ決められた手続きしか実行できなかった。本稿では、分散プログラムのプロトタイピング環境として、特定の環境に依存せずに、動的な変更・拡張が可能であるサーバと通信プロトコルを提案し、分散システムに適用した場合の有効性を述べる。