JP2004362019A - Information recommendation device, information recommendation method, information recommendation program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To recommend wide information by sharing history information among information sources and recommend information by considering each user's trends of using the information sources, in a device and method for recommending information by using history information of a plurality of information sources. <P>SOLUTION: The information recommendation device has, as a recommendation processing program, a means 6 for extracting a keyword from each piece of information, a means 7 for calculating a weight of the keyword from a history of access to all information sources 101 to 104, means 8 for calculating a weight of the keyword from a history of access to each information source, and a means 9 for calculating a recommendation level of each piece of information from the two types of weights of the keyword. Information recommendation is based on the use history of all the information sources, and also adjusted according to differences between the user's trends of using the information sources. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

であり、前記のようにコンテンツ情報が「ニュース２３」の場合は、「ニュース」というキーワードを含み、図３のキーワード評価テーブル２２から、その総合履歴得点が１８ということが分かり、単純得点は１８となる。
【００４７】
一方、一つの情報源αの得点Ｙは、

で算出され、情報源がテレビの場合は、図３のキーワード評価テーブル２２で「ニュース」のキーワードのテレビでの履歴得点が２であることから、１８＋２＝２０で、図４における「ニュース２３」のテレビでの得点が２０ということになる。
【００４８】
図４のコンテンツ評価テーブル２３の得点は、図３のキーワード評価テーブル２２が変更されるたびに変更される。また、新規のコンテンツ情報が入ってくる度にも算出される。
【００４９】
（推薦するコンテンツ情報の選び方）
図４のコンテンツ評価テーブル２３を参照し、図１の推薦度算出手段９を利用して、ユーザに推薦するコンテンツ情報を選出する。すなわち、入力部１からユーザによる推薦要求があった場合に、図４のコンテンツ評価テーブル２３から評価得点の高いコンテンツ情報を優先的に推薦する。このとき、単純得点ではなく、各情報源での得点を基に選出する。
【００５０】
たとえば、ユーザからの推薦要求が、「おすすめのテレビ番組を教えて欲しい」という内容ならば、図４のテレビの得点の中で一番得点の高いコンテンツ情報が選出される。たとえば「ニュース２３」のテレビの得点が２０であるのに対し、「ミュージックステーション」の得点が２４であることから、「ミュージックステーション」が選ばれる。
【００５１】
なお、単純得点では「ニュース２３」が１８で、「ミュージックステーション」が１６であり、「ニュース２３」のほうが高いが、これは、これらのコンテンツ情報に含まれる「ニュース」、「音楽」というキーワードで比較すると、「ニュース」をキーワードとする「ニュース２３」のほうが履歴得点が高いためである。
【００５２】
従来技術では、このような総合の履歴数が多いものが優先的に推薦されていたが、本実施の形態では、図３のキーワード評価テーブル２２から、「ニュース」はテレビでの得点数は低く、逆に「音楽」はテレビでの得点数が高いことがわかる。これによって、ユーザにテレビのコンテンツ情報を推薦する場合は、「音楽」を含むコンテンツ情報が優先されてもよいことが推測される。
【００５３】
複数の情報源の履歴を利用することによって、履歴情報が少ない状態でもよりユーザの好みに合致した情報を推薦できる可能性があるが、ユーザにはさまざまなメディア情報源を利用するときにも、個々のユーザの好みがあり、それを考慮した推薦方法が望ましい。メディア情報源毎において興味のある分野が変わる可能性もあり、このような場合にも対応できることが望まれる。
【００５４】
そこで、本発明では、総合の履歴数と、その情報源における履歴数を考慮した評価得点を算出することによって、メディア情報源全体での興味と、ある情報源内での興味を考慮した推薦を行うことが可能になる。全体的な興味の偏りと、メディア情報源間での興味の偏りを利用し、多様なメディア情報源の推薦に対応できる可能性がある。
【００５５】
ユーザからの推薦要求が、情報源を指定せず、「おすすめのコンテンツ情報を教えて欲しい」という内容ならば、図４の各情報源の得点のなかから高いものが優先される。この場合、得点が３８の「メールマガジンＮｏ．０２」が推薦される。
【００５６】
ユーザが情報源を指定しなかった場合、時間、使用端末などの状況によって、推薦するのでもよい。例えば、ユーザの使用している機器にテレビ機能がない場合は、テレビのコンテンツ情報を除いたものから選出したり、ユーザの今現在いる環境が、音をだせない状況ならば、メールコンテンツ情報を優先したりなどが考えられる。
【００５７】
なお、前述のように、単純得点は、従来技術との比較のために説明に用いたが、実際には使用しない。
【００５８】
【情報推薦処理の流れ】
図５は、本発明による情報推薦処理の流れの一例を示すフローチャート図である。ステップＳ４０１において、ユーザによる推薦要求、もしくは新着情報の受信、ユーザのアクセス履歴情報等が入力された場合には、ステップＳ４０２に進み、該入力が推薦要求かどうかを判別し、推薦要求でない場合は、ステップＳ４０３に進み、新規情報か否（すなわちアクセス履歴情報）かを判別する。
【００５９】
