JP3799905B2 - Keyword visualization method and apparatus - Google Patents

Keyword visualization method and apparatus Download PDF

Info

Publication number
JP3799905B2
JP3799905B2 JP30366399A JP30366399A JP3799905B2 JP 3799905 B2 JP3799905 B2 JP 3799905B2 JP 30366399 A JP30366399 A JP 30366399A JP 30366399 A JP30366399 A JP 30366399A JP 3799905 B2 JP3799905 B2 JP 3799905B2
Authority
JP
Japan
Prior art keywords
keyword
signal
user
unit
information
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Expired - Fee Related
Application number
JP30366399A
Other languages
Japanese (ja)
Other versions
JP2000200290A (en
Inventor
誠 平原
敏樹 金道
夏樹 岡
秀行 吉田
武彦 志田
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Panasonic Corp
Panasonic Holdings Corp
Original Assignee
Panasonic Corp
Matsushita Electric Industrial Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Panasonic Corp, Matsushita Electric Industrial Co Ltd filed Critical Panasonic Corp
Priority to JP30366399A priority Critical patent/JP3799905B2/en
Publication of JP2000200290A publication Critical patent/JP2000200290A/en
Application granted granted Critical
Publication of JP3799905B2 publication Critical patent/JP3799905B2/en
Anticipated expiration legal-status Critical
Expired - Fee Related legal-status Critical Current

Links

Images

Landscapes

  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Description

【0001】
【発明の属する技術分野】
本発明は、情報が必要か否かのユーザ入力によりユーザの興味を獲得する情報フィルタ装置で得られた情報をもとに、関係のあるキーワードをユーザに理解し易く配置し、表示することができるキーワードの可視化方法及びその装置に関するものである。
【0002】
【従来の技術】
近年の情報通信の社会基盤の進展に伴い、情報通信網の大規模化と通信量の著しい増大に対応する技術として、情報フィルタ装置(特開平9-288683号公報)の開発が進められている。この背景には、今日、個人が処理可能な情報量に対して、個人がアクセスできる情報量が上回るようになっていることがある。このため、大量の情報の中から、ユーザが必要に思う情報を適切に検索できる情報フィルタ装置が脚光を浴びている。
【0003】
【発明が解決しようとする課題】
例えば、特開平9-288683号公報の情報フィルタ装置は、ユーザの興味を反映したキーワードを自動獲得し、それに基づいて情報に前記ユーザの興味を反映したスコアを計算し、スコアの大きい順で情報を提示するものである。
【0004】
しかしながら、自動獲得された前記ユーザの興味を反映したキーワードを、前記ユーザ自身が確認することができないという課題があった。
【0005】
本発明は上記従来の課題を解決するものであり、少なくとも、ユーザの興味の反映度合いを持ったキーワードを入力として受け、前記キーワードの各々に対し、形状および大きさおよび色を持ったキーワード体を設定し、前記キーワード体を配置して、ディスプレイ上に前記キーワード体と前記キーワードを表示することを特徴とするキーワードの可視化方法であって、前記ユーザの興味のあるキーワードを、情報の分野毎にまとめて配置し、提示するキーワードの可視化装置を提供することを目的とする。
【0006】
【課題を解決するための手段】
この目的を達成するために本発明のキーワードの可視化方法は、情報が必要か否かのユーザ入力によりユーザの興味を獲得する情報フィルタ装置で得られたキーワードと、枝切り肯定メトリック信号および/または枝切り否定メトリック信号を受け、前記キーワードの各々に対し、ユーザの興味を反映した形状および大きさおよび色を持ったキーワード体を設定し、前記キーワード体を配置して、ディスプレイ上に前記キーワード体と前記キーワードを表示するようにしたものである。
【0007】
この発明によれば、前記キーワードのキーワード体の形状および/または大きさおよび/または色を、ユーザの興味の反映度合いを用いて決定することができる上、前記キーワード間の関係を利用して、関係のあるキーワードのキーワード体を情報の分野毎にまとめて配置でき、さらに、ディスプレイ上に前記キーワード体と前記キーワードを理解し易く表示することができる。
【0024】
【発明の実施の形態】
請求項1に記載の発明は、少なくともユーザの興味を反映したキーワードとそのキーワードのユーザの興味の度合いを示す評価値を入力として受ける入力部と、前記キーワードの各々に対し、形状、大きさ及び色で表現したキーワード体を設定するキーワード体生成部と、前記キーワード体をランダムに配置するように配置座標を求め、前記キーワード体の設定値及び配置座標を配置データ記憶部に記憶する配置生成部と、ディスプレイ上に前記キーワード体を前記配置座標に基づいて配置すると共に、前記各キーワードのユーザの興味の度合いを示す評価値に応じて、キーワード体同士を近くに配置し、近くに配置されたキーワード体を一つのキーワード体群を形成し、前記キーワード体の表示の仕方が同一になるように変更する表示部とを具備することを特徴とするキーワードの可視化装置としたものであり、ユーザの興味を反映したキーワードを理解し易く表示できるという作用を有する。
【0028】
請求項に記載の発明は、請求項記載の自動配置装置において、前記ユーザの興味の度合いは、少なくとも1つのキーワードが付与された情報が必要か否かのユーザ入力により、キーワード毎にユーザの興味を獲得したことを特徴としたものであり、前記ユーザの興味を反映したキーワードを理解し易く表示できるという作用を有する。
【0029】
請求項に記載の発明は、請求項記載のキーワードの可視化装置において、キーワード体の表示の仕方は、対応するキーワードに対してのユーザの興味の反映度合いを用いて、前記キーワードのキーワード体の少なくとも形状、大きさ、色の少なくとも1つ決定することを特徴とすることを特徴としたものであり、前記ユーザの興味を反映したキーワードを理解し易く表示できるという作用を有する。
【0030】
請求項に記載の発明は、請求項に記載の可視化装置において、ディスプレイ上のキーワード体の表示は、キーワード体に重ねてキーワードを表示することを特徴としたものであり、キーワードを理解し易く表示できるという作用を有する。
【0032】
請求項に記載の発明は、請求項に記載のキーワードの可視化装置において、キーワードのユーザの興味の度合いを示す評価値は、少なくとも1つのキーワードが付与された情報が必要か否かのユーザ入力により、ユーザの興味を反映したキーワードについて作られた(各情報で同時に出現したキーワードに対する関係とその強さ(頻度)を)自己相関行列で示したメトリック信号(肯定メトリック信号/否定メトリック信号)を含むことを特徴としたものであり、前記ユーザの興味を反映したキーワードを理解し易く表示できるという作用を有する。
【0033】
請求項に記載の発明は、請求項記載のキーワードの可視化装置において、前記メトリック信号は、ユーザの興味の度合いの大きなキーワードのみとした枝切りメトリック信号(枝切り肯定メトリック信号/枝切り否定メトリック信号)であることを特徴としたものであり、ユーザの興味の度合いの小さいキーワードを削除することにより、前記ユーザの興味を反映したキーワードを理解し易く表示できるという作用を有する。
【0035】
請求項に記載の発明は、請求項に記載のキーワードの可視化装置において、前記一つのキーワード体群のキーワード体の表示色を同一とすることを特徴としたものであり、前記ユーザの興味を反映したキーワードを、キーワード群毎にまとめて理解し易く表示できるという作用を有する。
【0036】
請求項に記載の発明は、請求項に記載のキーワードの可視化装置において、前記入力部は、ユーザの興味を反映した情報として、ユーザの興味を反映したキーワードとそのキーワードのユーザの興味の度合いを示す評価値及びキーワード間の関係情報を示す分類ラベルを受け、前記配置生成部は、分類ラベルが同じキーワードのキーワード体が一つのキーワード体群を形成することを特徴としたものであり、前記ユーザの興味を反映したキーワードを、精度良く、前記キーワードの分野毎にまとめて理解し易く表示できるという作用を有する。
【0037】
請求項に記載の発明は、請求項記載のキーワードの可視化装置において、前記一つのキーワード体群のキーワード体の表示色を同一とすることを特徴としたものであり、前記ユーザの興味を反映したキーワードを、前記キーワードの分野毎にまとめて理解し易く表示できるという作用を有する。
【0038】
請求項10に記載の発明は、請求項に記載のキーワードの可視化装置において、前記キーワード間の関係情報は、少なくとも1つのキーワードが付与された情報が必要か否かのユーザ入力により、ユーザの興味を反映したキーワードについて作られた(各情報で同時に出現したキーワードに対する関係とその強さ(頻度)を)自己相関行列で示したメトリック信号(肯定メトリック信号/否定メトリック信号)から、キーワード間に関係があるキーワードを同一の分類ラベルを付けて求めた分類ラベルを含むことを特徴としたものであり、前記ユーザの興味を反映したキーワードを、前記キーワードの分野毎にまとめて理解し易く表示できるという作用を有する。
【0039】
請求項11に記載の発明は、請求項に記載のキーワードの可視化装置において、前記入力部は、ユーザの興味を反映した情報として、ユーザの興味を反映したキーワードとそのキーワードのユーザの興味の度合いを示す評価値(及びキーワード間の関係情報を示す分類ラベル)を受け、前記配置生成部は、前記キーワード体生成部で生成されたキーワード体の初期配置としてランダムに配置するように配置座標を生成する初期配置生成部と、前記キーワード体の配置座標と前記キーワードのユーザの興味の度合いを示す評価値から分類ラベルが同じキーワード体が1つのキーワード体群としてまとまるように前記配置の更新を繰り返し行なう配置更新決定部とを具備することを特徴としたものであり、前記ユーザの興味を反映したキーワードを、高速に、精度良く、前記キーワードの分野毎にまとめて理解し易く表示できるという作用を有する。
【0049】
以下、本発明の実施の形態について、図1から図25を用いて説明する。
【0050】
(実施の形態1)
図1は、本発明の実施の形態1のキーワード可視化システムのブロック図を示す。図1において、1000はユーザの興味を反映したキーワードを理解し易く配置し、表示する可視化装置、1001はメモリ、1002はCPU、1003はディスプレイ/キーボード、1004はキーワードを自動配置したデータを記録した配置データ記憶部、1005はプログラム等を読み込むためのFD(フロッピーディスク)ユニット、1006はI/Fユニット、1007はユーザの興味を反映した情報を生成する情報フィルタ装置、2は符号辞書記憶部、532は枝切り肯定メトリック記憶部、533は枝切り否定メトリック記憶部、538は分類ラベル記憶部、1008はI/Fユニット、1009は可視化装置が情報フィルタ装置1007からの情報を受けるためのネットワークで構成されている。
【0051】
ここで、ユーザの興味を反映した情報とは、ユーザの興味を反映したキーワード、後述の枝切り肯定メトリック信号、後述の枝切り否定メトリック信号、後述の分類ラベルベクトルであり、それぞれ符号辞書記憶部2(図2参照)、枝切り肯定メトリック記憶部532(図13参照)、枝切り否定メトリック記憶部533(図13参照)、分類ラベル記憶部538(図13参照)に記憶されるものとする。
【0052】
以下、本発明の第1の実施の形態について、図面を参照しながら説明する。まず、本発明の根幹であるキーワードの可視化方法およびその装置の説明に先立って、本発明への入力となるユーザの興味を反映したキーワード、枝切り肯定メトリック信号、枝切り否定メトリック信号、分類ラベルベクトルを生成する情報フィルタ装置について説明する。
【0053】
図2は前記情報フィルタ装置の構成を示すブロック図であり、図3はその構成と動作を分かりやすくするために機能単位にまとめたブロック図である。
【0054】
まず、図3を用いて、前記情報フィルタ装置の基本概念を説明する。
【0055】
前記情報フィルタ装置は、ユーザーがどんな「情報」を過去に必要としたかという履歴に関する記録を記憶した複数の記憶部2、5、6、8と、「情報」のフィルタリングを行う情報フィルタリングユニット50と、その情報フィルタリングユニット50により実際にフィルタリングされた未読の「情報」(ユーザーがまだ読んでいない情報)を蓄積しておく未読データ記憶部10と、ユーザーが当該未読「情報」を可視できるようにしたディスプレー等の分類インタフェースユニット70と、ユーザーがどんな「情報」を必要としたかという履歴に関する学習を行う学習ユニット52と、ユーザーがどんな「情報」を過去に必要としたかという履歴に関する記録を記憶した複数の記憶部5、6に格納された記録を複数の記録に分割するメトリック分割ユニット53と、分割された記録を用いて「情報」のフィルタリングを行う複数(図面には2個の場合を記載)の分類情報フィルタリングユニット55−1、55−2と、分類未読データ記憶部56−1、56−2とからなる。
【0056】
以下、上記構成の動作について説明する。なお、以下の説明では既にユーザーがどんな「情報」を過去に必要としたかという履歴は学習済みのこととして説明する。また、以下に単に「情報」と称するものには、当該「情報」に対応する1つ以上のキーワードが付されているものとする。そのキーワードとは、当該「情報」を構成する各単語の一部あるいは全体であっても良いし、当該「情報」を代表するために特別に付したものであっても良い。
【0057】
まず、情報フィルタリングユニット50に新たな「情報」が入力されると、情報フィルタリングユニット50は、記憶部2、5、6、8からユーザーがどのような「情報」を過去に必要としたかという記録を読みだし、前記新たな「情報」の必要性を必要性信号として定量的に評価する。
【0058】
同様に、分類情報フィルタリングユニット55も前記新たな「情報」の必要性を必要性信号として定量的に評価する。
【0059】
次に、その評価された新たな「情報」は、未読データ記憶部10および分類未読データ記憶部56に、情報フィルタリングユニット50並びに各分類情報フィルタリングユニット55−1、55−2によって計算された各必要性信号が大きい順に過去からの未読「情報」を含めて並ぶように前記入力された「情報」を当該順番に書き込む。そして、ユーザーが望めば、分類インタフェースユニット70では、ユーザーに必要性信号の大きい順に前記新たな「情報」を含めた未読「情報」を1つひとつ提示(例えば、データ表示装置34に表示)する。
【0060】
この際に、ユーザーに提示された前記新たな「情報」を含めた未読「情報」の1つひとつがユーザーにとって必要か不要かを示す教師信号をユーザーが分類インタフェースユニット70の教師信号入力端子105を介して入力することにより、分類インタフェースユニット70では、当該教師信号を受け取り、当該「情報」とその教師信号を学習ユニット52に送る。なお、このユーザーによる教師信号の入力は、学習ユニット52の学習能力をより高めるために実施するものであり、学習ユニット52の学習能力(ユーザーがどんな「情報」を過去に必要としたかという履歴の学習能力)が既に十分に高ければ行う必要はない。
【0061】
次に、学習ユニット52では、前記提示した「情報」とその教師信号を用いて記憶部2、5、6、8の履歴内容を書き換える。
【0062】
以上により、前記情報フィルタ装置は、より高い学習を通じてユーザーに適応し、ユーザーの求める「情報」を分類し、優先的に提示することができる。また、当然のことながら、学習を行っていない初期状態では、ユーザーがどんな「情報」を必要としているのか学習ユニット52では分からないので、全ての入力される新たな「情報」を分類インタフェースユニット70でユーザーが提示を受ける毎に上述したユーザーによる教師信号の入力は必要であるが、随時実施する学習を通じてやがてユーザーに適応し、ユーザーの求める「情報」を分類しかつ優先的に提示することができる。
【0063】
なお、ユーザーの求める「情報」を優先的に提示するとは、より具体的な使用例で述べれば、ある「情報」データベースの母集団Aを特定のキーワードで検索して「情報」の検索集合Bを得たとしても、当該検索集合Bの「情報」の全てがユーザーにとっては必要であるとは限らないし、またユーザーにとっては「情報」の全てが必要であってもその必要順位は当然存在することを前提としている。よって、必要から順に不要、あるいはその必要順位に従って、分類インタフェースユニット70のデータ表示装置34でユーザーの必要順に提示することを、ユーザーの求める「情報」を優先的に提示することを意味する。
【0064】
上で述べた如く、入力された「情報」にキーワードが添付されている場合を考える。一人のユーザーを考えると、そのユーザーが必要としている「情報」に高い頻度または確率で付いているキーワード集合Aと、不要としている「情報」に高い頻度または確率で付いているキーワード集合Bと、さらにはいずれにもよく付く、または付かないキーワード集合Cとを考えることができる。
【0065】
したがって、前記キーワード集合Aに属するキーワードには正の数値を、前記キーワード集合Bに属するキーワードには負の値を、前記キーワード集合Cに属するキーワードには値0をそれぞれ割り振る。
【0066】
そして、新たに入力された「情報」についている1つ以上のキーワードについてそれぞれが前記キーワード集合A、B、Cのどのキーワードグループに属するかを判定し、前記割り振られた値を積算するように構成する。
【0067】
このように構成すれば、前記新たに入力された「情報」に付いていた複数のキーワードを、キーワード集合Aに属するキーワードが数多く含まれた「情報」(ユーザーが必要とする可能性の高い情報)に対しては大きな正の値を示し、キーワード集合Bに属するキーワードが数多く付いている「情報」(ユーザーが不要とする可能性の高い情報)に対しては大きな負の値を示す数値に変換することができる。
【0068】
こうして、前記数値を用いてユーザーの必要性に予測することができる。このように、前記情報フィルタ装置は、提示した「情報」とその「情報」に関するユーザーの必要/不要の評価とからキーワード(キーワード共起を含む)への値の割り振りを自動的に行い精度の高い必要性信号の計算を実現し、精度高く必要性の高い順に「情報」を並べ変えることを実現している。
【0069】
そのために、ここでは、「情報」に付けられた複数のキーワードを一つのベクトルに変換し、ユーザーが必要とした場合と不要とした場合について、別々に前記ベクトルの自己相関行列を計算している。ユーザーが必要と答えた「情報」についていたキーワードから作られた自己相関行列MYを用いて、ベクトルVの長さSYを
SY=ΣΣMij・Vi・Vj
ij
と計算する。
【0070】
なお、以下、必要と答えた「情報」についていたキーワードから作られた自己相関行列MYを「肯定メトリック信号」、不要と答えた情報についていたキーワードから作られた自己相関行列MNを「否定メトリック信号」と呼び、長さSYを肯定信号と呼ぶ。
【0071】
この長さSYは、ベクトルVの元となった複数のキーワードの中に、ユーザーが必要とする「情報」によく含まれているキーワードが数多く含まれていれば、長さSYは大きな正の値をとり、そうでない場合には0に近い値をとるから、必要性信号を計算する上で有効である。
【0072】
メトリック分割部53は、こうして生成された肯定メトリック信号MYと否定メトリック信号MNとを複数(図3の場合は2つ)に分割し、分類情報フィルタリングユニット55−1、55−2に書き込む。
【0073】
分類情報フィルタリングユニット55は、分割された肯定メトリック信号と否定メトリック信号とを用いて、前記情報フィルタリングユニット50と同様の動作を行い、ユーザーに分類され、かつ必要性の高い順に「情報」を並べて提示する。
【0074】
つぎに、図2を用いて図3に示した情報フィルタリングユニット50に相当するブロックを、図4を用いて図3に示した学習ユニット52に相当するブロックについて、機能単位毎に説明する。
【0075】
まず、情報フィルタリングユニット50に相当するブロックの構成を説明する。
【0076】
情報フィルタリングユニット50は、個々の「情報」につけられた複数のキーワード(正確には、分類コードを含む文字列)をベクトルに変換する部分と、ユーザーがどんな「情報」を必要/不要としたという履歴を表現した肯定メトリック信号及び否定メトリック信号を用いてある種のスコアを表す肯定信号と否定信号を計算する部分と、この肯定信号と否定信号とから「情報」の必要性をよく反映する必要性信号を計算する部分と、この必要性信号の大きい順に情報を並べ変える部分からなる。以下、情報フィルタリングユニット50に相当するブロックの構成を、図2に即して説明する。
【0077】
図2において、1は「情報」に付けられたキーワードなどの複数の文字列をベクトルに変換するベクトル生成部、2はキーワードなどの複数の文字列をベクトルに変換するための符号辞書信号を記憶した符号辞書記憶部である。この符号辞書記憶部2に記憶された符号辞書信号は、(数1)に示すような「情報」についているキーワードなどの文字列Wを数字Cに変換する対応表をnofDCK個有するコードブックである。
【0078】
【数1】

