WO2010095169A1 - 情報推薦方法、そのシステム、及びサーバ - Google Patents
情報推薦方法、そのシステム、及びサーバ Download PDFInfo
- Publication number
- WO2010095169A1 WO2010095169A1 PCT/JP2009/000666 JP2009000666W WO2010095169A1 WO 2010095169 A1 WO2010095169 A1 WO 2010095169A1 JP 2009000666 W JP2009000666 W JP 2009000666W WO 2010095169 A1 WO2010095169 A1 WO 2010095169A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- item
- interest
- unrecommended
- boundary
- indifference
- Prior art date
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9537—Spatial or temporal dependent retrieval, e.g. spatiotemporal queries
Definitions
- the present invention relates to a method for recommending information such as content, and more particularly to a technique for recommending an unexpected item that is of interest to a user but cannot be easily assumed.
- a “collaborative filtering” method is known as a system for recommending information to be distributed when distributing information such as content (see Non-Patent Document 1).
- This collaborative filtering is a method of estimating a user profile (what type a user belongs to, etc.) from the similarity of past user histories and determining the next recommended item.
- Patent Document 1 a preference cluster and a non-preference cluster are created based on content that the user has viewed and content that has not been viewed, respectively, and are recommended in order from the one that is similar to the preference cluster and that is not similar to the non-preference cluster. The method is described.
- JP 2008-210101 A Toshiyuki Masui, “City Corner of Interface (93)-Bookshelf Operation,” Unix Magazine, Vol. 20, No. 12, 2005
- the collaborative filtering method is mainly a method of determining a recommended item with reference to an unrecommended part in the entire history of another user whose history is partially similar, and in general, an unexpected recommendation, that is, an interest to the user.
- an unexpected recommendation that is, an interest to the user.
- the recommendation degree is determined by referring to only the cluster of the content having the closest similarity, and when there are other similar clusters, it cannot be reflected. .
- the recommendation results are the same regardless of whether the preference is the same.
- the recommendation method is based on what is considered to be the most preferred, it is not possible to increase the recommendation order of content far from the preferred content. For this reason, it is difficult to make recommendations that are surprising.
- it is not considered to reflect the background situation (physiological phenomenon, human relationship, etc.) that occurs in relation to user behavior in the recommended content.
- An object of the present invention is to provide an information recommendation method capable of making a recommendation that is unexpected for a user, a system thereof, and the like.
- Another object of the present invention is to provide an information recommendation method that can reflect the background situation that occurs in relation to user behavior in the recommended content, and its system.
- a server including a processing unit and a storage unit is used, and the reaction of the user who has received information about the item is reflected in the selection criteria for the item to be recommended next.
- An information recommendation method and an information recommendation system for recommending in which a processing unit collects information that a user has reacted and information that there has been no reaction, and an item of interest that the user has reacted to an unrecommended item
- Information recommendation method and information recommendation system for determining the next recommended item to the user using the comparison result I will provide a.
- the processing unit calculates the recent interest distance and the recent indifference distance for the unrecommended item, calculates the boundary divergence degree of the unrecommended item from the difference between the recent interest distance and the recent indifference distance, and the boundary divergence degree is the smallest.
- An information recommendation method for determining an unrecommended item as the next recommended item is provided.
- the processing unit extends the unrecommended item similar to the interested item and the indifferent item as the interested item and the indifferent item, so that it is a boundary between the interested item and the indifferent item in all items to be recommended.
- an information recommendation method for estimating an indifference / interest boundary and determining an unrecommended item on the interest item side determined on or near the indifference / interest boundary or an unrecommended item on the indifference / interest boundary as a next recommended item.
- the processing unit uses attribute information (referred to as a field context) about a user's physiological phenomenon and a human relationship with the accompanying person. Estimate and provide an information recommendation method that reflects the context of the place in the item recommendation.
- attribute information referred to as a field context
- primary recommendation is performed for recommending information based on action history and a fixed user profile, and not only items that have been reacted but also information on items that have not been reacted are used together. Narrow down the recommended items.
- the result of primary recommendation is used to identify where the user's interest / indifference boundary is located in all items by similarity, and as a result, unrecommended items become user's interest / indifference. Identify where on the region. Then, as an item that is not particularly indifferent and maximizes the unexpectedness, an item near the boundary between interest and indifference is determined as a recommended item.
- the present invention it is possible to make a recommendation that is surprising to the user, and it is possible to reflect the background situation that occurs in relation to the user behavior in the recommended content.