新規情報の場合にはステップＳ４０４に進み、新規情報の中からコンテンツ情報を一つ取り出し、ステップＳ４０５において、上記取り出したコンテンツ情報に含まれるキーワードを抽出する。キーワードの抽出の仕方は、先に述べたが、テキスト情報に含まれる名詞を抜き出してキーワードとするか、もしくはキーワード評価テーブル２２（図３）に存在するキーワードでテキスト情報の中に含まれるものを検索するものでもよい。
【００６０】
ステップＳ４０６において、前記ステップＳ４０５で抽出されたキーワードと図３のキーワード評価テーブル２２を参照し、コンテンツ情報の情報源における得点を計算する。計算の仕方は先に述べた通りであり、ステップＳ４０７において前記得点を図４のコンテンツ評価テーブル２３の対応する情報源の得点に追加する。なお、コンテンツ情報の単純得点は従来技術との比較を行うために、先の説明の中で説明したが、実際には使用しないので、計算しなくてよい。そして一つ一つのコンテンツ情報に関して上記の処理（Ｓ４０４〜Ｓ４０７）を行い、ステップＳ４０８において総てのコンテンツ情報の処理が終了したと判別すると、エンドに至る。
【００６１】
前記ステップＳ４０３において新規情報でないと判別した場合、すなわち、アクセス履歴情報だった場合は、ステップＳ４１０に進み、キーワード抽出手段６により、アクセスしたコンテンツ情報からキーワードを一つ取り出し、図３のキーワード評価テーブル２２を変更する。すなわち、図５のステップＳ４１１において、前記第１の重み算出手段７によりそのキーワードの総合得点のカウントを増やし、ステップＳ４１２において、前記第２の重み算出手段８によりその情報源における得点を増やす。キーワードが複数ある場合には、前記ステップＳ４１０〜Ｓ４１３の処理を繰り返し、ステップＳ４１３において、総てのキーワードについての処理が終了したと判別すれば、ステップＳ４１４に進み、図４のコンテンツ評価テーブル２３を変更する。すわち、そのキーワードを含み、メディア情報源が同じコンテンツ情報の、メディア情報源における得点を増やし、処理を終了する。
【００６２】
図５の前記ステップＳ４０２で情報推薦要求であった場合、前記推薦度算出手段９を利用して、ステップＳ４１５〜Ｓ４１８の処理が行われる。すなわち、ステップＳ４１５では、情報推薦要求に、メディア情報源の指定があるかないかを判別する。指定があった場合はステップＳ４１７に進み、指定された情報源から推薦情報を選択し、指定がなかった場合は、ステップＳ４１６に進み、全情報源から推薦情報を選択する。
【００６３】
情報の選択の仕方は先に述べたが、図４のコンテンツ評価テーブル２３を参照し、情報源における得点の高いコンテンツ情報を優先的に選択する。たとえばテレビ番組であれば、現在時刻に近いコンテンツ情報の優先度を上げたり、新しい情報ほど優先度を上げたり等、使用している機器がもつ機能や、ユーザの状況によって優先度を変更してもよい。そしてステップＳ４１６又はステップＳ４１７において推薦情報を選択したら、Ｓ４１８で情報を推薦して処理を終了させる。
【００６４】
以上、本発明の処理の流れの一例を述べたが、このように、複数の情報源の利用履歴を利用することによって、他の情報源で利用した興味のある分野の情報が別の情報源でも推薦することができ、履歴の少ない情報源においても個人の興味に合致した情報推薦が行うことが可能である。
【００６５】
さらに本発明では、全体のアクセス履歴だけでなく、情報源毎の履歴も利用して評価値を計算することにより、ユーザの情報源の利用傾向を考慮した情報推薦を行うことが可能である。これによって、メディア情報源ごとに利用する情報分野が異なるユーザにもよりよい情報推薦を行うことができる。
【００６６】
【発明の実施の形態２】
本願請求項２に係る発明に対応する実施の形態であり、履歴情報からキーワードの重みを算出する際に、前記実施の形態１では、キーワードの出現回数のみによって得点を算出していたが、本件実施の形態２では、そのコンテンツ情報によってキーワードの重みを変えて算出するために、前記図１のキーワード抽出手段６内に、キーワード毎に重み付けを行なう機能を備えている。その推薦方法を図５のフローチャートに基づいて説明する。
【００６７】
図５のステップＳ４１０において、たとえばテレビドラマの視聴履歴の場合、ＥＰＧのデータの中のタイトル、ジャンル、番組情報からキーワードを取り出すが、番組情報に出演者の情報も含まれていた場合に、出演者の中で主演の人の重みは大きく、助演の人の重みは小さくする。
【００６８】
たとえば図６及び図７により説明すると、テレビの履歴において、キーワードが俳優Ｂのところで、前述の図３では２回テレビで視聴していることから、得点は単純に２となっているが、該実施の形態２では、それが２回とも助演級だった場合は、重みを１回０．２とし、２回見ているので、図６のように俳優Ｂの得点は０．４となる。したがって、総合得点も、図３の俳優Ｂの総合得点３に対し、図７ではウェブでの得点１．０にテレビの得点０．４を加え、総合得点１．４となっている。
【００６９】
このように、番組のなかで出演場面が多く、ユーザが見ている人の重みを大きくしており、これによって、ユーザが見ている人の重みが大きくなり、その人が出演する番組の推薦度が高くなる。
【００７０】
たとえば、ユーザがタレントＴ１とタレントＴ２が好きで、それぞれが主演するドラマＤ１とＤ２を見ていたとする。さらに、どちらのドラマＤ１、Ｄ２にも助演としてタレントＴ３が出演していた場合、単に履歴ごとにカウントしていたのでは、タレントＴ３の重みが増えてしまうので、このような場合に本実施の形態のように、キーワードの重みに変化を付加することは有効である。
【００７１】
なお、主演かどうかの判断は、番組情報の出演者のなかで名前が最初の方に載っている人を主演であると判断する。普通ＥＰＧのデータには、番組情報の中で出演者が先に載っていて、その後に番組の説明文が続く。テレビドラマだけでなく、映画でもよい。
【００７２】
該実施の形態２による推薦処理は、必ずしも２種類のキーワードの重みを利用しなくてもよく、また、複数のメディア情報源の推薦の場合だけではなく、単独のメディア情報源内での推薦にも適応できる。したがって、履歴情報からキーワードを取り出してその重みを算出し、ユーザの好みを把握する場合のみに用いることができる。その場合の処理の流れは、図５のステップＳ４１０からＳ４１４までの処理となる。
【００７３】
【発明の実施の形態３】
本願請求項３に係る発明に対応する実施の形態であり、ユーザの好みを示すキーワードに基づいて、ユーザの好みに合致する情報を推薦する情報推薦装置であって、図１の推薦度算出手段９として、推薦対象となる情報に対し、該情報に含まれるキーワード毎に重み付けを変えて推薦度を算出する手段を備えている。
【００７４】
すなわち、図５のステップＳ４０５で新規コンテンツ情報からキーワードを取り出し、図３のキーワード評価テーブル２２を参照し、図５のステップＳ４０６でコンテンツ情報の評価を行うときに、コンテンツ情報におけるキーワードの重みを考慮して評価を行う。
【００７５】