Figure 0003799905
【0079】
ベクトル生成部1はキーワード数信号nofKsとnofKs個のキーワード信号からなるキーワード群信号Ks=(K[1],・・・,K[nofKs])とを受けキ ーワード群信号Ksと前記符号辞書信号DCKを用いてベクトル信号Vに変換する。3はスコア計算部で、ユーザーに提示された「情報」を必要/不要と評価した結果から計算された肯定メトリック信号MY、否定メトリック信号MNを用いて、ベクトル生成部1で変換された2つのベクトル信号Vの長さ、肯定信号SYと否定信号SNに変換する。5は(nofDCK×nofDCK)行列である前記肯定メトリック信号MYを記憶する肯定メトリック記憶部、6は(nofDCK×nofDCK)行列である前記否定メトリック信号MNを記憶する否定メトリック記憶部である。8は判定パラメータ信号Cを記憶する判定パラメータ記憶部、7は前記肯定信号SYと前記否定信号SNを受け前記判別パラメータ記憶部8から判定パラメータ信号Cを読み出し必要性信号Nと信頼性信号Rを計算する必要性計算部である。
【0080】
9は「情報」の本文である情報データDとキーワード数信号nofKsとキーワード群信号Ks と必要性信号Nと信頼性信号Rとを所定の手続きに従って後述する未読データ記憶部10に書き込む未読データ書き込み制御部、10は(数2)に示す未読データ、すなわち、前記「情報」の本文である情報データDと前記キーワード数信号nofKsと前記キーワード群信号Ksと前記必要性信号Nと前記信頼性信号Rとからなる最大nofURD個の未読データを記憶する未読データ記憶部である。
【0081】
【数2】
Figure 0003799905
【0082】
13は(数3)に示す最大nofTD個の教師データ信号を記憶する教師データ記憶部である。
【0083】
【数3】
Figure 0003799905
【0084】
を記憶する教師データ記憶部である。
【0085】
次に、図3で示した分類インタフェースユニット70のブロックの構成を説明する。
【0086】
図3において、70はデータ読み出し開始信号入力端子103から制御信号DOを受け、未読データ記憶部10または分類未読データ記憶部56−1および56−2の中から使用者によって選択されたいずれかの記憶部から未読データ信号URD[1]を読み出し、表示信号DDをデータ表示装置34に出力し、その表示信号DDがユーザーにとって必要か否かを示す教師信号Tを教師信号入力端子105よりユーザーから受け、前記教師信号Tと前記未読データ信号URD[1]のキーワード数信号nofKs[1]とキーワ ード群信号Ks[1]とを所定の手続きに従って教師データ記憶部13に書き込む分類インタフェースユニットである。
【0087】
次に、図3で示した学習ユニット52に相当するブロックの構成を説明する。学習ユニット52は、ユーザーから入力された教師信号Tを用いて肯定/否定メトリック信号を修正するメトリック学習を行う部分と、肯定/否定信号から必要性信号を計算するためのパラメータ、判定パラメータ信号を修正する部分からなり、各部分は学習制御部によって制御される。
【0088】
図4に示したメトリック学習を行う部分の構成は次のようである。図4において、19は肯定メトリック記憶部5に記憶された肯定メトリック信号MYと前記否定メトリック記憶部6に記憶された否定メトリック信号MNとを修正するメトリック学習部である。このメトリック学習部19は、教師データ記憶部13から前記教師データTDを読み出し、情報フィルタユニット50のベクトル生成部1と同じ機能である学習用ベクトル生成部20で複数のキーワードをベクトルに変換し、自己相関行列を計算することで、肯定/否定メトリック信号を修正する。
【0089】
判定パラメータ信号の学習を行う部分の構成は次のようである。
【0090】
図4において、22は学習用肯定信号計算部221と学習用否定信号計算部222とからなる学習用スコア計算部である。この学習用スコア計算部において、221は学習用ベクトル生成部20からの学習用ベクトル信号を受け学習用肯定信号LSYを計算する学習用肯定信号計算部、222は学習用ベクトル生成部20からの学習用ベクトル信号を受け学習用否定信号LSNを計算する学習用否定信号計算部である。21は学習制御部14からの判定パラメータ学習制御信号PLCを受けて所定の方法で判定パラメータ記憶部8の判定パラメータ信号を書き換える判定面学習部、14は学習開始信号LSを受けてスイッチ16、17、18とメトリック学習部19と学習用ベクトル生成部20と学習用スコア計算部22と学習用否定信号計算部23と判定面学習部21とを制御する学習制御部である。
【0091】
以上のように構成された情報フィルタ装置について、各ユニットごとに図面を用いてその動作を説明する。
【0092】
情報フィルタ装置の好ましい初期状態の一例は、肯定メトリック信号MYと否定メトリック信号MNとを(nofDCK×nofDCK)零行列、未読データ記憶部10の未読データURD[i]の全ての必要性信号N[i](i=1,・・・,nofURD)を使用するハードウエアが表現可能な最小の値Vmin、教師データ記憶部13の教師データTD[j]の教師信号T[j]を全て−1とした状態である。
【0093】
最初に、情報フィルタリングユニット50の動作を説明する。
まず、情報データ入力端子100から情報データDが入力され、キーワード数信号入力端子101から情報データに付けられたキーワードの個数を表すキーワード数信号nofKsが入力され、キーワード信号入力端子102から複数のキーワー ドであるキーワード群信号Ks=(K[1],K[2],・・・,K[nofKs])が入力される。
【0094】
ベクトル生成部1によってキーワード群信号Ksは、文字列の集まりからベクトル信号Vへと変換される。この変換によって、キーワード群信号の類似性をベクトルの距離として計算できるようになる。
ベクトル生成部1の動作を図5に示すフローチャートを参照しながら説明する。まず、キーワード数信号nofKsとキーワード群信号Ksを受けると(図5ステッ プ(イ))、内部のベクトル信号V=(V[1],V[2],・・・,V[nofDic])を(0,0,・・・,0)に、キーワードカウンタ信号iを1にセットする(同図ステップ(ロ)、(ハ))。次に、辞書カウンタ信号jを0セットした後辞書カウンタ信号jを1だけ増やす(同図ステップ(ニ))。
【0095】
次に、内部にnofDCK個の符号辞書信号DCKを有する辞書記憶部2から辞書カウンタjが指定するキーワードと数字からなる符号辞書信号DCK[j]を読み出し、符号辞書信号DCKの文字列部分W[j]とi番目のキーワード信号K[i]とを比較する(同図ステップ(ホ))。両者が等しくない場合には、辞書カウンタjを1だけ増やす(同図ステップ(ヘ))。両者が一致するか、辞書カウンタjの値が辞書記憶部2に格納された符号辞書信号の個数nofDiCと等しくなるまで図4ステップ(ホ)〜(ト)の処理を繰り返す(同図ステップ(ト))。
【0096】
キーワード信号K[i]と等しいW[j]が見つかると、ベクトル信号のj番目の成分V[j]を1にし(同図ステップ(チ))、キーワードカウンタ信号iを1だけ増やす(同図ステップ(リ))。以下、同様の処理をキーワードカウンタ信号iがキーワード数信号nofKsより大きくなるまで実行する(同図ステップ (ヌ))。
【0097】
こうして、ベクトル生成部1において、文字列信号からなるキーワード信号の集合体であるキーワード群信号Ksは、0と1でコード化されたnofDCK個の成分を持ったベクトル信号Vに変換される。
【0098】
次に、肯定信号計算部31は、キーワード群信号Ksに過去にユーザーの必要とした情報に含まれていたキーワードが数多く含まれる場合に、大きな値となる肯定信号SYを計算する。この目的のために、肯定信号計算部31は、前記ベクトル信号Vを受けて、肯定メトリック記憶部5から肯定メトリック信号MYを読み出し、肯定信号SYを(数4)により計算する。
【0099】
【数4】
Figure 0003799905
【0100】
否定信号計算部32は、キーワード群信号Ksに過去にユーザーの不要とした情報に含まれていたキーワードが数多く含まれる場合に、大きな値となる否定信号SNを計算する。この目的のために、否定信号計算部32は、否定メトリック記憶部6から否定メトリック信号MNを読み出し、否定信号SNを(数5)により計算する。
【0101】
【数5】
Figure 0003799905
【0102】
肯定メトリック信号MYと否定メトリック信号MNは、後述するようにキーワード群信号Ksとユーザーの応答に基づいて決められる。このように計算された肯定信号SYと否定信号SNを用いて、図11に示したように縦軸に肯定信号SYをとり横軸に否定信号SNをとった2次元空間上の1点に、情報データDを対応させることができる。この2次元空間における情報データDの分布は、ユーザーが必要とするもの(○で表示)は主に左上部に分布し、ユーザーが不要とするもの(×で表示)は主に右下部に分布するようになる。したがって、図12に示したように適切な係数Cを定めることにより、ユーザーが必要とる情報データDと不要な情報データDとを分離できる。
【0103】
さらに、以下に述べるこの係数Cを用いて計算される必要性信号Nは、上述の2次元空間で左上にある程、すなわち、必要性の高いと予測される情報データDほど大きな値となる。したがって、必要性信号Nの大きい順に情報データDを並べて提示すれば、ユーザーは必要な情報を効率よく手に入れることができる。必要性信号Nと直交する方向の信頼性信号Rは、大まかにはキーワード群信号Ksに含まれていたキーワードのうちどのくらいのキーワード信号が辞書に含まれていたかを示す信号である。したがって、この信頼性信号Rの大きさは、情報フィルタが計算した必要性信号Nがどれだけ信頼できるのかを示す。
【0104】
次に、必要性計算部7は、前記肯定信号計算部31から出力される前記肯定信号SYと前記否定信号計算部32から出力される前記否定信号SNとを受け、判定パラメータ記憶部8から判定パラメータ信号Cを読み出し、過去必要であった情報についていたキーワードが多数あり、不要であった情報についていたキーワードがほとんどない時に大きな値となる必要性信号Nを
N=SY−C・SN
と計算し、信頼性信号Rを
R=C・SY+SN
と計算する。
【0105】
未読データ書き込み制御部9の動作を、図6に示したフローチャートを参照しながら説明する。まず、それぞれの入力端子から前記情報データDと前記キーワード数信号nofKsと前記キーワード群信号Ksと受け、必要性計算部7から前記 必要性信号Nと前記信頼性信号Rとを受け、未読データ部指示端子110から出力する未読データ処理信号WIを0から1に変える(図6ステップ(イ))。次に、i=1とし(同図ステップ(ロ))、未読データ記憶部10に記憶された未読データURD[i]の必要性信号N[i](i=1,・・・,nofURD)を順次読み出し、前記必要性信号Nと比較し(同図ステップ(ハ))、前記必要性信号Nが未読データURD[i]の必要性信号N[i]より大きくなる(N≧N[i])最初の未読データの番号i1を検出する(同図ステップ(ニ)(ホ))。
【0106】
i1番目以降の未読データを
URD[i+1]=URD[i] i=i1,・・・,nofURD
と置き換え(同図ステップ(ヘ)〜(リ))、その後、i1番目の未読データURD[i1]を
N[i1] =N
R[i1] =R
nofKs[i1] =nofKs
Ks[i1] =Ks
D[i1] =D
と前記必要性信号N等で置き換える(同図ステップ(ヌ))。この置き換えが終了すると、未読データ部指示端子110から出力する未読データ部指示信号WIを0に戻し(同図ステップ(ル))、処理を終了する。
【0107】
次に、未読データUDRを読みだし、ユーザーの応答(教師信号T)を付加して教師データ信号TDをつくる分類インタフェースユニット70について説明する。分類インタフェースユニット70の動作を図7に示したフローチャートを参照しながら説明する。
【0108】
以下、好ましい実施形態では、未読データ記憶部10は、分類ラベル信号Catが0である分類未読データ記憶部として取り扱われる。以下では、分類インタフェースユニットの動作の説明の際には、未読データ記憶部10は分類ラベル信号Catが0である分類未読データ記憶部として取り扱う。
【0109】
データ読み出し開始信号入力端子103から、データ読み出し開始信号DOが入力される(図7ステップ(イ))。分類インタフェースユニット70は、アクセスするべき未読データ記憶部10または分類未読データ記憶部カテゴリ56を指定する分類ラベル信号Catを受け取るべく予め定められた時間待機する(同図ステップ(ロ))。前記時間内に分類ラベル信号Catが入力されなかった場合、分類インタフェースユニットは、分類ラベル信号Catに標準値(好ましい実施の形態では、Cat=0)を設定する(同図ステップ(ニ))。前記時間内に分類ラベル信号Catに値が入力されるか、または標準値が設定されると、分類ラベル信号Catで指定される分類未読データ記憶部56−Catから、1番目の未読データURD[1]を読み出し(同図ステップ(ホ))、未読データの必要性信号N[1]が最小値Vminより大きい場合には、未読データ信号URD[1]の情報信号D[1]を表示情報信号DDとしてデータ表示装置34に出力し、待機する(同図ステップ(へ)、(ト))。未読データの必要性信号N[1]が最小値Vminに等しい場合には、表示情報信号DDを「データなし」としてデータ表示端子104に出力し、待機する(同図ステップ(チ))。
【0110】
ユーザー(図示せず)は、データ表示装置34に表示された表示情報信号DDを見て、それが必要な情報である場合には教師信号T=1、必要でない場合には教師信号T=0、処理を終了する場合には教師信号T=−1として、教師信号入力端子105に返す(同図ステップ(リ))。教師信号T=−1の場合、処理を終了し、教師信号T≠−1の場合には(同図ステップ(ル))、未読データ出力制御部11は、教師データ記憶部13の(数2)で表わされる教師データを
TD[i]=TD[i−1], i=2,・・・,nofTD
と置き換え(同図ステップ(ヲ))、1番目の教師データTD[1]を前記教師信号Tと前記未読データのキーワード数信号nofKs[1]とキーワード群信号Ks[1]とを用いて
T[1] =T
TnofKs[1] =nofKs[1]
TKs[1] =Ks[1]
と置き換え(同図ステップ(ワ)、(ヨ))、前記未読データ記憶部10の未読データURDを
URD[i]=URD[i+1], i=1,・・・,(nofURD−1)とし(同図ステップ(タ)、(レ))、nofURD番目の未読データの必要性信号をN[nofURD]=(最小値Vmin)
とする(同図ステップ(ツ)、(ネ)、(ナ))。
【0111】
次に、学習ユニット52の動作について図8〜図20に示したフローチャートを参照しながら説明する。
【0112】
図8に学習制御部14の動作の概略を示すフローチャート示し、詳しく説明する。
【0113】
図8において、まず、学習開始信号入力端子106から学習開始信号LSが入力され、学習制御部指示信号出力端子107から出力される学習制御部指示信号LIを0から1に変え(図8ステップ(イ))、処理中を示す。次に、スイッチ16とスイッチ17とスイッチ18とをメトリック学習部19と学習用ベクトル生成部20が接続する様に切り替える(同図ステップ(ロ))。
【0114】
次に、図8のステップ(ハ)に対応するメトリック学習部19を動作し(同図ステップ(ハ)、判定面学習部21を動作させた後(同図ステップ(ニ))、LIを0として(同図ステップ(ホ))、処理を終了する。
【0115】
次に、メトリック学習部19がユーザーの応答(教師信号T)とキーワード群信号Ksとを用いて、肯定/否定メトリック信号を修正する動作を図9を用いて説明する。
【0116】
図9は、メトリック学習部19の動作のフローチャートであり、同図において、前記学習制御部14からメトリック学習制御信号MLCを受けた(図9ステップ(イ))メトリック学習部19は、肯定メトリック記憶部5から肯定メトリック信号MYを、否定メトリック記憶部6から否定メトリック信号MNをそれぞれ読み出す。
【0117】
次に、メトリック学習部19は、教師データカウンタcの値を1にする(同図ステップ(ロ))。次に教師データ記憶部13からc番目の教師データ信号TD[c]を読み出し(同図ステップ(ハ))、教師データTD[c]の教師信号T[c]を調べる。前記教師信号T[c]が−1でない場合(T≠−1)には(同図ステップ(ニ))、教師データTD[c]のキーワード数信号TnofKs[c] とキーワード群信号TKs[c]とを出力する(同図ステップ(ホ))。前記教師データTD[c]のキーワード数信号TnofKs[c]とキーワード群TKs[ c]とを受けた学習用ベクトル生成部20は、前述の情報フィルタリングユニット50のベクトル生成部1と同様の動作を行い、学習用ベクトル信号LVを出力する(同図ステップ(ヘ))。メトリック学習部19は、前記学習用ベクトル信号LVを受け、前記教師データTD[c]の教師信号T[c]がT=1である場合には(同図ステップ(ト))、肯定メトリック信号MYを
MY[i][j]=MY[i][j]+LV[i]・LV[j]
(ここで、i,j=1〜nofDiC)
と修正する(同図ステップ(チ))。
【0118】
この処理により、肯定メトリック信号MYは、ユーザーが必要とした情報データDについていたキーワード信号(複数)に対して大きな値を持つようになる。その結果、前述の肯定信号SYが、ユーザーが必要とする情報データDに対して大きくなるようになる。否定メトリック信号MNも以下のように同様の処理がなされる。
【0119】
前記教師データTD[c]の教師信号T[c]がT=0である場合には、否定メトリック信号MNを
MN[i][j]=MN[i][j]+LV[i]・LV[j]
(ここで、i,j=1〜nofDiC)
と修正する(同図ステップ(リ))。ついで、教師データカウンタの値を
c=c+1
と1だけ増やす(同図ステップ(ヌ))。
【0120】
以下、メトリック学習部19は、同様の動作を、教師データTD[c]の教師信号T[c]がT[c]=−1になるかまたはc=nofTDとなるまで繰り返す。T[c]=−1またはc=nofTDとなると(同図ステップ(ヲ))、メトリック学習の処理を終了し、メトリック学習制御信号MLCを学習制御部14に送る。
【0121】
学習制御部14は、メトリック学習部19からのメトリック学習制御信号MLCを受けて、スイッチ16を学習用ベクトル生成部20とスコア計算部22とが接続するように切り替え、スイッチ17とスイッチ18を学習用ベクトル生成部20と判定面学習部21とが接続するように切り替える。学習制御部14は、判定面学習制御信号PLCを判定面学習部21に送る。
【0122】
次に、判定面学習部21について、図10を用いて詳しく説明する。
判定面学習部21は、図12に示したように、肯定信号SYと否定信号SNを用いて2次元空間上に表現されたユーザーが必要とする情報データDとユーザーが不要とする情報データDとをもっともよく分離する係数Cを求めるものである。
【0123】
この目的のために、図10に示したフローチャートに従って詳しく説明する。
【0124】
まず、前記判定面学習制御信号PLCを受けて(図10ステップ(イ))、教師データカウンタcの値を1にする(同図ステップ(ロ))。教師データ記憶部13からc番目の教師データ信号TD[c]を読み出し(同図ステップ(ハ))、教師データTD[c]の教師信号T[c]を調べる(同図ステップ(ニ))。前記教師信号T[c]が−1でない場合(T≠−1)には、教師データTD[c]のキーワード数信号TnofKs[c]とキーワード群信号TKs[c]とを出力す る(同図ステップ(ホ))。前記教師データTD[c]のキーワード数信号TnofKs[c]とキーワード群TKs[c]とを受けた学習用ベクトル生成部20は 、前述した情報フィルタリングユニット50のベクトル生成部1と同様の動作を行い、学習用ベクトル信号LVを出力する。
【0125】
学習用スコア計算部22は、前述した情報フィルタリングユニット50のスコア計算部3と同様の動作を行い、学習用肯定信号LSY[c]と学習用否定信号LSN[c]とを出力し、判定面学習部21がそれを受ける(同図ステップ(ヘ))。前記学習用肯定信号LSY[c]と前記学習用否定信号LSN[c]と教師データTD[c]の教師信号T[c]と判定面学習用信号TC[c]=(T[c],LSN[c],LSY[c])を内部の記憶素子に記憶する(同図ステップ(ト))。そして、教師データカウンタの値を
c=c+1
と1だけ増やす(同図ステップ(チ))。
【0126】
以下、判定面学習部21は、同様の動作を、教師データTD[c]の教師信号T[c]がT[c]=−1になるかまたはc=nofTD+1となるまで繰り返す(同図ステップ(リ))。T[c]=−1またはc=nofTDとなると、学習用肯定信号LSY[c]計算等の処理を終了する。
【0127】
次に、判定面学習部21は、内部の記憶素子に記憶された判定面学習用信号TC[c](c=1、・・・)は、横軸をLSN[c]、縦軸をLSY[c]とし、T[c]=1を○、T[c]=0を×で示すと、図11に示すような分布となる。これらのうち、教師信号T[c]=1であるものと前記教師信号T[c]=0であるものとが、図11に示したように最もよく分離できる判定パラメータCを、山登り法によって計算する(同図ステップ(ヌ))。次に前記判定パラメータCを判定パラメータ記憶部8に書き込み、学習制御部14に判定面学習制御信号PLCを送り(同図ステップ(ル))、処理を終了する。
【0128】
学習制御部14は、判定面学習部21から判定面学習制御信号PLCを受け、学習制御部指示信号を待機中を示す値にし、処理を終了する。
【0129】
図11に示したように、上述の2つのメトリック信号を用いてキーワード群信号を肯定信号SYと否定信号SNとで表される2次元空間上で、ユーザーが必要とする情報は主に左上に、不要な情報は右下に分布するようになる。したがって、上記ように適切な係数Cを用いて必要性信号をN=SY−C・SNとすれば、必要性信号は、ユーザーが必要とする情報に対して大きな値をとるようになる。
【0130】
以上のような動作により、肯定メトリック記憶部5と否定メトリック記憶部6には、使用者の興味を反映した行列(肯定メトリック信号MYと否定メトリック信号MN)が生成される。
【0131】
こうして生成された前記行列を適切に分割することにより、使用者ごとに適した興味の分類を実現することができる。具体的には、メトリック分割ユニット53は、前記肯定メトリック信号MYと前記否定メトリック信号MNとを読み出し、これを処理、分割した分類肯定メトリック信号ClassMYと分類否定メトリック信号ClassMNを生成し、分類情報フィルタユニット55−1、55−2内の分類肯定メトリック記憶部555と分類否定メトリック記憶部556に書き込む。
【0132】
図13は、メトリック分割ユニット53の構成を示すブロック図である。
図13において、54は分類にかかるパラメータを記憶する分類パラメータ記憶部であり、その記憶内容は分類パラメータ入力端子301からの入力によって書き換えることができる。
【0133】
同図において、531は分類パラメータ記憶部54から枝切閾値を読み出し肯定メトリック信号MYおよび否定メトリック信号MNのうち分類時にノイズとなりやすい成分を除く枝切り処理部である。532は枝切り処理をされた肯定メトリック信号である枝切り肯定メトリック信号CMYを記憶する枝切り肯定メトリック記憶部であり、533は枝切り処理をされた否定メトリック信号である枝切り否定メトリック信号CMNを記憶する枝切り否定メトリック記憶部である。534は前記枝切り肯定メトリック信号CMYと前記枝切り否定メトリック信号CMNとから枝切りメトリック信号CMを計算する枝切りメトリック合成部である。
【0134】
535は前記枝切りメトリック信号CMを用いて行う行列演算を制御する行列演算制御部であり、536は前記行列演算制御部535によって制御される行列演算部であり、537は前記行列演算部536への入力を記憶している行列演算入力記憶部である。538は前記行列演算の結果に基づき分類ラベルを成分とするベクトルを記憶する分類ラベル記憶部であり、539は前記分類ラベル記憶部538に記憶されたベクトルの成分と前記枝切り肯定メトリック信号CMY、前記枝切り否定メトリック信号CMNと分類パラメータ記憶部54から最大分類数を読み出し前記分類情報フィルタリングユニット55の分類肯定メトリック信号ClassMYと分類否定メトリック信号ClassMNとを設定する分類メトリック書き出し部である。
【0135】
以上の構成をとるメトリック分割ユニット53の動作について、図13から図17を参照しながら詳細に説明する。
【0136】
メトリック分割ユニット53は、学習開始信号入力端子106から学習開始信号LSが入力されると待機状態に入り、その後、学習制御部指示信号出力端子107に出力される学習制御部指示信号LIの値が1から0に変わり、学習ユニット52の動作が終了したことを示されると、図14から図17のフローチャートに示された以下の処理を開始する。
【0137】
図14は、枝切り処理部531の処理を示したフローチャートである。この処理は、肯定メトリック信号MYおよび否定メトリック信号MNの各成分のうち、あまり値の大きくないものを枝切りする処理である。これによって、行列で表現できる2つのメトリック信号は、互いに独立な部分行列の和として書けやすくなる。
【0138】
枝切り処理部531は、分類開始信号入力端子108から分類開始信号CStを受け取ると、分類パラメータ記憶部54から枝切閾値Nlowを読み出す(ステップ(イ))。なお、図2ならびに図3のように学習制御部指示信号LIを分類開始信号として用いることも好ましい。
【0139】
次に、行カウンタiと列カウンタjを0に設定する(ステップ(ロ)、ステップ(ハ))。そして、肯定メトリック記憶部5から肯定メトリック信号MYの(ij)成分MY[i][j]と否定メトリック記憶部6から否定メトリック信号MNの(ij)成分MN[i][j]とを読み出す(ステップ(ニ))。
【0140】
前記肯定メトリック信号MYの(ij)成分MY[i][j]と前記否定メトリック信号MNの(ij)成分MN[i][j]との和が前記枝切閾値Nlowとを比較する(ステップ(ホ))。前記和が、前記枝切閾値Nlowより大きければ、前記肯定メトリック信号MYの(ij)成分MY[i][j]を枝切り肯定メトリック信号CMYの(ij)成分CMY[i][j]に、前記否定メトリック信号MNの(ij)成分MN[i][j]を枝切り肯定メトリック信号CMNの(ij)成分CMN[i][j]に代入する(ステップ(へ))。
【0141】
前記和が、前記枝切閾値Nlowより小さければ、枝切り肯定メトリック信号CMYの(ij)成分CMY[i][j]および枝切り肯定メトリック信号CMNの(ij)成分CMN[i][j]に0を代入する(ステップ(へ))。
そして、列カウンタjを一つ増やす(ステップ(チ))。
【0142】
以下、同様の処理を、全ての成分に対して行う(ステップ(リ)〜(ル))。
【0143】
以上の枝切り処理が終了すると、枝切り処理部531は、枝切りメトリック合成部534にメトリック合成開始信号を送る。
【0144】
枝切りメトリック合成部534は、図15に記載したフローチャートにしたがって、枝切り肯定メトリック信号CMYと枝切り否定メトリック信号CMNを読み出し(ハ)、その和を計算し分類行列信号Classを生成し(二)、行列演算部536に代入する。
【0145】
そして、その処理が終了すると、行列演算制御部535に行列演算開始信号を送る。
【0146】
行列演算制御部535は、前記行列演算開始信号を受けると、前記分類行列信号Classによって独立に変換されるベクトルを検出する処理を行う。処理の詳細について、図16に示したフローチャートを参照しながら説明する。この一連の処理の目的は、分類行列によって互いに変換され得る入力ベクトルの成分を特定することにある。その意味は、互いに変換され得る入力ベクトルの成分には対応するキーワードが相互に関係あるとみなすことができる。したがって、互いに変換され得る入力ベクトルの成分をいくつかのグループに分割することは、使用者の興味をいくつかに分類することに対応すると見なすことができる。
【0147】
まず、行列演算制御部535は、分類ラベル記憶部538に記憶された分類ラベルベクトルCatVを0ベクトルに初期化する(ステップ(イ))。次に、カテゴリ信号Catに最初のカテゴリであることを示す値1を代入する(ステップ(ロ))。入力ベクトルを決定するに先立ち行列演算記憶部537に記憶された入力ベクトルCVを0ベクトルに初期化する(ステップ(ハ))。そして、成分カウンタiを0に設定する(ステップ(ニ))。
【0148】
成分カウンタiによって指定される分類ラベルベクトルCatVの第i成分CatV[i]が0に等しいか否かを判定する(ステップ(ホ))。等しい場合には、入力ベクトルCVの第i成分CV[i]に1を代入する(ステップ(へ))。これによって、ただ一つの成分が1であり、他の成分はすべて0という初期入力ベクトルが得られる。この入力ベクトルに分類行列信号Classを掛け、変換ベクトルCVtmpを得る(ステップ(ト))。変換ベクトルCtmpの全ての成分に対し、
Ctmp[i]= Θ(Ctmp[i])
という変換を行う(ステップ(チ)から(ル))。ここで、関数Θは、
Θ(x)= 1 x>0
0 その他の場合
という関数である。
【0149】
この成分が1または0である変換ベクトルCVtmpと入力ベクトルCVと比較する(ステップ(ヲ))。変換ベクトルCVtmpと入力ベクトルCVが異なる場合は、変換ベクトルCVtmpを入力ベクトルCVに代入し、ステップ(ト)に戻る(ステップ(ワ)。 変換ベクトルCVtmpと入力ベクトルCVが等しい場合は、分類ラベルベクトルCatVを
CatV=CatV+Cat・CV
と変更する(ステップ(カ))。そして、カテゴリ信号Catを1だけ増やし、ステップ(ハ)に戻る(ステップ(ヨ))。
【0150】
そして、分類ラベルベクトルのすべての成分の値が0でなくなると、処理の流れがステップ(ホ)から終了へ分岐する。
【0151】
以上の処理により、分類ラベルベクトルCatVは、枝切り肯定メトリック信号CMYと枝切り否定メトリック信号CMNとから作られた分類行列信号Classが4つの独立な行列からなる場合、例えば、
初期状態 CatV=(0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)
Cat=1 CatV=(1,0,1,0,0,0,0,0,1,0,1,0,0,0,0)
Cat=2 CatV=(1,2,1,0,2,0,2,0,1,0,1,2,2,0,0)
Cat=3 CatV=(1,2,1,3,2,3,2,0,1,3,1,2,2,0,3)
Cat=4 CatV=(1,2,1,3,2,3,2,4,1,3,1,2,2,4,3)
というように変化する。
【0152】
こうして、互いに変換され得る入力ベクトルの成分をいくつかのグループに分割することにより、使用者の興味をいくつかに分類することができる。次の処理として、この結果を用いて、分類情報フィルタリングユニット55を設定する。
【0153】
上のように分類ラベルベクトルCatVが得られると、行列演算制御部535は、分類メトリック書き出し部539に書き出し開始信号を送る。
【0154】
分類メトリック書き出し部539は、前記書き出し開始信号を受けて、図17のフローチャートに示された処理を開始する。この一連の処理の目的は、枝切り肯定メトリック信号CMYと枝切り否定メトリック信号CMNを、分類ラベルベクトルに基づいて、最大分類数NCmax以下の数に分割し、それぞれを分類情報フィルタリングユニット55の分類肯定メトリック記憶部555と分類否定メトリック記憶部556に書き込むことにある。
【0155】
まず、分類ラベルベクトルを読み出す(ステップ(イ))。次に、分類パラメータ記憶部54から最大分類数NCmaxを読み出す(ステップ(ロ))。図2に示した構成の場合、分類情報フィルタリングユニットが2個であるので、最大分類数NCmaxは2とすることが望ましい。次に、NCmax個の分類情報フィルタリングユニット55−1〜NCmaxの分類肯定メトリック信号ClassMY,分類否定メトリック信号ClassMNすべてを零行列に初期化する(ステップ(ハ))。
【0156】
行カウンタiを0にする(ステップ(ニ))。列カウンタjを0にする(ステップ(ホ))。そして、分類ラベルベクトルの第i成分CatV[i]が最大分類数NCmax以下であるか否かを判断する(ステップ(ヘ))。以下であれば、ステップ(ト)に進み、より大きければ、ステップ(ヲ)に進む。ステップ(ト)では、分類ラベルベクトルの第i成分CatV[i]と第j成分CatV[j]とを比較する。
【0157】
ステップ(ト)で、分類ラベルベクトルの第i成分CatV[i]と第j成分CatV[j]とが等しかった場合、これは2つの成分は、分類行列Classによって互いに変換される、つまり独立していないことを意味するから、枝切り肯定メトリック信号CMYの(ij)成分CMY[i][j]と枝切り否定メトリック信号CMNの(ij)成分CMN[i][j]とを読み出す(ステップ(チ))。そして、分類番号CatV[i]の分類情報フィルタリングユニット55−CatV[i]の分類肯定メトリック信号ClassMYの(ij)成分に前記枝切り肯定メトリック信号の(ij)成分を代入し、分類否定メトリック信号ClassMNの(ij)成分に前記枝切り否定メトリック信号の(ij)成分を代入する(ステップ(リ))。
【0158】
一方、ステップ(ト)において、2つの成分が等しくない場合、2つの成分、分類行列Classによって互いに変換されない、つまり2つの成分は独立していることを意味する。ステップ(チ)からステップ(リ)に処理を省略する。
【0159】
そして、列カウンタjを一つ増やし(ステップ(ヌ))、列カウンタjの値が分類ラベルベクトルの次元NofDCKと比較する(ステップ(ル))。列カウンタjの値が、分類ラベルベクトルの次元より小さければステップ(ト)に戻る。列カウンタjの値が、分類ラベルベクトルの次元以上であれば、ステップ(ヲ)に進む。
【0160】
ステップ(ヲ)では、行カウンタiの値を1だけ増やす。行カウンタiの値が分類ラベルベクトルの次元NofDCKと比較する(ステップ(ワ))。行カウンタiの値が、分類ラベルベクトルの次元より小さければステップ(ホ)に戻る。行カウンタiの値が、分類ラベルベクトルの次元以上であれば、処理を終了する。
【0161】
以上の処理により、分類情報フィルタリングユニット55はすべて適切に設定(分類肯定メトリック信号と分類否定メトリック信号がともに零行列ある場合を含む)される。
【0162】
各々の分類情報フィルタリングユニット55の動作は、情報フィルタリングユニット51の動作と同様であるので、説明を省略する。
【0163】
以上、本発明の説明に先立って、本発明への入力となるユーザの興味を反映したキーワード、枝切り肯定メトリック信号、枝切り否定メトリック信号、分類ラベルベクトルを生成する情報フィルタ装置について説明した。
【0164】
本発明のキーワードの可視化方法およびその装置は、こうした情報フィルタ装置で生成されたユーザの興味を反映したキーワード、枝切り肯定メトリック信号、枝切り否定メトリック信号、分類ラベルベクトルなどを読み込み、ユーザに理解し易いかたちでキーワードを表示することにある。
【0165】
ここで、キーワードそのものは文字の並びであって、形状、大きさ、色などを本来は有していないことに注意されたい。このため、本発明では、キーワード毎に、形状および大きさおよび色を持ったキーワード体を与える。つまり、キーワードに代わるキーワード体の配置位置を決定し、そのキーワード体の配置位置を利用して、キーワードを表示することになる。
【0166】
以下では、図1のように構成された可視化システムにおける可視化装置1000の動作を、図18のフローチャートを用いて説明する。ここで、全キーワード体の配置座標をひとまとめにした集合を配置Sと呼び、2次元空間に配置する場合、キーワード体i(=1,...,nofDCK)の配置座標(中心座標)を(sx(i)、sy(i))で表すと、配置S={sx(i)、sy(i)|i=1、...、nofDCK}と書くことができる。なお、以下では、2次元空間に配置する場合を説明するが、3次元空間に配置する場合にも容易に拡張できる。
【0167】
まず、情報フィルタ装置1007内の符号辞書記憶部2に記憶されているユーザの興味を反映したキーワードi(=1,...,nofDCK)をネットワーク1009を介して読み込む(イ)。ここで、以下の説明を簡単にするため、キーワードは、ユーザーの興味の反映度合いが強い順に並んでいるものとする。これは、例えば、特開平9−288683号公報に開示されているキーワードiのキーワードコスト信号KD(i)の値(ユーザーの興味の反映度合いが強いキーワードほど大きな値をとる)を利用すれば容易に実現できるため、一般性を失わない。
【0168】
次に、情報フィルタ装置1007内の分類ラベル記憶部538から分類ラベルベクトルCatVをネットワーク1009を介して読み込む(ロ)。キーワードiには、分類ラベルベクトルCatVの第i成分である分類ラベルCatV(i)が設定される。
【0169】
次に、情報フィルタ装置1007内の枝切り肯定メトリック記憶部532から枝切り肯定メトリック信号CMYをネットワーク1009を介して読み込む(ハ)。キーワードiとjとのペアには、枝切り肯定メトリック信号の第(i、j)成分である枝切り肯定メトリック要素信号CMY(i,j)が設定される。
【0170】
次に、情報フィルタ装置1007内の枝切り否定メトリック記憶部532から枝切り否定メトリック信号CMNをネットワーク1009を介して読み込む(ニ)。キーワードiとjとのペアには、枝切り否定メトリック信号の第(i,j)成分である枝切り否定メトリック要素信号CMN(i,j)が設定される。
【0171】
次に、配置領域(この配置領域内にキーワード体を配置する)、配置S、温度T、キーワード体などの初期設定を行なう(ホ)。配置Sの初期化に関しては、例えば、全キーワード体の配置座標を配置領域内でランダムに設定した初期配置や、分類ラベルが同じキーワードのキーワード体の配置座標を配置領域内で近くに設定した初期配置を初期の配置Sとすればよい。また、温度Tの初期化に関しては、0より大きな適当な値を与えればよい。この他、後述のコスト関数F(S)のパラメータや、後述の温度更新係数βなどの初期化も行なう。
【0172】
また、キーワード体の初期設定としては、形状、大きさ、色を決定する。上述のとおり、キーワードは、ユーザーの興味の反映度合いが強い順に並んでいるものとしているため、例えば、キーワード体iの形状は正(i+2)角形などとすれば、キーワードが重要でなくなればなくなるほど、そのキーワードのキーワード体は円形に近づくことになり、ユーザはキーワードの重要性を一目で理解できることになる。
【0173】
また、たとえ、キーワード体の形状を一定にしても、大きさを変化させることで、ユーザはキーワードの重要性を一目で理解できることになる。たとえば、キーワード体iの大きさを、(キーワード体(i−1)の大きさ)>(キーワード体iの大きさ)>(キーワード体(i+1)の大きさ)の関係を満たすように設定すれば、理解のしやすさが高まることになる。さらに、たとえ、キーワード体の形状や大きさを一定にしても、色を変化させることで、ユーザはキーワードの重要性を一目で理解できることになる。もちろん、分類ラベルが同じキーワードのキーワード体を同一色にすることなども考えられ、この場合のユーザの利点は、同じ分野のキーワードが一目で理解できることにある。以上のように、キーワード体の形状および/または大きさおよび/または色を、ユーザの興味の反映度合いを用いて決定することは、ユーザの理解を助けるためにきわめて重要である。
【0174】
次に、カウンタcountを0セットし(ヘ)、配置Sを更新する(ト)。配置Sの更新には、まず、キーワード体またはキーワード体群を選択し、次に、選択されたキーワード体またはキーワード体群の配置座標を配置領域内で変更することで配置候補S’を生成し、最後に、配置Sのコストと配置候補S’のコストとの比較を通じて配置候補S’を次の時刻の配置Sとするかどうかを決定すればよい。
【0175】
配置Sを評価するコスト関数F(S)としては、最も簡単には、例えば、(数6)とすればよい。
【0176】
【数6】
Figure 0003799905
【0177】
ここで、A、Bは正の定数である。また、d(i,j,S)は配置Sにおけるキーワード体iとjとの間の距離である。さらに、o(i,j,S)は、配置Sにおいて、キーワード体iとjとの間に重なりがある場合に1の値を、そうでない場合に0の値をとる関数である。こうしたコスト関数F(S)を利用して、例えば、(数7)に示すように以下の確率で配置候補S’を次の時刻の配置Sとすればよい。
【0178】
【数7】
Figure 0003799905
【0179】
ここで、Tは上記のとおり温度パラメータである。
【0180】
キーワード体群の決定については、たとえば、分類ラベルが同じキーワードのキーワード体を一つのキーワード体群とすることが考えられる。
【0181】
さらには、キーワード体iとjとの間の距離d(i,j,S)を利用して、キーワード体群の決定を行なうこともできる。例えば、(CMY(i,j)+CMN(i,j))が0より大きなキーワード体iとjに対し、d(i,j,S)の時間変化H(i,j)が小さい場合、つまり、キーワード体iとjの配置位置が安定している場合、それらをキーワード体群とすればよい。ここで、d(i,j,S)の時間変化H(i,j)としては、例えば、(数8)で計算できる。
【0182】
【数8】
Figure 0003799905
【0183】
ここで、εは正の定数である。また、Sは現在の配置を、S’は次の時刻の配置((数7)で受理された配置候補)を示す。
【0184】
このH(i,j)がある閾値以下である場合に、キーワード体iとjとで一つのキーワード体群を構成する。また、これらキーワード体群のいくつかをひとまとめにした新たなキーワード体群も構成できる。具体的には、2つのキーワード体群に共通のキーワード体が含まれる場合、それら2つのキーワード体群をひとまとめにして新たなキーワード体群を構成し、これを繰り返すことで実現される。
【0185】
以上のように構成されたキーワード体群を単位として配置座標を変更する場合には、そのキーワード体群に含まれるキーワード体間の相対的な配置座標を保存することとする。これにより、高速に配置修正が行なわれ、短時間で良好な配置結果を得ることができる。
【0186】
なお、配置Sがどのように変化しているかを確認するためには、更新後の配置Sをディスプレイ1003に表示すればよい。
【0187】
ステップ(ト)での配置Sの更新に引き続き、ステップ(チ)では、カウンタcountに1を加える。
【0188】
次に、ステップ(リ)では、count<nofDIKであれば(Y)、ステップ(ト)に戻り、そうでなければ(N)、ステップ(ヌ)に進む。
【0189】
ステップ(ヌ)では、終了条件を満たしているかどうかを判断する。例えば、配置SのコストF(S)の変化が小さくなった場合に終了すればよい。
【0190】
また、上記カウンタcountのような、カウンターを設けておき、繰り返しの最大回数をあらかじめ設定し、カウンターがその最大回数を超えた時点で終了してもよい。
【0191】
ステップ(ヌ)で終了条件を満たしていない場合は(N)、ステップ(ル)に進み、温度Tを更新する。例えば、温度更新定数βを0<β<1の範囲で設定し、(数9)と更新することなどが考えられる。
【0192】
【数9】
Figure 0003799905
【0193】
ステップ(ヌ)で終了条件を満たした場合には(Y)、ステップ(ヲ)に進み、配置Sを配置データ記憶部1004に記憶し、必要ならばFD1005にも記憶して終了する。なお、配置Sがどのようになったかを確認するためには、つまり、配置結果を確認するためには、配置Sをディスプレイ1003に表示すればよい。
【0194】
以上のキーワードの可視化方法を、ユーザーの興味を反映したキーワードを含む例題に適用した。図19は初期配置(簡単のため、ランダム配置とした)を示し、図20は配置結果を示す。図中の正方形はキーワード体であり、キーワード体上の単語がキーワードである。簡単のため、ここでは、キーワード体の形状を正方形とし、大きさは一定とした。
【0195】
また、図中のキーワード体を結ぶ直線は、キーワード間の関係を示している。配置結果である図20から、関係あるキーワード体がクラスタを形成するように配置されていることがわかる。
(実施の形態2)
以下では、本発明の実施の形態2におけるキーワード可視化システムの一例を、図21のブロック図を用いて説明する。なお、実施の形態1での説明と同様に、2次元空間に配置する場合を説明するが、3次元空間に配置する場合にも容易に拡張できる。また、図21においては、図1と共通で説明上特に記す必要のないブロック(例えばCPU1002など)ならびにバス線は、図面が煩雑にならないようにと、省略してある。
【0196】
図21において、2100はユーザの興味を反映した情報(前記ユーザの興味を反映したキーワード、前記分類ラベルベクトル、前記枝切り肯定メトリック信号、前記枝切り否定メトリック信号)を生成する情報フィルタ装置で、前記ユーザの興味を反映したキーワードを記憶する符号辞書記憶部2、前記分類ラベルベクトルを記憶する分類ラベル記憶部538、前記枝切り肯定メトリック信号を記憶する枝切り肯定メトリック記憶部532、前記枝切り否定メトリック信号を記憶する枝切り否定メトリック記憶部533より構成される。
【0197】
2101は前記ユーザの興味を反映したキーワードをディスプレイ上に見やすく配置するための可視化装置である。2102は後記の初期設定部2103で設定されるパラメータ値(デフォルト値)をユーザが変更できるようにするための設定入力部、2103は配置領域(この領域内にキーワード体を配置する)の属性(大きさ、形状、色)、コスト関数の係数、初期温度や温度変更係数など、可視化装置2101の動作に必要となる種々のパラメータ値をデフォルト値または設定入力部2102から入力された値に設定するための初期設定部、2104は符号辞書記憶部2からのユーザーの興味を反映したキーワード、分類ラベル記憶部538からの分類ラベルベクトルを受け、キーワード体を生成するキーワード体生成部である。
【0198】
2105はキーワード体生成部2104からのキーワード体、初期設定部2103からの配置領域を受け、キーワード体の初期配置を生成し、かつ、後記コスト計算部2107で計算される前記初期配置のコストを保持する初期配置生成部、2106は枝切り肯定メトリック記憶部532からの枝切り肯定メトリック信号と枝切り否定メトリック記憶部533からの枝切り否定メトリック信号とを受け、枝切りメトリック信号を生成する枝切りメトリック信号生成部、2107は初期設定部2103からのコスト関数の係数、枝切りメトリック信号生成部2106からの枝切りメトリック信号、初期配置生成部2105や後記配置候補生成部2109からの配置Sを受け、そのコストを計算するコスト計算部、2108は現時点での配置とそのコストを保持する配置保持部、2109は配置保持部2108で保持されている配置を変更して配置候補を生成し、かつ、コスト計算部2107で計算される前記配置候補のコストを保持する配置候補生成部である。
【0199】
2110は配置保持部2108からの現時点での配置とそのコストおよび配置候補生成部2109からの配置候補とそのコストとを受け、さらに、後記温度更新部2111で保持される温度を受け、次の時刻の配置を決定し、その配置とコストとを配置保持部2108に送る配置更新決定部であり、2111は初期設定部2103からの初期温度や温度変更係数、配置更新決定部2110での配置更新回数などを受け、温度を更新・保持する温度更新部、2112は配置保持部で保持される配置を配置データ記憶部1004に出力する配置出力部である。
【0200】
以下では、各部の詳細を説明していくが、情報フィルタ装置2100内の符号辞書記憶部2、分類ラベル記憶部538、枝切り肯定メトリック記憶部532、枝切り否定メトリック記憶部533については、実施の形態1と同様であるため、ここでは省略することとする。
【0201】
設定入力部2102は、初期設定2103で設定されるパラメータ値(デフォルト値)をユーザが変更できるようにするためのもので、キーボード、マウス、音声マイクなどが利用できる。
【0202】
初期設定部2103は、可視化装置2101の動作に必要となる種々のパラメータの値を設定する。各パラメータの値には、設定入力部2102でユーザーが設定したパラメータ値を用いる。ユーザーによる設定がなされていないパラメータには、デフォルト値が設定される。初期設定部2103で設定されるパラメータとしては、初期配置生成部2105や配置候補生成部2109で必要となる配置領域(この領域内にキーワード体を配置する)の属性(大きさ、形状、色)、コスト計算部2107で必要となるコスト関数の係数、温度更新部2111で必要となる初期温度や温度変更係数などがある。
【0203】
キーワード体生成部2104は、符号辞書記憶部2からのユーザーの興味を反映したキーワード、分類ラベル記憶部538からの分類ラベルベクトルCatVを受け、キーワード体を生成する。符号辞書記憶部2から読み込んだ各キーワードi(i=1,...,nofDCK)毎に1つのキーワード体iが生成されることになる。例えば、キーワード体の形状を円形とする場合には、各キーワード体iの形状を、キーワードiの重要性(キーワードの重要性を評価するには、例えば、特開平9−288683号公報に開示されているキーワードiのキーワードコスト信号KD(i)の値(ユーザーの興味の反映度合いが強いキーワードほど大きな値をとる)を利用できる)を反映した半径を持つ円形にすればよい。さらには、キーワード体iの分類ラベルCatV(i)(分類ラベルベクトルCatVの第i要素)を利用して、分類ラベルが同じキーワード体群毎に色を割り当てることもできる。
【0204】
初期配置生成部2105は、初期設定部2103で設定された配置領域と、キーワード体生成部2104で生成したnofDCK個のキーワード体i(i=1,...,nofDCK)を受け、配置Sを初期化する。なお、キーワード体i(=1,...,nofDCK)の配置座標(中心座標)を(sx(i)、sy(i))で表すと、配置SはS={sx(i)、sy(i)|i=1、...、nofDCK}と書くことができる。ここで、sx(i)とsy(i)はそれぞれキーワード体iのX座標値とY座標値である。具体的な配置Sの初期化方法としては、キーワード体iが配置領域内に収まるようにsx(i)とsy(i)をランダムに設定するなどが考えられる。また、分類ラベルが同じキーワード体群がまとまるように配置Sを初期設定してもよい。このようにして生成された初期配置Sはコスト計算部2107に送られ、そのコストF(S)が計算される。初期配置生成部2105では、初期配置Sに加え、そのコストF(S)も保持することとする。
【0205】
枝切りメトリック信号生成部2106は、枝切り肯定メトリック記憶部532からの枝切り肯定メトリック信号CMY(i,j)(i,j=1,...,nofDCK)と枝切り否定メトリック記憶部533からの枝切り否定メトリック信号CMN(i,j)(i,j=1,...,nofDCK)とを受け、枝切りメトリック信号CM(i,j)(i,j=1,...,nofDCK)を生成する。具体的には、枝切りメトリック信号CM(i,j)を枝切り肯定メトリック信号CMY(i,j)と枝切り否定メトリック信号CMN(i,j)との要素毎の和などとすればよい。
【0206】
コスト計算部2107は、初期設定部2103からのコスト関数の係数、枝切りメトリック信号生成部2106からの枝切りメトリック信号CM(i,j)(i,j=1,...,nofDCK)を受け、コスト関数F(S)を生成する。CM(i,j) =CMY(i,j)+CMN(i,j)である場合には、コスト関数F(S)を(数6)のように生成すればよい。(数6)のコスト関数F(S)における係数A,Bの値は初期設定部2103で設定された値を用いることとする。コスト計算部2107は、このように生成されたコスト関数F(S)を用いて、初期配置生成部2105や後記配置候補生成部2109から入力される配置SのコストF(S)を計算し、出力する機能も持っている。
【0207】
配置保持部2108は現時点での配置SとそのコストF(S)を保持する機能を持つ。したがって、配置保持部2108は、まず最初(時刻0)に初期配置生成部2108で生成された初期配置SとそのコストF(S)を保持し、以降は後述の配置更新決定部2110から出力される配置SとそのコストF(S)を保持することになる。
【0208】
配置候補生成部2109は、配置保持部2108で保持されている配置Sを受け、それを変更して配置候補S’を生成する。配置Sの変更の仕方としては、例えば、あるキーワード体iを選択し、そのsx(i)やsy(i)に乱数を加えるなどで実現できる。また、複数のキーワード体(キーワード体群)を選択し、同様の操作を行なうことも考えられる。配置候補生成部2109は、こうして生成された配置候補S’と、そのS’をコスト計算部2107に送ることで得られるコストF(S’)を保持する機能を持つ。
【0209】
配置更新決定部2110は、配置保持部2108で保持されている配置SとそのコストF(S)、配置候補生成部2109で保持されている配置S’とそのコストF(S’)、さらには、後述の温度更新部で保持される温度Tを受け、次の時刻の配置Sを決定する。具体的には、例えば、(数7)の確率で、配置候補S’を次の時刻の配置Sとし、残りの確率で、現在の配置Sを次の時刻の配置Sとすればよい。
【0210】
配置更新決定部2110で決定された次の時刻の配置SとそのコストF(S)は、配置保持部2108で現在の時刻の配置SとそのコストF(S)として保持される。つまり、配置更新決定部2110から配置保持部2108へのフィードバックによりループが形成され、配置Sが時々刻々と変化することになる。
【0211】
温度更新部2111は初期設定部2103からの初期温度Tや温度変更係数βを受け、現在の温度Tを更新・保持する。例えば、配置更新決定部2110での配置更新回数をカウントし、カウント数があらかじめ決められた閾値以上になったとき、(数9)にしたがって温度Tを更新すればよい。
【0212】
配置出力部2112は配置保持部2108で保持される配置Sをディスプレイやハードディスクなどに出力するために配置データ記憶部1004へと配置データを書き込む。
【0213】
以上の構成により、ユーザーの興味を反映したキーワードについて、関係のあるキーワードのキーワード体を情報の分野毎にまとめて配置、表示することができる。
【0214】
(実施の形態3)
以下では、本発明の実施の形態3における情報可視化編集システムについて、以下に説明する。情報を可視化してその結果をユーザの求めに応じて編集することができる情報可視化編集システムの一例を、図22のブロック図と配置表示の好ましい例を示した図23とを用いて説明する。なお、実施の形態1、実施の形態2での説明と同様に、2次元空間に配置する場合を説明するが、3次元空間に配置する場合にも容易に拡張できる。また、図22においても、図21と同様に図1と共通で説明上特に記す必要のないブロック(例えばCPU1002など)ならびにバス線は、図面が煩雑にならないようにと、省略してある。
【0215】
図22において、2201は配置Sに含まれるキーワード体(ならびにその集合体)を切断する切断線DLを入力する切断線入力部であり、2202は、前期切断線DLと配置データ記憶部1004から配置Sとを受け配置Sをいくつかの部分に切断した切断配置DSを計算する配置データ切断部であり、2203は配置Sに含まれるキーワード体(ならびにその集合体)を接続する接合線CLを入力する接合線入力部であり、2204は前記接合線CLを受け、前記キーワード体(ならびにその集合体)を接続した配置Sを計算する配置接合部であり、その他のブロックは、実施の形態1ならびに実施の形態2に記載したものと同じである。さらに、キーワード体とキーワード体とを結ぶ直線をここでは接続線と呼ぶ。
【0216】
なお、キーワード体(ならびにその集合体)を編集するための編集入力は、前記切断線DLと前記接合線CLに限るものではない。切断個所、接合個所を示すものは一定の領域を設定する図形であっても構わない。
【0217】
以下、以上のような構成をとった可視化装置2101の動作について説明する。
【0218】
実施の形態2で説明にしたがって、配置データ記憶部1004に配置Sが書き込まれる。配置Sはモニタ1003に、図23に示したような形式で出力される。ユーザは、モニタ1003の出力を見ながら、切断線入力部2201を操作して図24に示したような配置Sを切断する切断線DLを入力する。コンピュータのマウスは、前記切断線入力部2201ならびに前記接合線入力部2203の好ましい例の一つである。前記切断線DLを受けた配置データ切断部2202は、前記配置Sの接続線の内、前記切断線DLと交差する接続線CL(id,jd)を検出し、枝切り肯定メトリック記憶部532に記憶された枝切り肯定メトリック信号CMYの(id,jd)成分CMY(id,jd)と(jd,id)成分CMY(jd,id)とを0に設定する。この処理をすべての交差する接続線に対して行う。これによって、図25の上部に示したように、図23では一塊であったキーワード体の集団が2つに分割された配置S1、S2を得ることができる。これは、一つのカテゴリを2つのカテゴリに分割することに対応しており、実施の形態1に記載したメトリック合成処理以下の処理を行うことにより、分割された配置S1、S2に対応する新たな分類ラベルベクトルを得られる。
【0219】
また、ユーザは、モニタ1003の出力を見ながら、接合線入力部2203を操作して図24に示したような配置Sの部分を接続する接合線CLを入力する。前記接合線CLを受けた配置データ接合部2204は、前記配置Sのキーワード体の内、前記接合線CLによって結ばれたキーワード体icとキーワード体jcを検出し、枝切り肯定メトリック記憶部532に記憶された枝切り肯定メトリック信号CMYの(ic,jc)成分CMY(ic,jc)と(jc,ic)成分CMY(jd,id)とを1に設定する。
【0220】
この処理をすべての接続されたキーワード体に対して行う。これによって、図25左下に示したように、図23では2つであったキーワード体の集団が1つに接合された配置S3を得ることができる。これは、1つのカテゴリを2つのカテゴリに分割することに対応しており、実施の形態1に記載したメトリック合成処理以下の処理を行うことにより、接合された配置S3に対応する新たな分類ラベルベクトルを得られる。
【0221】
以上のように、配置Sに含まれるキーワード体(ならびにその集合体)を切断する切断線DLを入力する切断線入力部と、前記切断線DLと配置データ記憶部から配置Sとを受け配置Sをいくつかの部分に切断した切断配置DSを計算する配置データ切断部と、配置Sに含まれるキーワード体(ならびにその集合体)を接続する接合線CLを入力する接合線入力部と、前記接合線CLを受け、前記キーワード体(ならびにその集合体)を接続した配置Sを計算する配置接合部を設けることにより、可視化された分類結果をユーザの手で編集できる情報可視化分類装置を実現できる。
【0222】
【発明の効果】
以上のように本発明は、情報が必要か否かのユーザ入力によりユーザの興味を獲得する情報フィルタ装置で得られたキーワードと、枝切り肯定メトリック信号および枝切り否定メトリック信号の少なくとも一方を受け、前記キーワードの各々に対し、ユーザの興味を反映した形状および大きさおよび色を持ったキーワード体を設定し、前記キーワード体を配置して、ディスプレイ上に前記キーワード体と前記キーワードを表示するようにしたことで、前記キーワードのキーワード体の形状、大きさおよび色の少なくとも1つを、ユーザの興味の反映度合いを用いて決定することができる上、前記枝切り肯定メトリック信号および枝切り否定メトリック信号の少なくとも一方を利用して、関係のあるキーワードのキーワード体を情報の分野毎にまとめて配置でき、さらに、ディスプレイ上に前記キーワード体と前記キーワードを理解し易く表示することができるキーワードの可視化方法を提供することができる。
【図面の簡単な説明】
【図1】本発明の実施の形態1のキーワード可視化システムのブロック図
【図2】本発明の実施の形態1のキーワード可視化システムで用いる情報フィルタ装置のブロック図
【図3】本発明の実施の形態1のキーワード可視化システムで用いる情報フィルタ装置の概略を示すブロック図
【図4】同情報フィルタ装置におけるメトリック学習を行なう部分の構成を示すブロック図
【図5】同情報フィルタ装置のベクトル生成部の動作を説明するフローチャート
【図6】同情報フィルタ装置の未読データ書き込み制御部の動作を説明するフローチャート
【図7】同情報フィルタ装置の分類インタフェースユニットの動作を説明するフローチャート
【図8】同情報フィルタ装置の学習制御部の動作を説明するフローチャート
【図9】同情報フィルタ装置のメトリック学習部の動作を説明するフローチャート
【図10】同情報フィルタ装置の判定面学習部の動作を説明するフローチャート
【図11】同情報フィルタ装置の判定面学習部の動作を説明するための図
【図12】同情報フィルタ装置の判定面学習部の動作を説明するための図
【図13】同情報フィルタ装置のメトリック分割ユニットのブロック結線図
【図14】同情報フィルタ装置のメトリック分割ユニットの枝切り処理部の動作を説明するフローチャート
【図15】同情報フィルタ装置のメトリック分割ユニットの枝切りメトリック合成部の動作を説明するフローチャート
【図16】同情報フィルタ装置のメトリック分割ユニットの行列演算制御部の動作を説明するフローチャート
【図17】同情報フィルタ装置のメトリック分割ユニットの分類メトリック書き出し部の動作を説明するフローチャート
【図18】本発明の実施の形態1のキーワード可視化システムの動作を説明するフローチャート
【図19】本発明のキーワード可視化システムの初期配置例を示す図
【図20】本発明のキーワード可視化システムによる配置結果例を示す図
【図21】本発明の実施の形態2のキーワード可視化システムの構成を示すブロック図
【図22】本発明の実施の形態3の情報可視化編集システムの構成を示すブロック図
【図23】本発明の実施の形態3によって可視化されたキーワード配置の例を示す図
【図24】本発明の実施の形態3によって可視化されたキーワード配置ならびに切断線と接合線の入力例を示す図
【図25】本発明の実施の形態3によって編集されたキーワード配置の例を示す図
【符号の説明】
1 ベクトル生成部
2 辞書記憶部
3 スコア計算部
5 肯定メトリック記憶部
6 否定メトリック記憶部
7 必要性計算部
8 判定パラメータ記憶部
9 未読データ書き込み制御部
10 未読データ記憶部
11 未読データ出力制御部
12 教師データ制御部
13 教師データ記憶部
14 学習制御部
16、17、18 スイッチ
19 メトリック学習部
20 学習用ベクトル生成部
21 判定面学習部
22 スコア計算部
34 データ表示装置
50 情報フィルタリングユニット
51 インタフェースユニット
52 学習ユニット
53 メトリック分割ユニット
54 分類パラメータ記憶部
55 分類情報フィルタリングユニット
56 分類未読データ記憶部
60 データベース記憶部
61 データベース読み出し部
62 データベース再構築制御部
63 スイッチ
64 適応データベース書き込み部
65 適応データベース記憶部
70 分類インタフェースユニット
100 情報入力端子
101 キーワード数信号入力端子
102 キーワード信号入力端子
103 データ読み出し開始信号入力端子
105 教師信号入力端子
106 学習開始信号入力端子
107 学習制御部指示信号出力端子
108 分類開始信号入力端子
110 未読データ部指示端子
532 枝切り肯定メトリック記憶部
533 枝切り否定メトリック記憶部
538 分類ラベル記憶部
1000 可視化装置
1001 メモリ
1002 CPU
1003 ディスプレイ/キーボード
1004 配置データ
1005 FD
1006 I/F
1007 情報フィルタ装置
1008 I/F
1009 ネットワーク
2100 情報フィルタ装置
2101 可視化装置
2102 設定入力部
2103 初期設定部
2104 キーワード体生成部
2105 初期配置生成部
2106 枝切りメトリック信号生成部
2107 コスト計算部
2108 配置保持部
2109 配置候補生成部
2110 配置更新決定部
2111 温度更新部
2112 配置出力部
2201 切断線入力部
2202 配置データ切断部
2203 接合線入力部
2204 配置データ接合部[0001]
BACKGROUND OF THE INVENTION
The present invention can arrange and display related keywords in an easy-to-understand manner for the user based on information obtained by the information filter device that acquires the user's interest by whether or not information is necessary. The present invention relates to a method and a device for visualizing possible keywords.
[0002]
[Prior art]
With the progress of the social infrastructure of information communication in recent years, development of an information filter device (Japanese Patent Laid-Open No. 9-288683) is being promoted as a technology to cope with the large-scale information communication network and the significant increase in communication volume. . This is because the amount of information that can be accessed by an individual exceeds the amount of information that can be processed by the individual today. For this reason, an information filter device that can appropriately retrieve information that the user thinks necessary from a large amount of information has attracted attention.
[0003]
[Problems to be solved by the invention]
For example, the information filter device disclosed in Japanese Patent Laid-Open No. 9-288683 automatically acquires a keyword reflecting the user's interest, calculates a score reflecting the user's interest in the information based on the keyword, and obtains information in descending order of the score. Is presented.
[0004]
However, there is a problem that the user himself / herself cannot confirm a keyword reflecting the user's automatically acquired interest.
[0005]
The present invention solves the above-described conventional problems, and receives at least a keyword having a degree of reflection of the user's interest as an input, and has a keyword body having a shape, size and color for each of the keywords. A keyword visualization method comprising: setting, arranging the keyword body, and displaying the keyword body and the keyword on a display, wherein the keyword that the user is interested in is classified for each field of information. An object of the present invention is to provide a visualization device for keywords to be arranged and presented together.
[0006]
[Means for Solving the Problems]
In order to achieve this object, the keyword visualization method of the present invention includes a keyword obtained by an information filter device that acquires a user's interest by a user input as to whether or not information is necessary, a defragmentation positive metric signal, and / or A keyword body having a shape, size, and color reflecting the user's interest is set for each of the keywords upon receiving a debranching negative metric signal, the keyword body is arranged, and the keyword body is displayed on a display. And the keywords are displayed.
[0007]
According to the present invention, the shape and / or size and / or color of the keyword body of the keyword can be determined using the degree of reflection of the user's interest, and the relationship between the keywords is utilized. The keyword bodies of related keywords can be arranged together for each field of information, and the keyword bodies and the keywords can be displayed on the display in an easily understandable manner.
[0024]
DETAILED DESCRIPTION OF THE INVENTION
The invention according to claim 1 is an input unit that receives as input at least a keyword reflecting the user's interest and an evaluation value indicating the degree of interest of the user of the keyword, and for each of the keywords, the shape, size, and A keyword body generation unit that sets a keyword body expressed in color, and an arrangement generation unit that obtains arrangement coordinates so as to randomly arrange the keyword bodies, and stores the setting values and arrangement coordinates of the keyword bodies in an arrangement data storage unit And placing the keyword body on the display based on the placement coordinates, and according to an evaluation value indicating the degree of interest of the user of each keyword , Place keyword bodies close to each other, form a keyword body group of the keyword bodies placed nearby, Said each Keyword body group And a display unit that changes the display method to be the same, and has a function of easily displaying a keyword reflecting the user's interest. .
[0028]
Claim 2 The invention described in claim 1 In the described automatic placement device, The degree of interest of the user acquired the user's interest for each keyword based on whether or not the information to which at least one keyword is assigned is necessary. And has an effect that the keyword reflecting the user's interest can be easily displayed.
[0029]
Claim 3 The invention described in claim 1 In the keyword visualization device described, The method of displaying the keyword body is determined using at least one of the shape, size, and color of the keyword body of the keyword using the degree of reflection of the user's interest in the corresponding keyword. This is characterized in that the keyword reflecting the user's interest can be displayed in an easily understandable manner.
[0030]
Claim 4 The invention described in claim 1 In the visualization apparatus described in The display of the keyword body on the display is characterized by displaying the keyword over the keyword body, and the keyword can be displayed in an easily understandable manner. It has the action.
[0032]
Claim 5 The invention described in claim 1 In the keyword visualization device described in The evaluation value indicating the degree of interest of the keyword user was created for the keyword reflecting the user's interest based on the user input indicating whether or not the information to which at least one keyword was assigned (appeared simultaneously in each information). Metric signal (positive metric signal / negative metric signal) showing relationship to keyword and its strength (frequency) by autocorrelation matrix And a keyword that reflects the user's interests The reason It has the effect that it can be displayed easily.
[0033]
Claim 6 The invention described in claim 5 In the keyword visualization device described, The metric signal is a branching metric signal (branching positive metric signal / branching negative metric signal) that is only a keyword with a high degree of user interest. It is characterized by that, By deleting keywords that are less interesting to users, The keyword reflecting the user's interest can be displayed in an easily understandable manner.
[0035]
Claim 7 The invention described in claim 1 The keyword visualization device according to claim 1, wherein the display colors of the keyword bodies of the one keyword body group are the same, and keywords reflecting the user's interest are grouped for each keyword group. It has the effect that it can be displayed in an easily understandable manner.
[0036]
Claim 8 The invention described in claim 1 In the keyword visualization device described in the above, the input unit includes, as information reflecting the user's interest, a keyword reflecting the user's interest, an evaluation value indicating the degree of the user's interest in the keyword, and relationship information between the keywords. In response to the classification label, the layout generation unit is characterized in that the keyword bodies having the same classification label form one keyword body group, and the keyword reflecting the user's interest is accurately obtained. , It has the effect that it can be displayed together in an easy-to-understand manner for each field of the keyword.
[0037]
Claim 9 The invention described in claim 8 The keyword visualization device described above is characterized in that the display colors of the keyword bodies of the one keyword body group are the same, and keywords reflecting the user's interest are grouped for each field of the keywords. And can be displayed in an easily understandable manner.
[0038]
Claim 10 The invention described in claim 8 In the keyword visualization device described in the above, the relation information between the keywords is created for the keyword reflecting the user's interest by the user input as to whether or not the information to which at least one keyword is assigned is necessary (each information From the metric signal (positive metric signal / negative metric signal) indicated by the autocorrelation matrix (the relationship to the keywords that simultaneously appear in) and the strength (frequency) It is characterized in that the obtained classification label is included, and has an effect that keywords reflecting the user's interest can be displayed together in each keyword field for easy understanding.
[0039]
Claim 11 The invention described in claim 1 In the keyword visualization device according to, the input unit includes, as information reflecting the user's interest, a keyword reflecting the user's interest and an evaluation value indicating the degree of the user's interest in the keyword (and relationship information between keywords). The layout generation unit receives an initial placement generation unit that generates placement coordinates so as to be randomly placed as an initial placement of the keyword body generated by the keyword body generation unit, and the keyword body A layout update determining unit that repeatedly updates the layout so that keyword bodies having the same classification label are grouped as one keyword body group from the placement coordinates and an evaluation value indicating a degree of interest of the keyword by the user. The keywords that reflect the user's interests can be quickly and accurately added to the keyword. An effect that can be displayed easily understand grouped by de fields.
[0049]
Hereinafter, embodiments of the present invention will be described with reference to FIGS.
[0050]
(Embodiment 1)
FIG. 1 shows a block diagram of a keyword visualization system according to Embodiment 1 of the present invention. In FIG. 1, 1000 is a visualization device that arranges and displays keywords reflecting user interests in an easy-to-understand manner, 1001 is a memory, 1002 is a CPU, 1003 is a display / keyboard, and 1004 is a record of data in which keywords are automatically arranged. An arrangement data storage unit, 1005 is an FD (floppy disk) unit for reading a program, 1006 is an I / F unit, 1007 is an information filter device that generates information reflecting the user's interest, 2 is a code dictionary storage unit, 532 is a branch cut positive metric storage unit, 533 is a branch cut negative metric storage unit, 538 is a classification label storage unit, 1008 is an I / F unit, and 1009 is a network for the visualization device to receive information from the information filter device 1007. It is configured.
[0051]
Here, the information reflecting the user's interest includes a keyword reflecting the user's interest, a branching positive metric signal described later, a branching negative metric signal described later, and a classification label vector described later. 2 (see FIG. 2), a branch cut positive metric storage unit 532 (see FIG. 13), a branch cut negative metric storage unit 533 (see FIG. 13), and a classification label storage unit 538 (see FIG. 13). .
[0052]
Hereinafter, a first embodiment of the present invention will be described with reference to the drawings. First, prior to the description of the keyword visualization method and apparatus that are the basis of the present invention, keywords, branch-cutting positive metric signals, branch-cutting negative metric signals, and classification labels that reflect the user's interest as inputs to the present invention. An information filter device that generates vectors will be described.
[0053]
FIG. 2 is a block diagram showing the configuration of the information filter device, and FIG. 3 is a block diagram grouped into functional units for easy understanding of the configuration and operation.
[0054]
First, the basic concept of the information filter device will be described with reference to FIG.
[0055]
The information filter device includes a plurality of storage units 2, 5, 6, and 8 that store records relating to a history of what “information” the user has required in the past, and an information filtering unit 50 that performs filtering of “information”. And the unread data storage unit 10 that stores the unread “information” (information that the user has not yet read) actually filtered by the information filtering unit 50, and the user can view the unread “information”. A classification interface unit 70 such as a display, a learning unit 52 that learns about what “information” the user needed, and a record about what “information” the user needed in the past A method for dividing a record stored in a plurality of storage units 5 and 6 storing a plurality of records into a plurality of records Divide unit 53, a plurality of classification information filtering units 55-1 and 55-2 for filtering "information" using the divided records (two cases are shown in the drawing), and unread data for classification It consists of storage units 56-1 and 56-2.
[0056]
The operation of the above configuration will be described below. In the following description, it is assumed that the history of what “information” the user has required in the past has already been learned. In addition, one or more keywords corresponding to the “information” are attached to what is simply referred to as “information” below. The keyword may be a part or the whole of each word constituting the “information”, or may be specially added to represent the “information”.
[0057]
First, when new “information” is input to the information filtering unit 50, the information filtering unit 50 describes what “information” the user has required in the past from the storage units 2, 5, 6, and 8. The record is read and the necessity of the new “information” is quantitatively evaluated as a necessity signal.
[0058]
Similarly, the classification information filtering unit 55 also quantitatively evaluates the necessity of the new “information” as a necessity signal.
[0059]
Next, the evaluated new “information” is stored in the unread data storage unit 10 and the classification unread data storage unit 56 by the information filtering unit 50 and the classification information filtering units 55-1 and 55-2. The inputted "information" is written in the order so that the necessity signal is arranged in the order including the unread "information" from the past. If the user desires, the classification interface unit 70 presents unread “information” including the new “information” to the user one by one in descending order of necessity signals (for example, displayed on the data display device 34). .
[0060]
At this time, the user receives a teacher signal indicating whether each unread “information” including the new “information” presented to the user is necessary or unnecessary for the user, and the user inputs a teacher signal input terminal 105 of the classification interface unit 70. The classification interface unit 70 receives the teacher signal and sends the “information” and the teacher signal to the learning unit 52. The input of the teacher signal by the user is performed in order to further improve the learning ability of the learning unit 52, and the learning ability of the learning unit 52 (history of what “information” the user has required in the past). Is not necessary if the learning ability is already high enough.
[0061]
Next, the learning unit 52 rewrites the history contents of the storage units 2, 5, 6, and 8 using the presented “information” and its teacher signal.
[0062]
As described above, the information filter device can adapt to the user through higher learning, classify and preferentially present the “information” required by the user. Of course, in the initial state in which learning is not performed, the learning unit 52 does not know what “information” the user needs, and therefore, all new input “information” is classified into the classification interface unit 70. However, each time a user receives a presentation, it is necessary to input the teacher signal as described above. However, it is necessary to adapt to the user through learning conducted as needed, and to classify and preferentially present the “information” that the user wants. it can.
[0063]
It should be noted that “information” requested by the user is preferentially presented, as described in a more specific use example, by searching for a population A of a certain “information” database with a specific keyword and searching for a search set B of “information” However, not all of the “information” of the search set B is necessary for the user, and even if all of the “information” is necessary for the user, the necessary order naturally exists. It is assumed that. Therefore, presentation in the order of the user's necessity on the data display device 34 of the classification interface unit 70 in accordance with the necessity order in accordance with the necessity order means that the “information” requested by the user is preferentially presented.
[0064]
Consider the case where a keyword is attached to the input “information” as described above. Considering a single user, a keyword set A with high frequency or probability attached to “information” that the user needs, and a keyword set B attached with high frequency or probability to “information” that is unnecessary, Furthermore, it is possible to consider a keyword set C that is attached or not attached to both.
[0065]
Accordingly, a positive numerical value is assigned to a keyword belonging to the keyword set A, a negative value is assigned to a keyword belonging to the keyword set B, and a value 0 is assigned to a keyword belonging to the keyword set C.
[0066]
Further, it is configured to determine which keyword group of the keyword sets A, B, and C each belongs to one or more keywords included in the newly input “information”, and accumulate the allocated values. To do.
[0067]
With this configuration, a plurality of keywords attached to the newly input “information” are replaced with “information” including a large number of keywords belonging to the keyword set A (information that is likely to be required by the user). ) Is a large positive value, and for “information” (information that is likely to be unnecessary by the user) with many keywords belonging to the keyword set B, it is a numerical value indicating a large negative value. Can be converted.
[0068]
Thus, the user's needs can be predicted using the numerical values. As described above, the information filter device automatically assigns values to keywords (including keyword co-occurrence) based on the presented “information” and the user's necessary / unnecessary evaluation regarding the “information”. It realizes calculation of high necessity signals and rearranges “information” in order of high accuracy and necessity.
[0069]
Therefore, here, a plurality of keywords attached to “information” are converted into one vector, and the autocorrelation matrix of the vector is calculated separately when the user needs it and when it is not needed. . Using the autocorrelation matrix MY created from the keywords associated with “information” that the user answered is necessary, the length SY of the vector V is
SY = ΣΣMij ・ Vi ・ Vj
ij
And calculate.
[0070]
In the following, the autocorrelation matrix MY made from the keyword associated with “information” answered as “necessary” is referred to as “positive metric signal”, and the autocorrelation matrix MN produced from the keyword associated with information answered as unnecessary is referred to as “negative metric signal”. And the length SY is called an affirmative signal.
[0071]
The length SY is a large positive value if the keywords that are often included in the "information" required by the user are included in a plurality of keywords from which the vector V is based. Since it takes a value and takes a value close to 0 otherwise, it is effective in calculating the necessity signal.
[0072]
The metric divider 53 divides the positive metric signal MY and the negative metric signal MN generated in this way into a plurality (two in the case of FIG. 3), and writes them into the classification information filtering units 55-1 and 55-2.
[0073]
The classification information filtering unit 55 performs the same operation as the information filtering unit 50 using the divided positive metric signal and negative metric signal, and arranges “information” in the order of necessity and classification by the user. Present.
[0074]
Next, a block corresponding to the information filtering unit 50 shown in FIG. 3 using FIG. 2 and a block corresponding to the learning unit 52 shown in FIG. 3 using FIG. 4 will be described for each functional unit.
[0075]
First, the configuration of a block corresponding to the information filtering unit 50 will be described.
[0076]