具体的には、コンテンツ情報がテレビドラマ番組であった場合、主演の人の重みを大きくし、助演の人の重みを小さくして、コンテンツ情報の評価値を算出する。たとえば、図８で映画の評価を行なう場合に、俳優Ｂにおける重みは、前記図３のキーワード評価テーブル２２により、総合の重みが３、テレビにおける重みが２であるが、俳優Ｂがこの映画において助演であるとすると、本件実施の形態３では、その重みをそれぞれ１／２にする。それによって、評価値は次のような式で計算される。
【００７６】

【００７７】
この方法を用いると、例えば、ユーザがタレントＴを好きでその重みが高かった場合に、そのタレントＴが出演する番組が複数あった場合、そのタレントＴが主演する番組の推薦度を高くすることが可能である。
【００７８】
どれが主演かを判別する方法は、上記実施の形態２で述べている。どちらも例ではテレビドラマの主演の人の重みをあげることで説明しているが、テレビドラマや映画に限定されない。たとえば、テレビのタイトルに「タレントＴの料理番組」などのように、タイトルからタレントＴが中心であることがわかる場合には、そのタレントＴの重みを重くしてもよい。
【００７９】
タイトルからタレントＴを切り出すには、人名辞書を用いて切り出すかもしくは、番組説明の場所にタレントＴの名前が出演者として載っている場合は、そこからタレントＴを切り出すことができる。
【００８０】
前記の実施の形態２と同じく、この方法は２種類のキーワードの重みを利用しなくてもよく、また複数のメディア情報源の推薦の場合だけではなく、単独のメディア情報源内での推薦にも適応できる。したがって、キーワードの重みからユーザへの推薦度を算出する場合のみに用いることができる。その場合の処理の流れは、図５のステップＳ４０４からＳ４０９までとなる。
【００８１】
【発明の実施の形態４】
本願請求項４に係る発明に対応する実施の形態であり、図１の推薦度算出手段９は、各情報間で推薦度の差が０か一定値以内の場合には、情報源毎の利用時間の傾向に基づいて推薦順位を変更する機能を備えている。
【００８２】
すなわち、前記図５のステップＳ４１６において、全情報源から情報推薦を行うときに、上位の推薦度の差が一定値以内の場合に、利用時間傾向の短いメディア情報源の情報から推薦する。
【００８３】
図９は推薦方法を具体的に示すフローチャート図であり、ステップＳ４１６からコンテンツ情報の評価値が上位のものに注目し、ステップＳ５０１において、それらの評価値の差が少ないかどうかを判断する。得点差が０か一定値以内の場合、ステップＳ５０２でそれらの上位コンテンツ情報の中で複数のメディア情報源の種類が存在するかどうかを調べる。例えば、テレビのコンテンツ情報とウェブのコンテンツ情報が存在していたとすると、ステップＳ５０３に進み、それらの複数の情報源間で利用時間の傾向が異なるかどうかを判断し、異なる場合にはステップＳ５０４に進み、一つのコンテンツ情報に対してテレビとウェブの利用時間は、たとえばウェブのほうが短いことにより、利用時間傾向の短いウェブのコンテンツ情報を優先して推薦する。ステップＳ５０５でユーザが推薦したコンテンツ情報をチェックし終えたら、ステップＳ５０６に進み、他に上位の推薦コンテンツ情報があるかどうかを調べ、ＹＥＳであれば、ステップＳ５０２に戻り、ステップＳ５０３〜Ｓ５０６の処理を繰り返し、次に利用時間傾向の短いメディア情報源のコンテンツ情報を推薦する。
【００８４】
なお、前記図９の説明ではステップＳ４１６の流れから説明したが、この方法は複数の情報源の推薦において、キーワードの重みを情報源全体における重みと、情報源毎の重みと２種類用意しなくても適応できる方法であり、キーワードの重みからユーザへの推薦度を算出する場合のみに用いることができる。
【００８５】
また、メディア情報源毎ではなく、コンテンツ情報毎に重みを変えてもよい。たとえば、テレビ番組を視聴した場合、そのテレビ番組の番組時間と、利用時間が異なる場合は、重みを変えてもよい。１時間の番組を３０分しか見ていなかった場合は、１時間全部みたときよりも注目度が低いと考えて、重みを半分にしてもよい。
【００８６】
【発明の他の実施の形態】
利用時間を考慮する場合、たとえば図５のステップＳ４１０で履歴得点をカウントする際に、利用時間の極端に短い履歴は履歴としてカウントしないほうが望ましい。これは、テレビの場合、ザッピングしているときが考えられるからである。この場合の履歴を履歴としてカウントする利用時間の閾値は、情報源毎に異なるのが合理的である。
【００８７】
たとえば、テレビのほうがウェブより閾値が長いのは当然である。なぜならば、ウェブのプロ野球記事は数分で読めるが、テレビのプロ野球中継が数分以内というのは利用と言い難い。
【００８８】
以上のことから、メディア情報源毎に利用時間の傾向から閾値を定め、それよりも長い利用時間の履歴はカウントし、短い場合はカウントしないようにしてよい。
【００８９】
【発明の効果】
本発明では、メディア情報源全体のアクセス履歴数を利用することにより、他の情報源に興味ある分野の情報が別の情報源でも推薦することができる。また、全体のアクセス履歴数だけでなく、メディア情報源毎の履歴数も利用して評価値を計算することにより、メディア情報源ごとに利用する情報分野が異なるユーザにも対応することができる。これによって、少ない履歴でも情報源全体での興味を考慮した幅広い情報推薦が可能であり、また、情報源毎の利用傾向も考慮できることから、よりユーザのニーズにあった情報推薦をすることが可能である。
【図面の簡単な説明】
【図１】本発明による情報推薦装置の実施の一形態を示すブロック図である。
【図２】本発明による情報推薦装置の履歴情報格納部から抽出したキーワード履歴の一例であって、キーワード評価テーブル作成途中の状態を示す図である。
【図３】本発明による情報推薦装置のキーワード評価テーブルの一例を示す図である。
【図４】本発明による情報推薦装置のコンテンツ評価テーブルの一例を示す図である。
【図５】本発明による情報推薦装置の処理の流れの一例を示すフローチャート図である。
【図６】本発明による情報推薦装置の履歴情報格納部から抽出したキーワード履歴の別の例であって、キーワード評価テーブル作成途中の状態を示す図である。
【図７】図６に対応するキーワード評価テーブルを示す図である。
【図８】図６に対応するコンテンツ評価テーブルを示す図である。
【図９】本願の請求項４に係る発明に対応する情報推薦装置の処理の流れの一例を示すフローチャート図である。[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to an information recommendation apparatus, an information recommendation method, an information recommendation program, and a recording medium that analyze user preference from history information and recommend information that matches the user preference.
[0002]
[Prior art]
2. Description of the Related Art Conventionally, there is an apparatus and method for observing a user's past usage history, analyzing user's preference from history information, and recommending information based on an analysis result in order to recommend information that matches the user's preference. With this apparatus and method, it is possible to respond to a preference that the user does not notice or change of the user's preference compared to a device and a method that asks the user for the user's preference in advance and recommends information based on the user's answer. , It is possible to perform information recommendation that more closely matches the user's preference.
[0003]
However, many of the conventional information recommendation apparatuses and methods using a history, when recommending information for certain media information, use only the history of information provided from a target information source to perform the information recommendation. Is going. For example, only the TV viewing history is used to recommend recommended TV programs, and the histories of other information sources are not used. However, at present, there are various media information sources, and in addition to televisions, various content information such as the web and e-mail can be easily accessed, so the usage history information of various media information sources can be effectively utilized. Information recommendation apparatuses and methods are desired.
[0004]
In response to the request, a technology has been developed in which access histories of a plurality of media information sources are collectively managed and information is recommended based on the history information (see Patent Document 1 and the like). For example, recommending information on the homepage of a program being watched on TV, recommending a TV program related to a search engine often searched on the web, etc. The information can be recommended according to the preference of the user who did not exist. In other words, keywords are extracted from the information of each media information source, and various media information is used by evaluating whether new information matches the user's preference based on history information of a plurality of information sources corresponding to the keywords. Information recommendation.
[0005]
[Patent Document 1]
JP-A-11-7453.
[0006]
[Problems to be solved by the invention]
However, the apparatus and method for collectively managing the access histories of a plurality of media information sources described in Patent Document 1 and the like are described in Japanese Patent Application Laid-Open No. H11-157, since the weight for each keyword is set the same among a plurality of information sources. Information from the source is recommended with the same recommendation level. For example, just because you frequently watch the talent T on TV, you may not want to watch it on the web. Even in such a case, information including a keyword with a high access history as a whole is However, since it is recommended to the top, the talent T may be recommended to the top even when browsing the web.
[0007]
[Object of the invention]
According to the present invention, history information is shared between a plurality of information sources to perform a wide range of information recommendation, while considering the tendency of the user to use each information source, the information recommendation is performed as much as possible according to the user's desire. It is an object of the present invention to provide an information recommendation device and method which can perform the information recommendation.
[0008]
[Means for Solving the Problems]
In order to solve the above-mentioned problem, an information recommendation device according to claim 1 of the present application includes a unit for extracting a keyword from each information provided by a plurality of information sources, and a weight of a keyword from an access history to all information sources. Calculating means, calculating weights of keywords from access histories in respective information sources, and calculating a recommendation degree for each information based on the calculated weights of the two types of keywords. , And is configured to recommend information that matches the user's preference based on the calculated degree of recommendation.
[0009]
In the above configuration, the information is mainly media information such as television (or radio) viewing information, web browsing information, mail usage information, or electronic payment information, and corresponds to each information such as a TV broadcasting station or the Internet. Provided by information sources. Note that the information on the electronic settlement is credit settlement, Internet shopping, shopping information using electronic money, and the like, and includes information such as a product name and a price.
[0010]
In addition, each piece of information in the media information, for example, in the case of a television, program information or the like is referred to as “content or content information”. The keyword refers to attribute information of each piece of content information and a character string included in each piece of content information. Examples of keywords are the title, cast name, genre name, etc. of a certain television program, and the keywords are extracted based on text information obtained from an EPG (electronic program information).
[0011]
In order to calculate an evaluation value for new information of a certain media information source α, the evaluation value is calculated as follows from the keyword Nx (x = 1 to n) included in the new information and weight information for the keyword. The weight of a keyword is basically determined by the number of appearances, and the weight increases as the number of appearances increases.
[0012]
P (α) =
((Weight of keyword N1 in all media information sources) ∪ (weight of keyword N1 in media information source α))
∪ ((weight of keyword N2 in all media information sources) ∪ (weight of keyword N2 in media information source α))
∪ ・ ・ ・
∪ ((weight of keyword Nn in all media information sources) ∪ (weight of keyword Nn in media information source α))
[0013]
In this way, by using the history of another media information source, it is possible to more accurately grasp the user's preference that could not be understood from the history of a single media information source.
[0014]
In addition, by calculating the weight of each media information source, it is possible to take into account the preference of keywords in the media information source, and it is possible to perform information recommendation that more closely matches the user's preference.
[0015]
The information recommendation device according to claim 2 is configured to extract a keyword from the history content information and to change the weight of the keyword for each content information when calculating the weight of the keyword. As a result, the user's preference for the keyword can be considered, and the user's preference can be grasped more accurately.
[0016]
For example, in a program of a certain TV drama, when calculating the weight of each keyword using "performers" as keywords, in the case of the leading role, the weight of the keywords is increased because there are many scenes appearing in the program, and conversely, In the case of, since the number of appearance scenes in the program is small, the weight of the keyword is reduced. Therefore, the weight of the keyword for a certain performer is determined to be more preferred by the user when the performer is watching the program starring, and the weight of the keyword is increased.
[0017]
The information recommendation device according to claim 3 extracts keywords from content information to be recommended, and considers the importance of each keyword in the content information when calculating a recommendation degree of the content information based on the weight information. , Each of which is calculated by changing the weight of the keyword, so as to calculate a recommendation degree that matches the user's preference.
[0018]
For example, when there is a TV drama program to be recommended, when the recommendation degree is calculated from the weight information using the performer as a keyword, the weight of the leading person is large, and the weight of the supporting person is small, and the recommendation degree is small. Is calculated. That is, in the content information to be recommended, the weight of a person who is considered to have many appearance scenes is increased. As a result, even if there are a plurality of programs in which the user prefers a talent, a program with more appearances is recommended.
[0019]
The information recommendation device according to claim 4, when calculating and recommending an information recommendation degree for a plurality of media information, when there is no difference in recommendation degree between content information having a higher recommendation degree or when a certain degree of content information is recommended. When the score is within the score difference (for example, 1 or 2), the recommendation order is changed and recommended based on the tendency of the use time for each media information source.