The information filtering unit 50 converts a plurality of keywords (exactly, character strings including classification codes) attached to individual “information” into vectors, and what “information” is necessary / unnecessary by the user. It is necessary to reflect the necessity of "information" from the positive signal and negative signal that calculate the positive signal and negative signal representing a certain score using positive metric signal and negative metric signal representing the history, and the positive signal and negative signal. A part for calculating the sex signal and a part for rearranging the information in descending order of the necessity signal. Hereinafter, the configuration of a block corresponding to the information filtering unit 50 will be described with reference to FIG.
[0077]
In FIG. 2, 1 is a vector generation unit that converts a plurality of character strings such as keywords attached to “information” into vectors, and 2 stores a code dictionary signal for converting a plurality of character strings such as keywords into vectors. The code dictionary storage unit. The code dictionary signal stored in the code dictionary storage unit 2 is a code book having nofDCK correspondence tables for converting a character string W such as a keyword attached to “information” as shown in (Expression 1) into a number C. .
[0078]
[Expression 1]
Figure 0003799905
[0079]
The vector generation unit 1 receives the keyword group signal Kof = (K [1],..., K [nofKs]) including the keyword number signal nofKs and nofKs keyword signals, and the keyword group signal Ks and the code dictionary signal. Conversion to vector signal V using DCK. Reference numeral 3 denotes a score calculation unit that uses the positive metric signal MY and the negative metric signal MN calculated from the result of evaluating that the “information” presented to the user is necessary / unnecessary, and the two converted by the vector generation unit 1 The length of the vector signal V is converted into a positive signal SY and a negative signal SN. 5 is a positive metric storage unit that stores the positive metric signal MY that is a (nofDCK × nofDCK) matrix, and 6 is a negative metric storage unit that stores the negative metric signal MN that is a (nofDCK × nofDCK) matrix. 8 is a determination parameter storage unit for storing the determination parameter signal C, 7 is the determination signal signal C read from the determination parameter storage unit 8 upon receipt of the positive signal SY and the negative signal SN, and the necessity signal N and the reliability signal R are read out. It is a necessity calculation part to calculate.
[0080]
9 is an unread data write in which information data D, which is the main body of “information”, a keyword number signal nofKs, a keyword group signal Ks, a necessity signal N, and a reliability signal R are written in an unread data storage unit 10 to be described later according to a predetermined procedure. The control unit 10 is the unread data shown in (Expression 2), that is, the information data D which is the text of the “information”, the keyword number signal nofKs, the keyword group signal Ks, the necessity signal N, and the reliability signal. This is an unread data storage unit for storing at most nofURD unread data consisting of R.
[0081]
[Expression 2]
Figure 0003799905
[0082]
Reference numeral 13 denotes a teacher data storage unit that stores a maximum of notTD teacher data signals shown in (Equation 3).
[0083]
[Equation 3]
Figure 0003799905
[0084]
Is a teacher data storage unit.
[0085]
Next, the block configuration of the classification interface unit 70 shown in FIG. 3 will be described.
[0086]
In FIG. 3, 70 receives a control signal DO from the data read start signal input terminal 103 and is selected by the user from the unread data storage unit 10 or the classification unread data storage units 56-1 and 56-2. The unread data signal URD [1] is read from the storage unit, the display signal DD is output to the data display device 34, and a teacher signal T indicating whether or not the display signal DD is necessary for the user is received from the user from the teacher signal input terminal 105. The classification interface unit writes the teacher signal T, the keyword number signal nofKs [1] of the unread data signal URD [1], and the keyword group signal Ks [1] into the teacher data storage unit 13 according to a predetermined procedure. is there.
[0087]
Next, the configuration of a block corresponding to the learning unit 52 shown in FIG. 3 will be described. The learning unit 52 includes a part for performing metric learning for correcting an affirmative / negative metric signal using a teacher signal T input from a user, a parameter for calculating a necessity signal from the affirmative / negative signal, and a determination parameter signal. It consists of parts to be corrected, and each part is controlled by a learning control unit.
[0088]
The configuration of the part that performs the metric learning shown in FIG. 4 is as follows. In FIG. 4, reference numeral 19 denotes a metric learning unit that corrects the positive metric signal MY stored in the positive metric storage unit 5 and the negative metric signal MN stored in the negative metric storage unit 6. The metric learning unit 19 reads the teacher data TD from the teacher data storage unit 13, converts a plurality of keywords into vectors by the learning vector generation unit 20 having the same function as the vector generation unit 1 of the information filter unit 50, Correct the positive / negative metric signal by calculating the autocorrelation matrix.
[0089]
The configuration of the part that learns the determination parameter signal is as follows.
[0090]
In FIG. 4, a learning score calculation unit 22 includes a learning positive signal calculation unit 221 and a learning negative signal calculation unit 222. In this learning score calculation unit, 221 receives a learning vector signal from the learning vector generation unit 20 and calculates a learning positive signal LSY, and 222 learns from the learning vector generation unit 20. It is a learning negative signal calculation part which receives the vector signal for calculation and calculates the negative signal for learning LSN. 21 is a determination surface learning unit that receives the determination parameter learning control signal PLC from the learning control unit 14 and rewrites the determination parameter signal in the determination parameter storage unit 8 by a predetermined method, and 14 receives the learning start signal LS and switches 16 and 17 , 18, a metric learning unit 19, a learning vector generation unit 20, a learning score calculation unit 22, a learning negative signal calculation unit 23, and a determination surface learning unit 21.
[0091]
About the information filter apparatus comprised as mentioned above, the operation | movement is demonstrated for each unit using drawing.
[0092]
An example of a preferable initial state of the information filter device is that the positive metric signal MY and the negative metric signal MN are (nofDCK × nofDCK) zero matrix, and all the necessary signals N [N] of the unread data URD [i] in the unread data storage unit 10 i] (i = 1,..., nofURD) The minimum value Vmin that can be expressed by the hardware using the hardware, and the teacher signal T [j] of the teacher data TD [j] in the teacher data storage unit 13 are all −1. It is in the state.
[0093]
First, the operation of the information filtering unit 50 will be described.
First, the information data D is input from the information data input terminal 100, the keyword number signal nofKs indicating the number of keywords attached to the information data is input from the keyword number signal input terminal 101, and a plurality of keywords are input from the keyword signal input terminal 102. Keyword group signal Ks = (K [1], K [2],..., K [nofKs]).
[0094]
The keyword group signal Ks is converted from a collection of character strings into a vector signal V by the vector generation unit 1. By this conversion, the similarity of the keyword group signal can be calculated as a vector distance.
The operation of the vector generation unit 1 will be described with reference to the flowchart shown in FIG. First, upon receiving the keyword number signal nofKs and the keyword group signal Ks (step (b) in FIG. 5), the internal vector signal V = (V [1], V [2],..., V [nofDic]). Is set to (0, 0,..., 0), and the keyword counter signal i is set to 1 (steps (b) and (c) in the figure). Next, after the dictionary counter signal j is set to 0, the dictionary counter signal j is incremented by 1 (step (d) in the figure).
[0095]
Next, the code dictionary signal DCK [j] consisting of a keyword and a number designated by the dictionary counter j is read from the dictionary storage unit 2 having nofDCK code dictionary signals DCK therein, and the character string portion W [ j] and the i-th keyword signal K [i] are compared (step (e) in the figure). If they are not equal, the dictionary counter j is incremented by 1 (step (f) in the figure). Steps (e) to (g) in FIG. 4 are repeated until both values match or the value of the dictionary counter j is equal to the number of code dictionary signals nofDiC stored in the dictionary storage unit 2 (step (G) in FIG. )).
[0096]
When W [j] equal to the keyword signal K [i] is found, the j-th component V [j] of the vector signal is set to 1 (step (H) in the figure), and the keyword counter signal i is increased by 1 (FIG. 5). Step (Re)). Thereafter, the same processing is executed until the keyword counter signal i becomes larger than the keyword number signal nofKs (step (n) in the figure).
[0097]
Thus, in the vector generation unit 1, the keyword group signal Ks, which is an aggregate of keyword signals made up of character string signals, is converted into a vector signal V having nofDCK components encoded with 0 and 1.
[0098]
Next, the affirmation signal calculation unit 31 calculates an affirmation signal SY having a large value when the keyword group signal Ks includes many keywords that have been included in the information required by the user in the past. For this purpose, the affirmation signal calculation unit 31 receives the vector signal V, reads the affirmation metric signal MY from the affirmation metric storage unit 5, and calculates the affirmation signal SY by (Equation 4).
[0099]
[Expression 4]
Figure 0003799905
[0100]
The negative signal calculation unit 32 calculates the negative signal SN having a large value when the keyword group signal Ks includes a large number of keywords that have been included in the information unnecessary for the user in the past. For this purpose, the negative signal calculation unit 32 reads the negative metric signal MN from the negative metric storage unit 6 and calculates the negative signal SN by (Equation 5).
[0101]
[Equation 5]
Figure 0003799905
[0102]
The positive metric signal MY and the negative metric signal MN are determined based on the keyword group signal Ks and the user response, as will be described later. Using the affirmative signal SY and the negative signal SN calculated in this way, as shown in FIG. 11, at one point on the two-dimensional space where the vertical axis represents the positive signal SY and the horizontal axis represents the negative signal SN. Information data D can be made to correspond. The distribution of the information data D in this two-dimensional space is mainly what the user needs (indicated by ○) is distributed in the upper left, and what the user does not need (indicated by ×) is mainly distributed in the lower right. Will come to do. Therefore, by determining an appropriate coefficient C as shown in FIG. 12, the information data D required by the user and the unnecessary information data D can be separated.
[0103]
Further, the necessity signal N calculated using this coefficient C described below becomes larger as it is located at the upper left in the above-described two-dimensional space, that is, the information data D predicted to be more necessary. Therefore, if information data D are arranged and presented in descending order of necessity signal N, the user can efficiently obtain necessary information. The reliability signal R in the direction orthogonal to the necessity signal N is a signal indicating how many keyword signals are included in the dictionary among the keywords included in the keyword group signal Ks. Therefore, the magnitude of the reliability signal R indicates how reliable the necessity signal N calculated by the information filter is.
[0104]
Next, the necessity calculator 7 receives the positive signal SY output from the positive signal calculator 31 and the negative signal SN output from the negative signal calculator 32, and determines from the determination parameter storage unit 8. When the parameter signal C is read and there are many keywords related to information that was necessary in the past, and there is almost no keyword related to unnecessary information, the necessity signal N that becomes a large value is obtained.
N = SY-C / SN
And calculate the reliability signal R
R = C ・ SY + SN
And calculate.
[0105]
The operation of the unread data write control unit 9 will be described with reference to the flowchart shown in FIG. First, the information data D, the keyword number signal nofKs, and the keyword group signal Ks are received from the respective input terminals, the necessity signal N and the reliability signal R are received from the necessity calculator 7, and the unread data portion The unread data processing signal WI output from the instruction terminal 110 is changed from 0 to 1 (step (a) in FIG. 6). Next, i = 1 is set (step (b) in the figure), and the necessity signal N [i] (i = 1,..., NotURD) of the unread data URD [i] stored in the unread data storage unit 10. Are sequentially read and compared with the necessity signal N (step (c) in the figure), the necessity signal N becomes larger than the necessity signal N [i] of the unread data URD [i] (N ≧ N [i ]) The number i1 of the first unread data is detected (steps (d) (e) in the figure).
[0106]
i1 and subsequent unread data
URD [i + 1] = URD [i] i = i1,..., NotURD
(Steps (f) to (ri) in the figure), and then the i1th unread data URD [i1]
N [i1] = N
R [i1] = R
nofKs [i1] = nofKs
Ks [i1] = Ks
D [i1] = D
And the necessity signal N or the like (step (nu) in the figure). When this replacement is completed, the unread data portion instruction signal WI output from the unread data portion instruction terminal 110 is returned to 0 (step (L) in the figure), and the processing is ended.
[0107]
Next, the classification interface unit 70 that reads the unread data UDR and adds a user response (teacher signal T) to generate the teacher data signal TD will be described. The operation of the classification interface unit 70 will be described with reference to the flowchart shown in FIG.
[0108]
Hereinafter, in a preferred embodiment, the unread data storage unit 10 is treated as a classified unread data storage unit whose classification label signal Cat is 0. Hereinafter, when explaining the operation of the classification interface unit, the unread data storage unit 10 is treated as a classification unread data storage unit having a classification label signal Cat of 0.
[0109]
A data read start signal DO is input from the data read start signal input terminal 103 (step (a) in FIG. 7). The classification interface unit 70 waits for a predetermined time to receive the classification label signal Cat specifying the unread data storage unit 10 or the classification unread data storage unit category 56 to be accessed (step (b) in the figure). If the classification label signal Cat is not input within the time, the classification interface unit sets a standard value (Cat = 0 in the preferred embodiment) in the classification label signal Cat (step (d) in the figure). When a value is input to the classification label signal Cat within the time period or a standard value is set, the first unread data URD [] from the classification unread data storage unit 56-Cat specified by the classification label signal Cat. 1] is read (step (e) in the figure), and if the unread data necessity signal N [1] is larger than the minimum value Vmin, the information signal D [1] of the unread data signal URD [1] is displayed. The signal DD is output to the data display device 34 and waits (steps (f) and (g) in the figure). When the unread data necessity signal N [1] is equal to the minimum value Vmin, the display information signal DD is output as “no data” to the data display terminal 104 and waits (step (H) in FIG. 5).
[0110]
A user (not shown) looks at the display information signal DD displayed on the data display device 34, and if it is necessary information, the teacher signal T = 1. If not, the teacher signal T = 0. When the process is terminated, the teacher signal T = −1 is returned to the teacher signal input terminal 105 (step (I) in the figure). When the teacher signal T = −1, the process is terminated. When the teacher signal T ≠ −1 (step (L) in the figure), the unread data output control unit 11 stores the equation (2) in the teacher data storage unit 13. ) For teacher data
TD [i] = TD [i-1], i = 2,..., NotTD
(Step (W) in the figure) The first teacher data TD [1] is replaced with the teacher signal T, the keyword number signal nofKs [1] of the unread data and the keyword group signal Ks [1].
T [1] = T
TnofKs [1] = nofKs [1]
TKs [1] = Ks [1]
(Steps (W) and (Y) in the figure), the unread data URD in the unread data storage unit 10 is replaced with
It is assumed that URD [i] = URD [i + 1], i = 1,..., (NofURD-1) (steps (t) and (re) in the figure), and the necessity signal of the nofURD-th unread data is represented by N [nofURD ] = (Minimum value Vmin)
(Steps (tu), (ne), (na) in the figure).
[0111]
Next, the operation of the learning unit 52 will be described with reference to the flowcharts shown in FIGS.
[0112]
FIG. 8 shows a flowchart showing an outline of the operation of the learning control unit 14, which will be described in detail.
[0113]
In FIG. 8, a learning start signal LS is first input from the learning start signal input terminal 106, and the learning control unit instruction signal LI output from the learning control unit instruction signal output terminal 107 is changed from 0 to 1 (step (FIG. 8)). A)), showing the process. Next, the switch 16, the switch 17, and the switch 18 are switched so that the metric learning unit 19 and the learning vector generation unit 20 are connected (step (b) in the figure).
[0114]
Next, the metric learning unit 19 corresponding to step (c) in FIG. 8 is operated (step (c) in FIG. 8), the determination surface learning unit 21 is operated (step (d) in FIG. 8), and LI is set to 0. (Step (e) in the figure), the process is terminated.
[0115]
Next, an operation in which the metric learning unit 19 corrects the positive / negative metric signal using the user response (teacher signal T) and the keyword group signal Ks will be described with reference to FIG.
[0116]
FIG. 9 is a flowchart of the operation of the metric learning unit 19. In FIG. 9, the metric learning unit 19 that receives the metric learning control signal MLC from the learning control unit 14 (step (b) in FIG. 9) stores the positive metric memory. The positive metric signal MY is read from the unit 5, and the negative metric signal MN is read from the negative metric storage unit 6.
[0117]
Next, the metric learning unit 19 sets the value of the teacher data counter c to 1 (step (b) in the figure). Next, the c-th teacher data signal TD [c] is read from the teacher data storage unit 13 (step (c) in the figure), and the teacher signal T [c] of the teacher data TD [c] is examined. When the teacher signal T [c] is not −1 (T ≠ −1) (step (D) in the figure), the keyword number signal TnofKs [c] and the keyword group signal TKs [c] of the teacher data TD [c]. ] Is output (step (e) in the figure). The learning vector generation unit 20 that receives the keyword count signal TnofKs [c] and the keyword group TKs [c] of the teacher data TD [c] performs the same operation as the vector generation unit 1 of the information filtering unit 50 described above. The learning vector signal LV is output (step (f) in the figure). The metric learning unit 19 receives the learning vector signal LV, and when the teacher signal T [c] of the teacher data TD [c] is T = 1 (step (g) in the figure), a positive metric signal MY
MY [i] [j] = MY [i] [j] + LV [i] · LV [j]
(Where i, j = 1 to notDiC)
(Step (C) in the figure).
[0118]
By this processing, the positive metric signal MY has a larger value than the keyword signal (plurality) for the information data D required by the user. As a result, the affirmative signal SY described above becomes larger than the information data D required by the user. The negative metric signal MN is similarly processed as follows.
[0119]
When the teacher signal T [c] of the teacher data TD [c] is T = 0, the negative metric signal MN is
MN [i] [j] = MN [i] [j] + LV [i] · LV [j]
(Where i, j = 1 to notDiC)
(Step (re) in the figure). Next, the value of the teacher data counter
c = c + 1
And increase by 1 (step (nu) in the figure).
[0120]
Thereafter, the metric learning unit 19 repeats the same operation until the teacher signal T [c] of the teacher data TD [c] becomes T [c] = − 1 or c = nofTD. When T [c] = − 1 or c = ofTD (step (wo) in the figure), the metric learning process is terminated and the metric learning control signal MLC is sent to the learning control unit 14.
[0121]
The learning control unit 14 receives the metric learning control signal MLC from the metric learning unit 19, switches the switch 16 so that the learning vector generation unit 20 and the score calculation unit 22 are connected, and learns the switch 17 and the switch 18. It switches so that the vector generation part 20 and the determination surface learning part 21 may connect. The learning control unit 14 sends a determination surface learning control signal PLC to the determination surface learning unit 21.
[0122]
Next, the determination surface learning unit 21 will be described in detail with reference to FIG.
As shown in FIG. 12, the determination plane learning unit 21 uses the positive signal SY and the negative signal SN to express the information data D required by the user and the information data D unnecessary for the user. The coefficient C that best separates and is obtained.
[0123]
For this purpose, it will be described in detail according to the flowchart shown in FIG.
[0124]
First, upon receiving the determination surface learning control signal PLC (step (b) in FIG. 10), the value of the teacher data counter c is set to 1 (step (b) in FIG. 10). The c-th teacher data signal TD [c] is read from the teacher data storage unit 13 (step (c) in the figure), and the teacher signal T [c] of the teacher data TD [c] is examined (step (d) in the figure). . When the teacher signal T [c] is not −1 (T ≠ −1), the keyword number signal TnofKs [c] and the keyword group signal TKs [c] of the teacher data TD [c] are output (same as above). Figure step (e)). The learning vector generation unit 20 that receives the keyword count signal TnofKs [c] and the keyword group TKs [c] of the teacher data TD [c] performs the same operation as the vector generation unit 1 of the information filtering unit 50 described above. The learning vector signal LV is output.
[0125]
The learning score calculation unit 22 performs the same operation as the score calculation unit 3 of the information filtering unit 50 described above, outputs a learning positive signal LSY [c] and a learning negative signal LSN [c], and determines the determination surface. The learning unit 21 receives it (step (f) in the figure). The learning positive signal LSY [c], the learning negative signal LSN [c], the teacher signal T [c] of the teacher data TD [c], and the determination plane learning signal TC [c] = (T [c], LSN [c], LSY [c]) are stored in the internal storage element (step (g) in the figure). And the value of the teacher data counter
c = c + 1
And increase by 1 (step (C) in the figure).
[0126]
Thereafter, the determination surface learning unit 21 repeats the same operation until the teacher signal T [c] of the teacher data TD [c] becomes T [c] = − 1 or c = nofTD + 1 (step in the figure). (Li)). When T [c] = − 1 or c = ofTD, processing such as calculation of the learning positive signal LSY [c] is terminated.
[0127]
Next, the determination surface learning unit 21 determines that the determination surface learning signal TC [c] (c = 1,...) Stored in the internal storage element is LSN [c] on the horizontal axis and LSY on the vertical axis. When [c] is indicated, T [c] = 1 is indicated by ◯, and T [c] = 0 is indicated by ×, the distribution shown in FIG. 11 is obtained. Among these, the determination parameter C that can be best separated between the teacher signal T [c] = 1 and the teacher signal T [c] = 0 as shown in FIG. Calculate (step (nu) in the figure). Next, the determination parameter C is written into the determination parameter storage unit 8, a determination plane learning control signal PLC is sent to the learning control unit 14 (step (L) in the figure), and the process ends.