[0020]
For example, if the recommendation level of a certain TV program and the recommendation level of a web page do not change much, it is considered that the average usage time (viewing time) of the TV program and the average usage time of the web page are shorter than the usage time of the web page. Web pages are recommended first. After the use is completed, a television program may be recommended. In this way, by making a recommendation based on the usage tendency, it becomes possible to recommend a lot of content information to the user.
[0021]
Embodiment 1 of the present invention
[Information recommendation device]
[Overall basic configuration]
FIG. 1 is a block diagram showing an embodiment of an information recommendation device according to the present invention. In FIG. 1, the information recommendation device has an input unit 1 for receiving a recommendation request, and an arithmetic control unit having a CPU 2, a memory 3, and the like. A first storage unit 10 for a program in which various programs for information recommendation are written, a second storage unit 12 for data for storing various data, a receiving unit 13 for receiving various information, A display unit 15 for displaying recommendation information is provided.
[0022]
The receiving unit 13 is connectable to a plurality of information sources 101, 102, 103, 104, etc., receives various information provided from each of the information sources 101, 102, 103, 104 and transmits the information to the arithmetic control unit 4. It is supposed to. Examples of the information sources 101, 102, 103, and 104 include a television broadcast information source, a web information source, a mail information source related to electronic mail, and an information source such as electronic payment.
[0023]
Various processing means (programs) 6, 7, 8, and 9 used for the information recommendation processing according to the present invention are written in the first storage unit 10 for the program. That is, a keyword extracting means (program) 6 for extracting a keyword from each piece of information received by the receiving unit 13, and a first weight calculation for calculating a weight of the keyword from a history of accesses to all the information sources 101, 102, 103 and 104. Means (program) 7, second weight calculating means (program) 8 for calculating the weight of the keyword for each information source from the access history to each information source 101, 102, 103, 104, and the two calculated types A recommendation degree calculating means (program) 9 for calculating a recommendation degree for each information based on the weight of the keyword is written. The information recommendation means (programs) 6, 7, 8, and 9 are loaded into the memory 3 and refer to each data in the data second storage unit 12 and under the control of the CPU 2. Be executed.
[0024]
The second data storage unit 12 includes a history information storage unit (management unit) 21, a keyword evaluation table 22, a content evaluation table 23, and the like.
[0025]
In the history information storage unit 21, the history of the content information used by the user is stored as text information for each of the information sources 101, 102, 103, and 104. The text information is received by the receiving unit 13 and stored in the history information storage unit 21 as described above. However, when the apparatus has a function of a television, a web, or the like, information received from the outside may be used. May be text content from the broadcast content information received by the receiving unit 13 and store the text information in the history storage unit 21.
[0026]
When the apparatus does not have a function such as a television and a web, the receiving unit 13 receives history information from another device having such functions via a network, and stores the history information in the history storage unit 21.
[0027]
The receiving unit 13 receives new information from the external service information source 105 and selects information recommended to the user from the new information. The new information is, for example, EPG data, new homepage information, recommended homepage information of this week, sales ranking information, unread mail, and the like. Therefore, the receiving unit 13 receives broadcast content information used by the user in the present apparatus, history information of content information used by the user in another device, and new information.
[0028]
FIG. 3 shows an example of the keyword evaluation table 22. The execution of the keyword extraction means 6, the first weight calculation means 7 and the second weight calculation means 8 of FIG. The number of extracted various keywords is written for each information source (integrated) and each information source, and is updated each time information is received.
[0029]
FIG. 4 shows an example of the content evaluation table 23. The execution of the keyword extraction means 6, the first weight calculation means 7 and the second weight calculation means 8 of FIG. The name of the content information, the keyword included in the content information, the type of the information source, the calculated simple score, the score at the information source, and the like are written, and are sequentially updated each time information is received.
[0030]
The keyword entered in the keyword evaluation table 22 of FIG. 3 may be an additional type in which the noun included in the history information is added each time the history information is entered, or a fixed type in which the keyword is determined in advance.
[0031]
[Specific configuration example]
The conceptual configuration has been described above. Next, the information recommendation device will be described with specific examples of content information. In FIG. 1, the history information storage unit 21 stores the history of the used content information in the form of text information as described above. Character information such as a program description is stored. Such character information can be extracted from text information such as an electronic program guide (EPG). When browsing the web, the title and address information of the homepage are stored. In the case of an electronic mail, the title, the address of the transmission / reception destination, the main mail, and the like are stored. In the case of electronic payment information, the name of the purchased product, the name of the store, the amount, and the like are stored.
[0032]
FIG. 2 shows a state before keywords included in the content information stored in the history information storage unit 21 in FIG. 1 are extracted and managed collectively in the keyword evaluation table 22 in FIG. , Web, e-mail, electronic payment) 101, 102, 103, and 104.
[0033]
(Keyword evaluation table)
The keyword evaluation table 22 shown in FIG. 3 collectively manages the keywords included in the content information of television, web, e-mail, and electronic payment. The score based on the history at each information source and the history at all information sources are displayed. It manages both the combined scores. For example, a score of "0" indicates that the information source did not use the content information including the keyword. Here, when the history score is counted, only those that have been used for a certain period of time or more are targeted. If the usage time is extremely short, it is more accurate to not count.
[0034]
Explaining with a specific example, in FIG. 3, the keyword “news” has a total score of 18, but among them, the history score of the information source 104 (television) is 2, and the keyword of the information source 102 (web) is The score is 8, which indicates that the content information including the keyword "news" is more frequently used on the Web than on television. In other words, users often use content information including the keyword "news", but do not watch much news on television, and frequently check news information on the web and news flowing in e-mails. Can be inferred.
[0035]
In addition, content information including a keyword of "music" is often used, but since the score on the television is high, it can be inferred that the user frequently watches a music program.
[0036]
Similarly, the keyword “movie” has a high score in e-mail and electronic payment. From this, we do not often watch movies on TV, but it is guessed that we frequently exchange topics about movies on e-mail, watch movies at movie theaters from electronic settlement information, rent rental videos, etc. it can.