[0128]
The learning control unit 14 receives the determination surface learning control signal PLC from the determination surface learning unit 21, sets the learning control unit instruction signal to a value indicating standby, and ends the process.
[0129]
As shown in FIG. 11, in the two-dimensional space in which the keyword group signal is expressed by the positive signal SY and the negative signal SN using the two metric signals described above, the information required by the user is mainly in the upper left. Unnecessary information is distributed in the lower right. Therefore, if the necessity signal is N = SY−C · SN using the appropriate coefficient C as described above, the necessity signal takes a large value for the information required by the user.
[0130]
Through the operation as described above, a matrix (positive metric signal MY and negative metric signal MN) reflecting the user's interest is generated in the positive metric storage unit 5 and the negative metric storage unit 6.
[0131]
By appropriately dividing the matrix generated in this way, it is possible to realize an interest classification suitable for each user. Specifically, the metric division unit 53 reads the positive metric signal MY and the negative metric signal MN, processes and generates the divided positive classification metric signal ClassMY and negative classification metric signal ClassMN, and the classification information filter The data is written in the classification positive metric storage unit 555 and the classification negative metric storage unit 556 in the units 55-1 and 55-2.
[0132]
FIG. 13 is a block diagram showing a configuration of the metric division unit 53.
In FIG. 13, 54 is a classification parameter storage unit for storing parameters related to classification, and the stored contents can be rewritten by input from the classification parameter input terminal 301.
[0133]
In the figure, reference numeral 531 denotes a branching processing unit which reads a branching threshold value from the classification parameter storage unit 54 and removes components that easily become noise during classification from the positive metric signal MY and the negative metric signal MN. Reference numeral 532 denotes a branch cut positive metric storage unit that stores a branch cut positive metric signal CMY that is a positive metric signal subjected to branch cut processing. Reference numeral 533 denotes a branch cut negative metric signal CMN that is a negative metric signal subjected to branch cut processing. Is a branch-cut negative metric storage unit. Reference numeral 534 denotes a branching metric synthesizer that calculates a branching metric signal CM from the branching positive metric signal CMY and the branching negative metric signal CMN.
[0134]
Reference numeral 535 denotes a matrix calculation control unit that controls a matrix calculation performed using the debranching metric signal CM, 536 denotes a matrix calculation unit controlled by the matrix calculation control unit 535, and 537 to the matrix calculation unit 536. Is a matrix operation input storage unit that stores the inputs. 538 is a classification label storage unit that stores a vector having a classification label as a component based on the result of the matrix operation, and 539 is a vector component stored in the classification label storage unit 538 and the debranching positive metric signal CMY, It is a classification metric writing unit that reads the maximum number of classifications from the branch cut negative metric signal CMN and the classification parameter storage unit 54 and sets the classification positive metric signal ClassMY and the classification negative metric signal ClassMN of the classification information filtering unit 55.
[0135]
The operation of the metric division unit 53 having the above configuration will be described in detail with reference to FIGS.
[0136]
When the learning start signal LS is input from the learning start signal input terminal 106, the metric division unit 53 enters a standby state, and then the value of the learning control unit instruction signal LI output to the learning control unit instruction signal output terminal 107 is When it is changed from 1 to 0 and it is indicated that the operation of the learning unit 52 is completed, the following processing shown in the flowcharts of FIGS. 14 to 17 is started.
[0137]
FIG. 14 is a flowchart showing processing of the branch cut processing unit 531. This process is a process of branching out the components of the positive metric signal MY and the negative metric signal MN that are not so large. This makes it easy to write two metric signals that can be expressed as a matrix as a sum of sub-matrices independent of each other.
[0138]
Upon receiving the classification start signal CSt from the classification start signal input terminal 108, the branch cut processing unit 531 reads the branch cut threshold value Nlow from the classification parameter storage unit 54 (step (A)). In addition, it is also preferable to use the learning control unit instruction signal LI as the classification start signal as shown in FIGS.
[0139]
Next, the row counter i and the column counter j are set to 0 (step (b), step (c)). Then, the (ij) component MY [i] [j] of the positive metric signal MY and the (ij) component MN [i] [j] of the negative metric signal MN are read from the positive metric storage unit 5. (Step (d)).
[0140]
The sum of the (ij) component MY [i] [j] of the positive metric signal MY and the (ij) component MN [i] [j] of the negative metric signal MN compares the branching threshold Nlow (step) (E)). If the sum is larger than the branching threshold Nlow, the (ij) component MY [i] [j] of the positive metric signal MY is changed to the (ij) component CMY [i] [j] of the branching positive metric signal CMY. The (ij) component MN [i] [j] of the negative metric signal MN is substituted into the (ij) component CMN [i] [j] of the debranching positive metric signal CMN (step (To)).
[0141]
If the sum is smaller than the branch cut threshold Nlow, the (ij) component CMY [i] [j] of the branch cut positive metric signal CMY and the (ij) component CMN [i] [j] of the branch cut positive metric signal CMN. 0 is substituted into (Step (To)).
Then, the column counter j is incremented by one (step (H)).
[0142]
Thereafter, the same processing is performed for all components (steps (L) to (L)).
[0143]
When the above branching processing is completed, the branching processing unit 531 sends a metric synthesis start signal to the branching metric synthesis unit 534.
[0144]
The branching metric synthesis unit 534 reads the branching positive metric signal CMY and the branching negative metric signal CMN according to the flowchart shown in FIG. 15 (c), calculates the sum thereof, and generates the classification matrix signal Class (two). ), And is substituted into the matrix calculation unit 536.
[0145]
When the processing is completed, a matrix calculation start signal is sent to the matrix calculation control unit 535.
[0146]
When receiving the matrix calculation start signal, the matrix calculation control unit 535 performs a process of detecting a vector that is independently converted by the classification matrix signal Class. Details of the processing will be described with reference to the flowchart shown in FIG. The purpose of this series of processing is to specify the components of the input vector that can be converted into each other by the classification matrix. The meaning can be considered that the corresponding keywords are mutually related to the components of the input vectors that can be converted to each other. Thus, dividing the components of the input vector that can be converted into each other into several groups can be considered as corresponding to classifying the user's interest into several groups.
[0147]
First, the matrix calculation control unit 535 initializes the classification label vector CatV stored in the classification label storage unit 538 to a zero vector (step (A)). Next, a value 1 indicating the first category is assigned to the category signal Cat (step (b)). Prior to determining the input vector, the input vector CV stored in the matrix operation storage unit 537 is initialized to 0 vector (step (c)). Then, the component counter i is set to 0 (step (d)).
[0148]
It is determined whether or not the i-th component CatV [i] of the classification label vector CatV specified by the component counter i is equal to 0 (step (e)). If equal, 1 is substituted into the i-th component CV [i] of the input vector CV (step (f)). As a result, an initial input vector in which only one component is 1 and all other components are 0 is obtained. This input vector is multiplied by the classification matrix signal Class to obtain a conversion vector CVtmp (step (g)). For all components of the transformation vector Ctmp,
Ctmp [i] = Θ (Ctmp [i])
(Step (C) to (L)). Where the function Θ is
Θ (x) = 1 x> 0
0 Other cases
It is a function.
[0149]
The conversion vector CVtmp whose component is 1 or 0 is compared with the input vector CV (step (W)). If the conversion vector CVtmp is different from the input vector CV, the conversion vector CVtmp is substituted into the input vector CV, and the process returns to step (G) (step (W)). If the conversion vector CVtmp and the input vector CV are equal, the classification label vector CatV
CatV = CatV + Cat · CV
(Step (f)). Then, the category signal Cat is increased by 1, and the process returns to step (c) (step (y)).
[0150]
When the values of all the components of the classification label vector are not 0, the processing flow branches from step (e) to end.
[0151]
With the above processing, the classification label vector CatV is obtained when the classification matrix signal Class generated from the debranching positive metric signal CMY and the debranching negative metric signal CMN is composed of four independent matrices.
Initial state CatV = (0,0,0,0,0,0,0,0,0,0,0,0,0,0,0)
Cat = 1 CatV = (1,0,1,0,0,0,0,0,1,0,1,0,0,0,0)
Cat = 2 CatV = (1,2,1,0,2,0,2,0,1,0,1,2,2,0,0)
Cat = 3 CatV = (1,2,1,3,2,3,2,0,1,3,1,2,2,0,3)
Cat = 4 CatV = (1,2,1,3,2,3,2,4,1,3,1,2,2,4,3)
It will change.
[0152]
In this way, the user's interests can be classified into several groups by dividing the components of the input vector that can be transformed into one another into several groups. As the next processing, the classification information filtering unit 55 is set using this result.
[0153]
When the classification label vector CatV is obtained as described above, the matrix operation control unit 535 sends a writing start signal to the classification metric writing unit 539.
[0154]
The classification metric writing unit 539 receives the writing start signal and starts the processing shown in the flowchart of FIG. The purpose of this series of processing is to divide the debranching positive metric signal CMY and the debranching negative metric signal CMN into numbers equal to or less than the maximum classification number NCmax based on the classification label vector, and classify each into the classification of the classification information filtering unit 55. The positive metric storage unit 555 and the classification negative metric storage unit 556 are written.
[0155]
First, the classification label vector is read (step (A)). Next, the maximum classification number NCmax is read from the classification parameter storage unit 54 (step (b)). In the configuration shown in FIG. 2, since the number of classification information filtering units is two, it is desirable that the maximum classification number NCmax is 2. Next, all the classification positive metric signal ClassMY and classification negative metric signal ClassMN of NCmax classification information filtering units 55-1 to NCmax are initialized to a zero matrix (step (c)).
[0156]
The row counter i is set to 0 (step (d)). The column counter j is set to 0 (step (e)). Then, it is determined whether or not the i-th component CatV [i] of the classification label vector is equal to or less than the maximum classification number NCmax (step (f)). If it is below, the process proceeds to step (g), and if larger, the process proceeds to step (wo). In step (g), the i-th component CatV [i] and the j-th component CatV [j] of the classification label vector are compared.
[0157]
In step (g), if the i-th component CatV [i] and the j-th component CatV [j] of the classification label vector are equal, this means that the two components are converted into each other by the classification matrix Class, that is, independent. This means that the (ij) component CMY [i] [j] of the debranching positive metric signal CMY and the (ij) component CMN [i] [j] of the debranching negative metric signal CMN are read (step (H)). Then, the (ij) component of the defragmentation positive metric signal is substituted for the (ij) component of the classification positive metric signal ClassMY of the classification information filtering unit 55-CatV [i] with the classification number CatV [i], and the classification negative metric signal The (ij) component of the debranching negative metric signal is substituted for the (ij) component of ClassMN (step (i)).
[0158]
On the other hand, if the two components are not equal in step (g), it means that the two components are not converted by the classification matrix Class, that is, the two components are independent. The processing is omitted from step (h) to step (li).
[0159]
Then, the column counter j is incremented by one (step (nu)), and the value of the column counter j is compared with the dimension NofDCK of the classification label vector (step (le)). If the value of the column counter j is smaller than the dimension of the classification label vector, the process returns to step (g). If the value of the column counter j is greater than or equal to the dimension of the classification label vector, the process proceeds to step (v).
[0160]
In step (W), the value of the row counter i is incremented by one. The value of the row counter i is compared with the dimension NofDCK of the classification label vector (step (W)). If the value of the row counter i is smaller than the dimension of the classification label vector, the process returns to step (e). If the value of the row counter i is greater than or equal to the dimension of the classification label vector, the process ends.
[0161]
With the above processing, all the classification information filtering units 55 are appropriately set (including the case where both the classification positive metric signal and the classification negative metric signal are zero matrices).
[0162]
Since the operation of each classification information filtering unit 55 is the same as the operation of the information filtering unit 51, description thereof is omitted.
[0163]
In the foregoing, prior to the description of the present invention, an information filter device that generates a keyword, a branch cut positive metric signal, a branch cut negative metric signal, and a classification label vector that reflect the user's interest as an input to the present invention has been described.
[0164]
The keyword visualization method and apparatus according to the present invention reads a keyword, a branch cut positive metric signal, a branch cut negative metric signal, a classification label vector, etc. reflecting the user's interest generated by such an information filter device, and is understood by the user. It is to display keywords in a form that is easy to do.
[0165]
Here, it should be noted that the keyword itself is a sequence of characters and does not originally have a shape, size, color, or the like. For this reason, in the present invention, a keyword body having a shape, a size, and a color is given for each keyword. That is, the arrangement position of the keyword body instead of the keyword is determined, and the keyword is displayed using the arrangement position of the keyword body.
[0166]
Hereinafter, the operation of the visualization apparatus 1000 in the visualization system configured as shown in FIG. 1 will be described with reference to the flowchart of FIG. Here, a set in which the arrangement coordinates of all keyword bodies are collected together is called an arrangement S. When arranging in a two-dimensional space, the arrangement coordinates (center coordinates) of the keyword bodies i (= 1,..., NofDCK) are ( sx (i), sy (i)), the arrangement S = {sx (i), sy (i) | i = 1,. . . , NofDCK}. In the following description, the case of arranging in a two-dimensional space will be described.
[0167]
First, the keyword i (= 1,..., NofDCK) reflecting the user's interest stored in the code dictionary storage unit 2 in the information filter device 1007 is read via the network 1009 (a). Here, in order to simplify the following description, it is assumed that the keywords are arranged in the descending order of the degree of reflection of the user's interest. This is easy, for example, by using the value of the keyword cost signal KD (i) of the keyword i disclosed in Japanese Patent Laid-Open No. 9-288683 (the keyword having a higher degree of user interest reflection takes a larger value). Therefore, generality is not lost.
[0168]
Next, the classification label vector CatV is read from the classification label storage unit 538 in the information filter device 1007 via the network 1009 (B). For the keyword i, a classification label CatV (i) that is the i-th component of the classification label vector CatV is set.
[0169]
Next, the branch cut positive metric signal CMY is read from the branch cut positive metric storage unit 532 in the information filter device 1007 via the network 1009 (C). The pair of keywords i and j is set with a branch cut positive metric element signal CMY (i, j), which is the (i, j) component of the branch cut positive metric signal.
[0170]
Next, the branch cut negative metric signal CMN is read from the branch cut negative metric storage unit 532 in the information filter apparatus 1007 via the network 1009 (d). For the pair of keywords i and j, a debranching negative metric element signal CMN (i, j) which is the (i, j) component of the debranching negative metric signal is set.
[0171]
Next, initial settings are made for an arrangement area (a keyword body is arranged in the arrangement area), an arrangement S, a temperature T, a keyword body, and the like (e). Regarding initialization of the arrangement S, for example, an initial arrangement in which the arrangement coordinates of all keyword bodies are randomly set in the arrangement area, or an initial arrangement in which the arrangement coordinates of the keyword bodies having the same classification label are set close in the arrangement area. The arrangement may be the initial arrangement S. For initialization of the temperature T, an appropriate value larger than 0 may be given. In addition, initialization of parameters of a cost function F (S) described later and a temperature update coefficient β described later is also performed.
[0172]
Further, as the initial setting of the keyword body, the shape, size, and color are determined. As described above, since the keywords are arranged in the descending order of the degree of reflection of the user's interest, for example, if the shape of the keyword body i is a regular (i + 2) square, the keyword is not important. As the keyword disappears, the keyword body of the keyword approaches a circle, and the user can understand the importance of the keyword at a glance.
[0173]
Moreover, even if the shape of the keyword body is fixed, the user can understand the importance of the keyword at a glance by changing the size. For example, the size of the keyword body i is set so as to satisfy the relationship of (the size of the keyword body (i−1))> (the size of the keyword body i)> (the size of the keyword body (i + 1)). This will improve understanding. Furthermore, even if the shape and size of the keyword body is constant, the user can understand the importance of the keyword at a glance by changing the color. Of course, it is also conceivable that keyword bodies having the same classification label have the same color, and the user's advantage in this case is that keywords in the same field can be understood at a glance. As described above, determining the shape and / or size and / or color of the keyword body using the degree of reflection of the user's interest is extremely important to help the user's understanding.
[0174]
Next, the counter count is set to 0 (f), and the arrangement S is updated (g). To update the arrangement S, first, a keyword body or keyword body group is selected, and then an arrangement candidate S ′ is generated by changing the arrangement coordinates of the selected keyword body or keyword body group within the arrangement area. Finally, it is only necessary to determine whether or not the arrangement candidate S ′ is to be the arrangement S at the next time through a comparison between the cost of the arrangement S and the cost of the arrangement candidate S ′.
[0175]
The simplest cost function F (S) for evaluating the arrangement S may be, for example, (Equation 6).
[0176]
[Formula 6]
Figure 0003799905
[0177]
Here, A and B are positive constants. D (i, j, S) is the distance between the keyword bodies i and j in the arrangement S. Further, o (i, j, S) is a function that takes a value of 1 when there is an overlap between the keyword bodies i and j in the arrangement S, and takes a value of 0 otherwise. Using such a cost function F (S), for example, as shown in (Expression 7), the placement candidate S ′ may be set as the placement S at the next time with the following probability.
[0178]
[Expression 7]
Figure 0003799905
[0179]
Here, T is a temperature parameter as described above.
[0180]
Regarding the determination of the keyword body group, for example, the keyword bodies having the same classification label may be considered as one keyword body group.
[0181]
Furthermore, the keyword body group can be determined using the distance d (i, j, S) between the keyword bodies i and j. For example, when (CMY (i, j) + CMN (i, j)) is greater than 0 for keyword bodies i and j, the time change H (i, j) of d (i, j, S) is small, that is, When the arrangement positions of the keyword bodies i and j are stable, they may be used as keyword body groups. Here, the time change H (i, j) of d (i, j, S) can be calculated by, for example, (Equation 8).
[0182]
[Equation 8]
Figure 0003799905
[0183]
Here, ε is a positive constant. In addition, S indicates the current arrangement, and S ′ indicates the arrangement at the next time (the arrangement candidate accepted in (Equation 7)).
[0184]
When this H (i, j) is below a certain threshold, the keyword bodies i and j form one keyword body group. In addition, a new keyword body group in which some of these keyword body groups are grouped can be configured. Specifically, when a common keyword body is included in two keyword body groups, a new keyword body group is formed by grouping these two keyword body groups, and this is repeated.
[0185]
When the arrangement coordinates are changed using the keyword body group configured as described above as a unit, the relative arrangement coordinates between the keyword bodies included in the keyword body group are stored. Thereby, the arrangement correction is performed at high speed, and a good arrangement result can be obtained in a short time.
[0186]
In order to check how the arrangement S is changed, the updated arrangement S may be displayed on the display 1003.
[0187]
Subsequent to the update of the arrangement S in step (g), 1 is added to the counter count in step (h).
[0188]
Next, in step (i), if count <nofDIK (Y), the process returns to step (g); otherwise (N), the process proceeds to step (n).
[0189]
In step (n), it is determined whether the end condition is satisfied. For example, the process may be ended when the change in the cost F (S) of the arrangement S becomes small.
[0190]
Alternatively, a counter such as the counter count may be provided, the maximum number of repetitions may be set in advance, and the process may be terminated when the counter exceeds the maximum number.
[0191]
If the end condition is not satisfied in step (nu) (N), the process proceeds to step (le) and the temperature T is updated. For example, it is conceivable to set the temperature update constant β in a range of 0 <β <1, and update it as (Equation 9).
[0192]
[Equation 9]
Figure 0003799905
[0193]
If the end condition is satisfied in step (nu) (Y), the process proceeds to step (wo), the arrangement S is stored in the arrangement data storage unit 1004, and if necessary, is also stored in the FD 1005 and the process ends. In order to confirm the arrangement S, that is, in order to confirm the arrangement result, the arrangement S may be displayed on the display 1003.
[0194]
The above keyword visualization method was applied to an example that includes keywords that reflect the user's interests. FIG. 19 shows an initial arrangement (for simplicity, a random arrangement), and FIG. 20 shows an arrangement result. A square in the figure is a keyword body, and a word on the keyword body is a keyword. For simplicity, the keyword body has a square shape and a constant size.
[0195]
Further, the straight line connecting the keyword bodies in the figure indicates the relationship between the keywords. It can be seen from FIG. 20 that is an arrangement result that related keyword bodies are arranged so as to form a cluster.
(Embodiment 2)
Below, an example of the keyword visualization system in Embodiment 2 of this invention is demonstrated using the block diagram of FIG. Although the case of arranging in a two-dimensional space will be described in the same manner as described in the first embodiment, it can be easily extended to the case of arranging in a three-dimensional space. In FIG. 21, the blocks (for example, the CPU 1002 and the like) and the bus lines that are common to FIG. 1 and need not be described are omitted so as not to make the drawing complicated.
[0196]
In FIG. 21, reference numeral 2100 denotes an information filter device for generating information reflecting a user's interest (keyword reflecting the user's interest, the classification label vector, the debranching positive metric signal, the debranching negative metric signal), Code dictionary storage unit 2 that stores keywords reflecting the user's interest, classification label storage unit 538 that stores the classification label vector, branch cut positive metric storage unit 532 that stores the branch cut positive metric signal, and the branch cut It is comprised from the debranching negative metric memory | storage part 533 which memorize | stores a negative metric signal.
[0197]
Reference numeral 2101 denotes a visualization device for arranging keywords reflecting the user's interest on the display in an easy-to-see manner. Reference numeral 2102 denotes a setting input unit for allowing a user to change a parameter value (default value) set in the initial setting unit 2103 described later. Reference numeral 2103 denotes an attribute of an arrangement area (a keyword body is arranged in this area). Various parameter values required for the operation of the visualization apparatus 2101 such as size, shape, color), cost function coefficient, initial temperature, temperature change coefficient, etc. are set to default values or values input from the setting input unit 2102. An initial setting unit 2104 for receiving a keyword reflecting the user's interest from the code dictionary storage unit 2 and a classification label vector from the classification label storage unit 538 and generating a keyword body.
[0198]
2105 receives the keyword body from the keyword body generation unit 2104 and the placement area from the initial setting unit 2103, generates the initial placement of the keyword body, and holds the initial placement cost calculated by the cost calculation unit 2107 described later. The initial arrangement generation unit 2106 receives a branch cut positive metric signal from the branch cut positive metric storage unit 532 and a branch cut negative metric signal from the branch cut negative metric storage unit 533, and generates a branch cut metric signal. The metric signal generation unit 2107 receives the cost function coefficient from the initial setting unit 2103, the branching metric signal from the branching metric signal generation unit 2106, and the arrangement S from the initial arrangement generation unit 2105 and the arrangement candidate generation unit 2109 described later. The cost calculation unit 2108 for calculating the cost and the current arrangement and An arrangement holding unit 2109 for holding a cost changes the arrangement held by the arrangement holding unit 2108 to generate an arrangement candidate, and an arrangement candidate for holding the cost of the arrangement candidate calculated by the cost calculation unit 2107 It is a generation part.
[0199]
2110 receives the current arrangement and its cost from the arrangement holding unit 2108 and the arrangement candidate and its cost from the arrangement candidate generation unit 2109, receives the temperature held by the temperature update unit 2111, and the next time An arrangement update determining unit that determines the arrangement of the image and sends the arrangement and cost to the arrangement holding unit 2108. Reference numeral 2111 denotes an initial temperature and temperature change coefficient from the initial setting unit 2103, and the number of arrangement updates in the arrangement update determining unit 2110. The temperature update unit 2112 that updates / holds the temperature and the like is an arrangement output unit that outputs the arrangement held by the arrangement holding unit to the arrangement data storage unit 1004.
[0200]
Hereinafter, details of each unit will be described, but the code dictionary storage unit 2, the classification label storage unit 538, the branch cut positive metric storage unit 532, and the branch cut negative metric storage unit 533 in the information filter device 2100 are implemented. Since it is the same as that of Form 1, it is omitted here.
[0201]
The setting input unit 2102 is for allowing a user to change a parameter value (default value) set in the initial setting 2103, and a keyboard, a mouse, a voice microphone, or the like can be used.
[0202]
The initial setting unit 2103 sets various parameter values necessary for the operation of the visualization apparatus 2101. The parameter value set by the user using the setting input unit 2102 is used as the value of each parameter. Default values are set for parameters that are not set by the user. Parameters set by the initial setting unit 2103 include attributes (size, shape, and color) of an arrangement area (a keyword body is arranged in this area) required by the initial arrangement generation unit 2105 and the arrangement candidate generation unit 2109. There are a cost function coefficient required by the cost calculation unit 2107, an initial temperature and a temperature change coefficient required by the temperature update unit 2111, and the like.
[0203]
The keyword body generation unit 2104 receives a keyword reflecting the user's interest from the code dictionary storage unit 2 and the classification label vector CatV from the classification label storage unit 538, and generates a keyword body. One keyword body i is generated for each keyword i (i = 1,..., NofDCK) read from the code dictionary storage unit 2. For example, when the shape of the keyword body is circular, the shape of each keyword body i is disclosed in Japanese Patent Application Laid-Open No. 9-288683. A circle having a radius that reflects the value of the keyword cost signal KD (i) of the keyword i that can be used (a keyword having a higher degree of user interest reflection can be used) can be used. Furthermore, using the classification label CatV (i) of the keyword body i (the i-th element of the classification label vector CatV), a color can be assigned to each keyword body group having the same classification label.
[0204]
The initial arrangement generation unit 2105 receives the arrangement area set by the initial setting unit 2103 and nofDCK keyword bodies i (i = 1,..., NofDCK) generated by the keyword body generation unit 2104, and sets the arrangement S. initialize. When the arrangement coordinates (center coordinates) of the keyword body i (= 1,..., NofDCK) are represented by (sx (i), sy (i)), the arrangement S is S = {sx (i), sy (I) | i = 1,. . . , NofDCK}. Here, sx (i) and sy (i) are the X coordinate value and the Y coordinate value of the keyword body i, respectively. As a specific method for initializing the arrangement S, sx (i) and sy (i) may be set at random so that the keyword body i is within the arrangement area. Further, the arrangement S may be initially set so that keyword body groups having the same classification label are collected. The initial arrangement S generated in this way is sent to the cost calculation unit 2107, and its cost F (S) is calculated. The initial arrangement generation unit 2105 holds the cost F (S) in addition to the initial arrangement S.
[0205]
The branch cut metric signal generation unit 2106 receives the branch cut positive metric signal CMY (i, j) (i, j = 1,..., NofDCK) from the branch cut positive metric storage unit 532 and the branch cut negative metric storage unit 533. Branch metric signal CM (i, j) (i, j = 1,...) Received from the chopping negative metric signal CMN (i, j) (i, j = 1,..., NofDCK) from , nofDCK). Specifically, the branching metric signal CM (i, j) may be the sum of each element of the branching positive metric signal CMY (i, j) and the branching negative metric signal CMN (i, j). .
[0206]
The cost calculation unit 2107 receives the coefficient of the cost function from the initial setting unit 2103 and the branching metric signal CM (i, j) (i, j = 1,..., NofDCK) from the branching metric signal generation unit 2106. In response, a cost function F (S) is generated. If CM (i, j) = CMY (i, j) + CMN (i, j), the cost function F (S) may be generated as in (Equation 6). The values set in the initial setting unit 2103 are used as the values of the coefficients A and B in the cost function F (S) of (Expression 6). The cost calculation unit 2107 calculates the cost F (S) of the arrangement S input from the initial arrangement generation unit 2105 and the postscript arrangement candidate generation unit 2109 using the cost function F (S) generated in this way, It also has a function to output.
[0207]
The arrangement holding unit 2108 has a function of holding the current arrangement S and its cost F (S). Therefore, the arrangement holding unit 2108 first holds the initial arrangement S generated by the initial arrangement generation unit 2108 and its cost F (S) at the first time (time 0), and thereafter outputted from the arrangement update determination unit 2110 described later. The arrangement S and its cost F (S) are held.
[0208]
The arrangement candidate generation unit 2109 receives the arrangement S held by the arrangement holding unit 2108, changes it, and generates an arrangement candidate S ′. For example, the arrangement S can be changed by selecting a keyword body i and adding a random number to the sx (i) or sy (i). It is also conceivable to select a plurality of keyword bodies (keyword body group) and perform the same operation. The placement candidate generation unit 2109 has a function of holding the placement candidate S ′ thus generated and the cost F (S ′) obtained by sending the S ′ to the cost calculation unit 2107.
[0209]
The arrangement update determination unit 2110 includes the arrangement S held by the arrangement holding unit 2108 and its cost F (S), the arrangement S ′ held by the arrangement candidate generation unit 2109 and its cost F (S ′), and further In response to a temperature T held by a temperature updating unit described later, an arrangement S at the next time is determined. Specifically, for example, the placement candidate S ′ may be the next time placement S with the probability of (Equation 7), and the current placement S may be the next time placement S with the remaining probability.
[0210]
The arrangement S at the next time determined by the arrangement update determination unit 2110 and its cost F (S) are held by the arrangement holding unit 2108 as the arrangement S at the current time and its cost F (S). That is, a loop is formed by feedback from the arrangement update determining unit 2110 to the arrangement holding unit 2108, and the arrangement S changes every moment.
[0211]
The temperature update unit 2111 receives the initial temperature T and the temperature change coefficient β from the initial setting unit 2103, and updates / holds the current temperature T. For example, the number of arrangement updates in the arrangement update determining unit 2110 is counted, and the temperature T may be updated according to (Equation 9) when the count number is equal to or greater than a predetermined threshold.
[0212]
The arrangement output unit 2112 writes the arrangement data into the arrangement data storage unit 1004 in order to output the arrangement S held by the arrangement holding unit 2108 to a display, a hard disk, or the like.
[0213]
With the above configuration, it is possible to arrange and display keyword bodies of related keywords for each field of information for keywords reflecting the user's interest.
[0214]
(Embodiment 3)
Below, the information visualization editing system in Embodiment 3 of this invention is demonstrated below. An example of an information visualization editing system that can visualize information and edit the result according to the user's request will be described with reference to the block diagram of FIG. 22 and FIG. 23 showing a preferable example of the arrangement display. In addition, the case where it arrange | positions in a two-dimensional space is demonstrated similarly to description in Embodiment 1 and Embodiment 2, However, It can expand easily also when arrange | positioning in a three-dimensional space. Also in FIG. 22, as in FIG. 21, blocks (such as the CPU 1002) and bus lines that are common to FIG. 1 and do not need to be described in the description are omitted so as not to make the drawing complicated.
[0215]
In FIG. 22, reference numeral 2201 denotes a cutting line input unit for inputting a cutting line DL for cutting the keyword body (and its aggregate) included in the arrangement S, and 2202 denotes arrangement from the previous cutting line DL and the arrangement data storage unit 1004. S is a placement data cutting unit that receives the S and cuts the placement S into several parts and calculates a cut placement DS. 2203 inputs a joint line CL that connects the keyword bodies (and their aggregates) included in the placement S. 2204 is a joint line input unit that receives the joint line CL and calculates a layout S that connects the keyword bodies (and aggregates thereof). This is the same as that described in the second embodiment. Further, a straight line connecting the keyword body and the keyword body is referred to herein as a connection line.
[0216]
The editing input for editing the keyword body (and its aggregate) is not limited to the cutting line DL and the joining line CL. The figure indicating the cutting part and the joining part may be a figure for setting a certain area.
[0217]
Hereinafter, the operation of the visualization apparatus 2101 having the above configuration will be described.
[0218]
As described in the second embodiment, the arrangement S is written in the arrangement data storage unit 1004. The arrangement S is output to the monitor 1003 in the format as shown in FIG. While viewing the output of the monitor 1003, the user operates the cutting line input unit 2201 to input the cutting line DL for cutting the arrangement S as shown in FIG. A computer mouse is one of preferred examples of the cutting line input unit 2201 and the joining line input unit 2203. The arrangement data cutting unit 2202 that has received the cutting line DL detects a connection line CL (id, jd) that intersects the cutting line DL among the connection lines of the arrangement S, and stores the connection line CL (id, jd) in the branch cut positive metric storage unit 532. The stored (id, jd) component CMY (id, jd) and (jd, id) component CMY (jd, id) of the stored branch cut positive metric signal CMY are set to zero. This process is performed for all intersecting connection lines. As a result, as shown in the upper part of FIG. 25, it is possible to obtain the arrangements S1 and S2 in which the group of keyword bodies, which was a lump in FIG. This corresponds to the division of one category into two categories, and new processing corresponding to the divided arrangements S1 and S2 is performed by performing the processing following the metric synthesis processing described in the first embodiment. A classification label vector can be obtained.
[0219]
In addition, the user operates the joint line input unit 2203 while viewing the output of the monitor 1003, and inputs the joint line CL that connects the portions of the arrangement S as shown in FIG. The placement data joining unit 2204 that has received the joint line CL detects the keyword body ic and the keyword body jc connected by the joint line CL among the keyword bodies of the placement S, and stores them in the branch cut positive metric storage unit 532. The (ic, jc) component CMY (ic, jc) and the (jc, ic) component CMY (jd, id) of the stored branch cut positive metric signal CMY are set to 1.
[0220]
This process is performed for all connected keyword bodies. As a result, as shown in the lower left of FIG. 25, it is possible to obtain an arrangement S3 in which a group of keyword bodies that were two in FIG. 23 are joined together. This corresponds to the division of one category into two categories, and a new classification label corresponding to the arrangement S3 joined by performing the processing below the metric synthesis processing described in the first embodiment. You can get a vector.
[0221]
As described above, the cut line input unit for inputting the cut line DL for cutting the keyword body (and its aggregate) included in the arrangement S, and the arrangement S received from the cut line DL and the arrangement data storage unit. An arrangement data cutting unit for calculating a cutting arrangement DS obtained by cutting the portion into several parts, a joining line input unit for inputting a joining line CL for connecting a keyword body (and its aggregate) included in the placement S, and the joining By providing an arrangement joint that receives the line CL and calculates the arrangement S connecting the keyword bodies (and the aggregates thereof), an information visualization classification apparatus that can edit the visualized classification result by the user's hand can be realized.
[0222]
【The invention's effect】
As described above, the present invention receives a keyword obtained by an information filter device that acquires a user's interest by user input as to whether or not information is necessary, and at least one of a branch cut positive metric signal and a branch cut negative metric signal. For each of the keywords, a keyword body having a shape, size and color reflecting the user's interest is set, the keyword body is arranged, and the keyword body and the keyword are displayed on a display. As a result, at least one of the shape, size, and color of the keyword body of the keyword can be determined using the degree of reflection of the user's interest, and the branch cut positive metric signal and the branch cut negative metric are determined. Use at least one of the signals to identify the keyword body of the relevant keyword for each field of information Stopped and can be arranged, further, it is possible to provide a visualization method of a keyword that can be displayed easily understand the keyword body and the keyword on the display.
[Brief description of the drawings]
FIG. 1 is a block diagram of a keyword visualization system according to a first embodiment of the present invention.
FIG. 2 is a block diagram of an information filter device used in the keyword visualization system according to the first embodiment of the present invention.
FIG. 3 is a block diagram showing an outline of an information filter device used in the keyword visualization system according to the first embodiment of the present invention.
FIG. 4 is a block diagram showing a configuration of a part that performs metric learning in the information filter device;
FIG. 5 is a flowchart for explaining the operation of the vector generation unit of the information filter device;
FIG. 6 is a flowchart for explaining the operation of an unread data write control unit of the information filter device;
FIG. 7 is a flowchart for explaining the operation of the classification interface unit of the information filter device;
FIG. 8 is a flowchart for explaining the operation of the learning control unit of the information filter device;
FIG. 9 is a flowchart for explaining the operation of the metric learning unit of the information filter device;
FIG. 10 is a flowchart for explaining the operation of the determination surface learning unit of the information filter device;
FIG. 11 is a diagram for explaining the operation of the determination surface learning unit of the information filter device;
FIG. 12 is a diagram for explaining the operation of the determination surface learning unit of the information filter device;
FIG. 13 is a block connection diagram of a metric division unit of the information filter device.
FIG. 14 is a flowchart for explaining the operation of the branching processing unit of the metric division unit of the information filter device;
FIG. 15 is a flowchart for explaining the operation of the branching metric synthesis unit of the metric division unit of the information filter device;
FIG. 16 is a flowchart for explaining the operation of the matrix calculation control unit of the metric division unit of the information filter device;
FIG. 17 is a flowchart for explaining the operation of the classification metric writing unit of the metric division unit of the information filter device;
FIG. 18 is a flowchart for explaining the operation of the keyword visualization system according to the first embodiment of the present invention;
FIG. 19 is a diagram showing an example of an initial arrangement of the keyword visualization system according to the present invention.
FIG. 20 is a diagram showing an example of an arrangement result by the keyword visualization system of the present invention.
FIG. 21 is a block diagram showing a configuration of a keyword visualization system according to the second embodiment of the present invention.
FIG. 22 is a block diagram showing a configuration of an information visualization editing system according to a third embodiment of the present invention.
FIG. 23 is a diagram showing an example of keyword arrangement visualized according to the third embodiment of the present invention.
FIG. 24 is a diagram showing an example of keyword arrangement and cutting line / joining line input visualized according to Embodiment 3 of the present invention;
FIG. 25 is a diagram showing an example of keyword arrangement edited by the third embodiment of the present invention.
[Explanation of symbols]
1 Vector generator
2 Dictionary storage
3 Score calculator
5 Positive metric storage
6 Negative metric storage
7 Necessity calculation section
8 Judgment parameter storage
9 Unread data write controller
10 Unread data storage
11 Unread data output controller
12 Teacher data control unit
13 Teacher data storage
14 Learning control unit
16, 17, 18 switches
19 Metric Learning Department
20 Learning vector generator
21 Judgment surface learning section
22 Score calculator
34 Data display device
50 Information filtering unit
51 Interface unit
52 learning units
53 Metric division unit
54 Classification parameter storage unit
55 Classification information filtering unit
56 Classification unread data storage
60 Database storage
61 Database reading part
62 Database reconstruction control unit
63 switch
64 Adaptive database writer
65 Adaptive database storage
70 Classification interface unit
100 Information input terminal
101 Keyword number signal input terminal
102 Keyword signal input terminal
103 Data read start signal input terminal
105 Teacher signal input terminal
106 Learning start signal input terminal
107 Learning control unit instruction signal output terminal
108 Classification start signal input terminal
110 Unread data section instruction terminal
532 Branch cut positive metric storage unit
533 Branch cut negative metric storage unit
538 Classification label storage
1000 Visualization device
1001 memory
1002 CPU
1003 Display / Keyboard
1004 Placement data
1005 FD
1006 I / F
1007 Information filter device
1008 I / F
1009 network
2100 Information filter device
2101 Visualization device
2102 Setting input section
2103 Initial setting section
2104 Keyword field generator
2105 Initial arrangement generation unit
2106 Debranching metric signal generator
2107 Cost calculator
2108 Arrangement holding unit
2109 Arrangement candidate generation unit
2110 Location update determination unit
2111 Temperature update unit
2112 Layout output unit
2201 Cutting line input section
2202 Arrangement data cutting unit
2203 Joining line input section
2204 Placement data junction