[0037]
As described above, even in matters of interest to the user, it is considered that the information sources used for the matters are biased, and by considering the bias of these information sources, the present invention provides a method of using the information sources of the user. According to the tendency, information can be recommended.
[0038]
The keyword evaluation table 22 extracts a keyword from the text information of the content information stored in the history information storage unit 21 of FIG. , 8, the number of histories is updated for each keyword and each media information source as shown in FIG.
[0039]
For example, when watching a television program, keywords are extracted from the title, genre, performer name, program description, and the like. The program description may be subjected to morphological analysis, and the nouns appearing may be extracted as keywords. When a music program is viewed, the keyword "music" is extracted from the genre name and added to the total score and the TV score.
[0040]
From the web browsing history, keywords are extracted from the title and text information of the viewed homepage and added to the total score and the web score, respectively. Morphological analysis may be performed on text information, and nouns may be extracted as keywords. A word having a large number of appearances may be preferentially used as a keyword. For example, when a newspaper article is viewed on the web, a score is added for the keyword “newspaper”.
[0041]
From the mail use history, the title and body of the mail are morphologically analyzed, keywords are extracted, and the total and the score of the mail are added for each keyword.
[0042]
From the information of the electronic settlement, a score is added as a keyword such as a product name, a store name, and an amount.
[0043]
In the above description, information such as television, web, e-mail, and electronic payment is used as an information source. However, the information source is not limited to these information sources. If you can get the keyword, you can also get the keyword from it.
[0044]
(Content evaluation table)
In the content evaluation table 23 shown in FIG. 4, for example, the content information "news 23" is a television program including the keyword "news", and therefore, referring to the keyword evaluation table 22 in FIG. A simple score and a TV score are calculated from the total history score and the TV history score.
[0045]
When new content information is received, a keyword is extracted from the content information, and a score is calculated by using the first and second weight calculation programs 7 and 8 in FIG. 1 and referring to the keyword evaluation table 22 in FIG. . The extraction of the keyword of the content information is as described above.
[0046]
In FIG. 4, a method of calculating a score in the case of “news 23” will be specifically described. The simple score X of the content evaluation table of the content information including N keywords is

When the content information is "news 23" as described above, the content includes the keyword "news", and the keyword evaluation table 22 of FIG. It becomes.
[0047]
On the other hand, the score Y of one information source α is

When the information source is a television, since the history score of the keyword of “news” on the television in the keyword evaluation table 22 in FIG. 3 is 2, 18 + 2 = 20, and “news 23” in FIG. Has a score of 20 on TV.
[0048]
The score of the content evaluation table 23 of FIG. 4 is changed each time the keyword evaluation table 22 of FIG. 3 is changed. It is also calculated each time new content information is received.
[0049]
(How to select recommended content information)
Referring to the content evaluation table 23 of FIG. 4, the content information recommended to the user is selected by using the recommendation degree calculating means 9 of FIG. That is, when there is a recommendation request from the input unit 1 by the user, content information having a high evaluation score is preferentially recommended from the content evaluation table 23 of FIG. At this time, the selection is made not based on the simple score but on the score at each information source.
[0050]
For example, if the recommendation request from the user is “I want you to tell me a recommended TV program”, the content information with the highest score among the scores of the TV in FIG. 4 is selected. For example, the score of “Music Station” is 24 while the score of “News 23” is 20 on the TV, so “Music Station” is selected.
[0051]
In the simple score, “News 23” is 18, “Music Station” is 16, and “News 23” is higher. This is because the keywords “News” and “Music” included in the content information are used. This is because “News 23” having “News” as a keyword has a higher history score.
[0052]
In the prior art, the one having such a large total number of histories is preferentially recommended. In the present embodiment, however, from the keyword evaluation table 22 of FIG. 3, “news” has a low score on the television. On the contrary, it can be seen that "music" has a high score on television. Thus, when recommending the content information of the television to the user, it is presumed that the content information including “music” may be given priority.
[0053]
By using the histories of multiple information sources, there is a possibility that information that matches the user's preference can be recommended even when the history information is small, but the user can use various media information sources, There is a preference for each user, and a recommendation method that takes that into consideration is desirable. The field of interest may change for each media information source, and it is desired to be able to cope with such a case.
[0054]
Therefore, in the present invention, by making an evaluation score in consideration of the total number of histories and the number of histories in the information source, the recommendation is performed in consideration of interest in the entire media information source and interest in a certain information source. It becomes possible. There is a possibility that it is possible to respond to recommendation of various media information sources by using the bias of the overall interest and the bias of interest among the media information sources.
[0055]
If the recommendation request from the user does not specify an information source but indicates "I want you to give recommended content information", the highest score among the information sources in FIG. 4 is given priority. In this case, “e-mail magazine No. 02” with a score of 38 is recommended.
[0056]
If the user does not specify the information source, the recommendation may be made according to the situation such as the time and the terminal used. For example, if the device used by the user does not have a TV function, select from items excluding TV content information, or if the current environment of the user is a situation where sound cannot be output, the mail content information is selected. Priority can be considered.
[0057]
As described above, the simple score is used for explanation for comparison with the conventional technique, but is not actually used.
[0058]
[Flow of information recommendation process]
FIG. 5 is a flowchart illustrating an example of the flow of the information recommendation process according to the present invention. In step S401, if a recommendation request by the user, reception of new arrival information, access history information of the user, or the like is input, the process proceeds to step S402, and it is determined whether the input is a recommendation request. The process proceeds to step S403 to determine whether the information is new information (that is, access history information).
[0059]
In the case of new information, the process proceeds to step S404, one content information is extracted from the new information, and in step S405, a keyword included in the extracted content information is extracted. As described above, the method of extracting the keyword is to extract a noun included in the text information and use it as a keyword, or to select a keyword existing in the keyword evaluation table 22 (FIG. 3) and included in the text information. You may search.
[0060]
In step S406, the score extracted in the information source of the content information is calculated with reference to the keyword extracted in step S405 and the keyword evaluation table 22 in FIG. The method of calculation is as described above. In step S407, the score is added to the score of the corresponding information source in the content evaluation table 23 of FIG. Note that the simple score of the content information has been described in the above description for comparison with the prior art, but it is not actually used, and thus need not be calculated. The above processing (S404 to S407) is performed for each piece of content information. When it is determined in step S408 that the processing of all pieces of content information has been completed, the process ends.