Claims (11)

少なくともユーザの興味を反映したキーワードとそのキーワードのユーザの興味の度合いを示す評価値を入力として受ける入力部と、前記キーワードの各々に対し、形状、大きさ及び色で表現したキーワード体を設定するキーワード体生成部と、前記キーワード体をランダムに配置するように配置座標を求め、前記キーワード体の設定値及び配置座標を配置データ記憶部に記憶する配置生成部と、ディスプレイ上に前記キーワード体を前記配置座標に基づいて配置すると共に、前記各キーワードのユーザの興味の度合いを示す評価値に応じて、キーワード体同士を近くに配置し、近くに配置されたキーワード体を一つのキーワード体群を形成し、前記各キーワード体群の表示の仕方が同一になるように変更する表示部とを具備することを特徴とするキーワードの可視化装置。An input unit that receives as input at least a keyword reflecting the user's interest and an evaluation value indicating the degree of interest of the user of the keyword, and a keyword body expressed by shape, size and color for each of the keywords A keyword body generation unit, an arrangement coordinate unit that obtains arrangement coordinates so as to arrange the keyword bodies at random, store a setting value and arrangement coordinates of the keyword body in an arrangement data storage unit, and the keyword body on a display According to the evaluation value indicating the degree of interest of the user of each keyword , the keyword bodies are arranged close to each other, and the keyword bodies arranged close to one keyword body group are arranged based on the arrangement coordinates. formed, characterized in that the the manner of display of each keyword element group and a display unit to change to be the same Visualization device of keywords. 前記ユーザの興味の度合いは、少なくとも1つのキーワードが付与された情報が必要か否かのユーザ入力により、キーワード毎にユーザの興味を獲得したことを特徴とする請求項1に記載のキーワードの可視化装置。  2. The keyword visualization according to claim 1, wherein the degree of interest of the user has acquired the user's interest for each keyword based on a user input indicating whether information to which at least one keyword is assigned is necessary. apparatus. キーワード体の表示の仕方は、対応するキーワードに対してのユーザの興味の反映度合いを用いて、前記キーワードのキーワード体の少なくとも形状、大きさ、色の少なくとも1つ決定することを特徴とする請求項1記載のキーワードの可視化装置。  The method of displaying a keyword body is characterized in that at least one of the keyword body of the keyword is determined using at least one of the shape, size and color of the keyword using the degree of reflection of the user's interest in the corresponding keyword. Item 1. A keyword visualization device according to item 1. ディスプレイ上のキーワード体の表示は、キーワード体に重ねてキーワードを表示することを特徴とする請求項1に記載のキーワードの可視化装置。  The keyword visualization apparatus according to claim 1, wherein the keyword body is displayed on the display by overlapping the keyword body with the keyword. キーワードのユーザの興味の度合いを示す評価値は、少なくとも1つのキーワードが付与された情報が必要か否かのユーザ入力により、ユーザの興味を反映したキーワードについて作られた(各情報で同時に出現したキーワードに対する関係とその強さ(頻度)を)自己相関行列で示したメトリック信号(肯定メトリック信号/否定メトリック信号)を含むことを特徴とする請求項1に記載のキーワードの可視化装置。  The evaluation value indicating the degree of interest of the keyword user was created for the keyword reflecting the user's interest based on the user input indicating whether or not the information to which at least one keyword was assigned (appeared simultaneously in each information). The keyword visualization device according to claim 1, further comprising a metric signal (positive metric signal / negative metric signal) indicating the relationship to the keyword and its strength (frequency) by an autocorrelation matrix. 前記メトリック信号は、ユーザの興味の度合いの大きなキーワードのみとした枝切りメトリック信号(枝切り肯定メトリック信号/枝切り否定メトリック信号)であることを特徴とした請求項5に記載のキーワードの可視化装置。  6. The keyword visualization device according to claim 5, wherein the metric signal is a branch metric signal (branch cut positive metric signal / branch cut negative metric signal) that includes only a keyword having a high degree of interest of a user. . 前記一つのキーワード体群のキーワード体の表示色を同一とすることを特徴とする請求項1に記載のキーワードの可視化装置。  The keyword visualization device according to claim 1, wherein display colors of the keyword bodies of the one keyword body group are the same. 前記入力部は、ユーザの興味を反映した情報として、ユーザの興味を反映したキーワードとそのキーワードのユーザの興味の度合いを示す評価値及びキーワード間の関係情報を示す分類ラベルを受け、
前記配置生成部は、分類ラベルが同じキーワードのキーワード体が一つのキーワード体群を形成することを特徴とする請求項1に記載のキーワードの可視化装置。
The input unit receives, as information reflecting the user's interest, a keyword reflecting the user's interest, an evaluation value indicating the degree of the user's interest in the keyword, and a classification label indicating the relationship information between the keywords,
2. The keyword visualization apparatus according to claim 1, wherein the layout generation unit forms one keyword body group with keyword bodies having the same classification label.
前記一つのキーワード体群のキーワード体の表示色を同一とすることを特徴とする請求項に記載のキーワードの可視化装置。9. The keyword visualization device according to claim 8 , wherein display colors of the keyword bodies of the one keyword body group are the same. 前記キーワード間の関係情報は、少なくとも1つのキーワードが付与された情報が必要か否かのユーザ入力により、ユーザの興味を反映したキーワードについて作られた(各情報で同時に出現したキーワードに対する関係とその強さ(頻度)を)自己相関行列で示したメトリック信号(肯定メトリック信号/否定メトリック信号)から、キーワード間に関係があるキーワードを同一の分類ラベルを付けて求めた分類ラベルを含むことを特徴とする請求項に記載のキーワードの可視化装置。The relationship information between the keywords is created for the keyword reflecting the user's interest by the user input whether or not the information to which at least one keyword is assigned is necessary (the relationship to the keyword that appears simultaneously in each information and its relationship) It includes a classification label obtained by assigning the same classification label to a keyword related to the keyword from a metric signal (intensity (frequency)) indicated by an autocorrelation matrix (positive metric signal / negative metric signal). The keyword visualization device according to claim 8 . 前記入力部は、ユーザの興味を反映した情報として、ユーザの興味を反映したキーワードとそのキーワードのユーザの興味の度合いを示す評価値(及びキーワード間の関係情報を示す分類ラベル)を受け、
前記配置生成部は、前記キーワード体生成部で生成されたキーワード体の初期配置としてランダムに配置するように配置座標を生成する初期配置生成部と、前記キーワード体の配置座標と前記キーワードのユーザの興味の度合いを示す評価値から分類ラベルが同じキーワード体が1つのキーワード体群としてまとまるように前記配置の更新を繰り返し行なう配置更新決定部とを具備することを特徴とする請求項1に記載のキーワードの可視化装置。
The input unit receives, as information reflecting the user's interest, a keyword reflecting the user's interest and an evaluation value indicating the degree of the user's interest in the keyword (and a classification label indicating relationship information between the keywords),
The arrangement generation unit includes an initial arrangement generation unit that generates arrangement coordinates so as to randomly arrange as an initial arrangement of the keyword body generated by the keyword body generation unit, an arrangement coordinate of the keyword body, and a user of the keyword The arrangement update determining unit that repeatedly updates the arrangement so that keyword bodies having the same classification label are grouped as one keyword body group based on an evaluation value indicating a degree of interest. Keyword visualization device.
JP30366399A 1998-11-06 1999-10-26 Keyword visualization method and apparatus Expired - Fee Related JP3799905B2 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP30366399A JP3799905B2 (en) 1998-11-06 1999-10-26 Keyword visualization method and apparatus

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
JP10-315880 1998-11-06
JP31588098 1998-11-06
JP30366399A JP3799905B2 (en) 1998-11-06 1999-10-26 Keyword visualization method and apparatus

Publications (2)

Publication Number Publication Date
JP2000200290A JP2000200290A (en) 2000-07-18
JP3799905B2 true JP3799905B2 (en) 2006-07-19

Family

ID=26563586

Family Applications (1)

Application Number Title Priority Date Filing Date
JP30366399A Expired - Fee Related JP3799905B2 (en) 1998-11-06 1999-10-26 Keyword visualization method and apparatus

Country Status (1)

Country Link
JP (1) JP3799905B2 (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP4536225B2 (en) * 2000-07-28 2010-09-01 富士通株式会社 Dynamic determination of keywords and their importance in message sending and receiving systems
GB2390704A (en) 2002-07-09 2004-01-14 Canon Kk Automatic summary generation and display

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP3244005B2 (en) * 1995-09-04 2002-01-07 松下電器産業株式会社 Information filtering apparatus and information filtering method

Also Published As

Publication number Publication date
JP2000200290A (en) 2000-07-18

Similar Documents

Publication Publication Date Title
JP2003140942A (en) Methods, systems, and articles of manufacture for soft hierarchical clustering of co-occurring objects
JP2006520948A (en) Method, system and data structure for searching for 3D objects
CN106294481B (en) A kind of air navigation aid and device based on map
JP3244005B2 (en) Information filtering apparatus and information filtering method
CN107004025A (en) Image retrieving apparatus and the method for retrieving image
CN106844666B (en) Self-adaptive time series data query method
JP6868576B2 (en) Event presentation system and event presentation device
US6233352B1 (en) Information processing method and apparatus
JP3799905B2 (en) Keyword visualization method and apparatus
JP2008305268A (en) Document classification device and classification method
CN117235281A (en) Multi-element data management method and system based on knowledge graph technology
JP6567720B1 (en) Data preprocessing device, data preprocessing method, and data preprocessing program
JPH0944518A (en) Method for structuring image data base, and method and device for retrieval from image data base
JP3645431B2 (en) Information search support device and information search support program storage medium
CN105893343A (en) Information management device and information management method
CN113343012B (en) News matching method, device, equipment and storage medium
JP5061147B2 (en) Image search device
US7240050B2 (en) Methods and apparatus for generating automated graphics using stored graphics examples
CN115408611A (en) Menu recommendation method and device, computer equipment and storage medium
KR20050058877A (en) Apparatus and method for motion data processing
CN113870998A (en) Interrogation method, device, electronic equipment and storage medium
JPH10240762A (en) Information filter device and data base re-constructing device and information filtering method and initialization method
JP3772401B2 (en) Document classification device
JPH11219365A (en) Image retrieving device
US20040086203A1 (en) Database registration system and database registration method

Legal Events

Date Code Title Description
RD01 Notification of change of attorney

Free format text: JAPANESE INTERMEDIATE CODE: A7421

Effective date: 20050629

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20050816

A521 Written amendment

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20051003

A131 Notification of reasons for refusal

Free format text: JAPANESE INTERMEDIATE CODE: A131

Effective date: 20060117

A521 Written amendment

Free format text: JAPANESE INTERMEDIATE CODE: A523

Effective date: 20060309

TRDD Decision of grant or rejection written
A01 Written decision to grant a patent or to grant a registration (utility model)

Free format text: JAPANESE INTERMEDIATE CODE: A01

Effective date: 20060404

A61 First payment of annual fees (during grant procedure)

Free format text: JAPANESE INTERMEDIATE CODE: A61

Effective date: 20060417

FPAY Renewal fee payment (event date is renewal date of database)

Free format text: PAYMENT UNTIL: 20100512

Year of fee payment: 4

LAPS Cancellation because of no payment of annual fees