[0061]
If it is determined in step S403 that the information is not new information, that is, if it is access history information, the process proceeds to step S410, in which one keyword is extracted from the accessed content information by the keyword extracting means 6, and the keyword evaluation table shown in FIG. Change 22. That is, in step S411 of FIG. 5, the count of the total score of the keyword is increased by the first weight calculating means 7, and in the step S412, the score of the information source is increased by the second weight calculating means 8. If there are a plurality of keywords, the process of steps S410 to S413 is repeated. If it is determined in step S413 that the processes for all the keywords have been completed, the process proceeds to step S414, and the content evaluation table 23 of FIG. change. That is, the score of the content information including the keyword and having the same media information source is increased in the media information source, and the process is terminated.
[0062]
If the request is an information recommendation request in step S402 in FIG. 5, the processing in steps S415 to S418 is performed using the recommendation degree calculating means 9. That is, in step S415, it is determined whether or not the information recommendation request includes designation of a media information source. If specified, the process proceeds to step S417, where recommended information is selected from the specified information source, and if not specified, the process proceeds to step S416, where recommended information is selected from all information sources.
[0063]
As described above, the method of selecting information is to refer to the content evaluation table 23 in FIG. 4 and preferentially select content information having a high score in the information source. For example, in the case of a TV program, the priority may be changed according to the function of the device being used, such as raising the priority of content information near the current time, raising the priority of newer information, or the situation of the user. Is also good. Then, when the recommendation information is selected in step S416 or step S417, the information is recommended in step S418, and the process ends.
[0064]
As described above, an example of the processing flow of the present invention has been described. As described above, by using the usage histories of a plurality of information sources, information of a field of interest used by another information source can be stored in another information source. However, it is possible to make an recommendation, and it is possible to make an information recommendation that matches an individual's interest even in an information source with a small history.
[0065]
Further, according to the present invention, it is possible to perform information recommendation in consideration of the tendency of the user to use the information source by calculating the evaluation value using not only the entire access history but also the history of each information source. As a result, better information recommendation can be performed for users who use different information fields for each media information source.
[0066]
Embodiment 2 of the present invention
This is an embodiment corresponding to the invention according to claim 2 of the present application. In the first embodiment, when calculating the weight of a keyword from history information, a score is calculated only by the number of appearances of a keyword. In the second embodiment, the keyword extracting means 6 shown in FIG. 1 has a function of performing weighting for each keyword in order to change the weight of the keyword according to the content information. The recommendation method will be described based on the flowchart of FIG.
[0067]
In step S410 in FIG. 5, for example, in the case of the viewing history of a TV drama, keywords are extracted from the title, genre, and program information in the EPG data. Among the performers, the weight of the leading person is large, and the weight of the supporting person is small.
[0068]
For example, referring to FIG. 6 and FIG. 7, the score is simply 2 because the keyword is actor B in the history of the television and the user has watched twice on the television in FIG. 3 described above. In the second embodiment, if it is the supporting class twice, the weight is set to 0.2 once and the player is watched twice, so that the score of actor B is 0.4 as shown in FIG. Accordingly, in FIG. 7, the total score of the actor B is 1.4, and the total score of the TV is 0.4, and the total score of the TV is 0.4.
[0069]
As described above, there are many appearance scenes in the program, and the weight of the person watching the user is increased. As a result, the weight of the person watching the user is increased, and the recommendation of the program in which the person appears is increased. The degree increases.
[0070]
For example, suppose that the user likes talent T1 and talent T2, and has watched dramas D1 and D2, each of which starred. Furthermore, when the talent T3 appears as an assist in both dramas D1 and D2, the weight of the talent T3 increases if the count is simply performed for each history. As in the embodiment, it is effective to add a change to the weight of the keyword.
[0071]
It should be noted that the determination as to whether or not it is the lead is made based on the person whose name appears first in the cast of the program information. In the EPG data, performers are listed first in the program information, followed by a description of the program. Not only TV drama but also movie.
[0072]
The recommendation processing according to the second embodiment does not necessarily need to use the weights of the two types of keywords, and is not limited to recommendation of a plurality of media information sources, but is also recommended for recommendation within a single media information source. Can adapt. Therefore, a keyword can be extracted from the history information, its weight can be calculated, and it can be used only when grasping the user's preference. The flow of the processing in that case is the processing from steps S410 to S414 in FIG.
[0073]
Third Embodiment of the Invention
This is an embodiment corresponding to the invention according to claim 3 of the present application, and is an information recommendation device that recommends information that matches a user's preference based on a keyword indicating the user's preference. As No. 9, there is provided means for calculating a recommendation degree by changing the weight for each of the keywords included in the information to be recommended for each of the keywords included in the information.
[0074]
That is, a keyword is extracted from the new content information in step S405 in FIG. 5, and the weight of the keyword in the content information is considered when the content information is evaluated in step S406 in FIG. 5 with reference to the keyword evaluation table 22 in FIG. And evaluate.
[0075]
Specifically, when the content information is a TV drama program, the weight of the leading person is increased and the weight of the supporting person is reduced, and the evaluation value of the content information is calculated. For example, when the movie is evaluated in FIG. 8, the weight of the actor B is 3 according to the keyword evaluation table 22 in FIG. If it is an auxiliary performance, in the third embodiment, the weight is set to 1/2 each. Thereby, the evaluation value is calculated by the following equation.
[0076]

[0077]
When this method is used, for example, when the user likes the talent T and the weight is high, when there are a plurality of programs in which the talent T appears, the recommendation degree of the program in which the talent T is starred is increased. Is possible.
[0078]
The method of determining which is the leading role has been described in the second embodiment. In both cases, the explanation is made by increasing the weight of the person who plays the main role of the TV drama, but the present invention is not limited to the TV drama or movie. For example, when the title of the television shows that the talent T is the center, such as “cooking program of the talent T”, the weight of the talent T may be increased.
[0079]
In order to cut out the talent T from the title, the talent T can be cut out using a personal name dictionary, or when the name of the talent T is listed as a performer in the place of the program description, the talent T can be cut out therefrom.
[0080]
As in the second embodiment, this method does not need to use the weights of the two types of keywords, and can be used not only for recommending a plurality of media information sources, but also for recommending within a single media information source. Can adapt. Therefore, it can be used only when the degree of recommendation to the user is calculated from the weight of the keyword. The flow of the processing in that case is from step S404 to S409 in FIG.
[0081]
Embodiment 4 of the present invention
This is an embodiment corresponding to the invention according to claim 4 of the present application, and the recommendation degree calculating means 9 in FIG. 1 uses the recommendation degree for each information source when the difference in recommendation degree between each information is 0 or within a certain value. It has a function to change the recommendation order based on the tendency of time.
[0082]
That is, in step S416 of FIG. 5, when information recommendation is performed from all information sources, if the difference between the higher recommendation degrees is within a certain value, the recommendation is made from the information of the media information source having a short usage time tendency.
[0083]
FIG. 9 is a flowchart specifically showing the recommendation method. From step S416, attention is paid to those having higher evaluation values of the content information, and in step S501, it is determined whether or not the difference between the evaluation values is small. If the score difference is 0 or within a certain value, it is checked in step S502 whether or not there are a plurality of media information source types in the upper content information. For example, if there is TV content information and web content information, the process proceeds to step S503, and it is determined whether the usage time trends among the plurality of information sources are different. If not, the process proceeds to step S504. As the use time of the television and the web for one piece of content information is shorter in the web, for example, the content information of the web having a shorter usage time tendency is preferentially recommended. After checking the content information recommended by the user in step S505, the process proceeds to step S506, and it is checked whether there is another recommended content information. If YES, the process returns to step S502, and the processes in steps S503 to S506 are performed. Is repeated, and content information of a media information source having a short usage time tendency is recommended.
[0084]
Note that, in the description of FIG. 9, the flow of step S416 has been described. However, in this method, in recommending a plurality of information sources, two types of keyword weights, that is, a weight for the entire information source and a weight for each information source, are not prepared. This is a method that can be applied even when the degree of recommendation to the user is calculated from the weight of the keyword.
[0085]
Further, the weight may be changed for each content information instead of for each media information source. For example, when watching a television program, if the program time of the television program is different from the usage time, the weight may be changed. If the user watches only one hour of the program for 30 minutes, the weight may be reduced to half assuming that the degree of attention is lower than that of watching the entire hour.
[0086]
Another embodiment of the present invention
When the usage time is taken into consideration, for example, when counting the history score in step S410 in FIG. 5, it is desirable not to count the history having an extremely short usage time as the history. This is because in the case of a television, zapping can be considered. In this case, it is reasonable that the threshold value of the usage time for counting the history as the history is different for each information source.
[0087]
For example, it is natural that television has a longer threshold than the web. The reason is that you can read professional baseball articles on the web in a few minutes, but it is hard to say that a live broadcast of a professional baseball on a television in less than a few minutes.
[0088]
From the above, a threshold value may be determined for each media information source based on the tendency of the usage time, and the history of the usage time longer than that may be counted, and if the usage time is shorter, the counting may not be performed.
[0089]
【The invention's effect】
According to the present invention, by utilizing the number of access histories of the entire media information source, information in a field of interest to another information source can be recommended by another information source. Also, by calculating the evaluation value using not only the total number of access histories but also the number of histories for each media information source, it is possible to cope with users who use different information fields for each media information source. As a result, it is possible to recommend a wide range of information considering the interests of the entire information source even with a small history, and it is also possible to consider the usage tendency of each information source, thereby making it possible to recommend information more suited to the needs of the user. It is.
[Brief description of the drawings]
FIG. 1 is a block diagram showing an embodiment of an information recommendation device according to the present invention.
FIG. 2 is an example of a keyword history extracted from a history information storage unit of the information recommendation device according to the present invention, showing a state in which a keyword evaluation table is being created.
FIG. 3 is a diagram showing an example of a keyword evaluation table of the information recommendation device according to the present invention.
FIG. 4 is a diagram showing an example of a content evaluation table of the information recommendation device according to the present invention.
FIG. 5 is a flowchart illustrating an example of a processing flow of the information recommendation device according to the present invention.
FIG. 6 is a diagram showing another example of the keyword history extracted from the history information storage unit of the information recommendation device according to the present invention, showing a state in which a keyword evaluation table is being created.
FIG. 7 is a diagram showing a keyword evaluation table corresponding to FIG. 6;
FIG. 8 is a diagram showing a content evaluation table corresponding to FIG. 6;
FIG. 9 is a flowchart showing an example of the flow of processing of the information recommendation device according to the invention of claim 4 of the present application.

Claims (7)

An information recommendation device that manages a user's past history with respect to information provided from a plurality of information sources, estimates a user's preference from history information, and recommends information that matches the user's preference,
Means for extracting a keyword from each of the provided information;
Means for calculating keyword weights from access histories to all information sources;
Means for calculating the weight of the keyword from the access history at each information source;
Means for calculating a recommendation degree for each information based on the calculated weights of the two types of keywords,
An information recommendation device characterized by recommending information matching the user's preference based on the calculated recommendation degree. An information recommendation device that manages a user's past history with respect to content information provided from an information source, estimates user preference from the history information, and matches information that matches the user preference,
Means having a function of extracting keywords from content information used by the user and weighting each extracted keyword;
A means for calculating a recommendation degree for each piece of information based on a keyword included in the information to be recommended;
An information recommendation device characterized by recommending information that matches a user's preference. An information recommendation device that recommends information that matches a user's preference based on a keyword indicating the user's preference,
An information recommendation apparatus comprising: means for calculating a recommendation degree by changing a weight for information to be recommended for each keyword included in the information. An information recommendation device that recommends information that matches a user's preference with respect to information provided from a plurality of information sources based on a keyword indicating the user's preference,
An information recommendation apparatus having a function of changing a recommendation order based on a tendency of use time for each information source when a difference between recommendation degrees among respective information is within a predetermined size. An information recommendation method for managing a user's past history for information provided from a plurality of information sources, estimating a user's preference, and recommending information that matches the user's preference,
Extracting a keyword from each information, calculating a keyword weight from the number of access histories to all information sources, and calculating a keyword weight from the number of access histories in each information source,
An information recommendation method characterized by recommending information matching the user's preference based on the calculated weights of the two types of keywords. An information recommendation program for operating a computer as the information recommendation device according to claim 1. A recording medium on which the information recommendation program according to claim 6 is recorded.

Images