- FIG. 1 is a diagram showing an overall system configuration in which a recommendation method and a recommendation system are executed.
- a user 100 possesses a terminal 101 such as a mobile phone, and uses this terminal 101 to access a server 104 via a network 103 via a network connection device 102 such as a wireless communication device.
- the server 104 incorporates a recommendation program, which will be described in detail later, and performs a recommendation / distribution process of contents and the like to the user 100.
- the server 104 has a configuration in which a database such as the item / attribute information 105, the user profile 106, and the user action history 107 is externally attached or built in.
- FIG. 2 is a diagram showing an example of the internal configuration of the terminal 101 in FIG.
- a display 200, an input device 202, and an antenna 204 are connected to an internal bus 206 such as a data bus via a video display unit 201, an interface (IF) 203, and a wireless communication unit 205, respectively.
- an arithmetic device 207 including a central processing unit (Central Processing Unit) and a memory 208 serving as a storage unit are connected to the internal bus 206.
- the memory 208 stores a display program for generating data to be displayed on the display 200.
- Central Processing Unit Central Processing Unit
- FIG. 3 is a diagram showing an example of the internal configuration of the server 104 in FIG.
- reference numerals 300, 301, 302, 303, 304, 305, 306, 307, 308 denote a display device, a video display unit, an input device, an interface (IF), an input device, an interface (IF), a data bus, etc.
- the server 104 is an ordinary computer.
- the server 104 is an arithmetic unit including a CPU, a memory, and a memory. The recommended program in the memory will be described in detail later.
- the storage device 310 connected to the internal bus 306 via the interface (IF) 309 stores the item / attribute information 105, the user profile 106, and the user action history 107 shown in FIG. 1 as a database.
- the server is connected to the network 103 via the communication unit 311.
- the first embodiment is an example of a recommendation method and a recommendation system based on a vector space type item classification.
- the similarity relationship expression of items is used as a spatial arrangement (vector) type, and an interest / indifference boundary region is calculated to recommend an item.
- a vector space is generated by expressing all items as vectors according to their attributes, and an interest / indifference boundary is determined in the vector space using the primary recommendation result, and items existing on or near the boundary.
- the server 104 having the system configuration shown in FIGS. 1 and 3 recommends various items such as contents to the terminal 101 possessed by the user 100 via the network, and further distributes them.
- all items to be recommended / distributed are regarded as item vectors, and various items are arranged in a vector space.
- the item vector is created using the item attributes as components.
- each genre is expressed as a number from 1 to 10
- price range is expressed as a number from 1 to 5
- purchase layer is also expressed as a number from 1 to 5.
- a set of three numbers is given to one item.
- the genre is 7, the price range is 3, the purchase layer is 5, and so on.
- the vector of the item is given by three components (three dimensions) of (7, 3, 5). It goes without saying that the same applies to the case of three or more components.
- the vector (7, 3, 5) may be expressed by assuming that the vector (7, 3, 5) is arranged at a corresponding position in the three-dimensional space.
- FIG. 4 is a diagram illustrating an overall flowchart in the server 104 of the recommendation system according to the first embodiment.
- This flow is a processing flow in the arithmetic unit 307 which is a processing unit of the server 104.
- an item vector is created based on the attributes of all recommended items as described above (401).
- a similarity is calculated for each item, and primary recommendation using the similarity is performed (402).
- the primary recommendation result one or more recommendations
- the boundary divergence degree of the unrecommended data is determined using the accumulated data (403).
- a secondary recommended item is determined using the boundary divergence degree (404), and the process ends (405).
- the primary recommendation using the similarity may use, for example, conventional collaborative filtering.
- the determination flow of the boundary divergence degree is started (500), the reaction data for each item that has been primarily recommended (402) is checked (501), and the presence / absence of an interest reaction within a certain time is checked (502).
- the item is classified as an interested item (503).
- the item is classified as an indifferent item (504) and accumulated, and all data are finished ( 505, 506), vectors of interest items and indifferent items in the item space are set as initial positions of interest and indifference areas (regions) (507).
- the presence or absence of this reaction is the presence / absence of access to the access destination corresponding to the primary recommended item, the presence / absence of electronic payment of the product related to the primary recommended item, the download of the discount ticket etc. of the product corresponding to the primary recommended item, etc.
- the processing unit can detect log information that can be regarded as a user's response to the primary recommended item.
- the nearest interest distance (the nearest initial position of interest, that is, the distance to the nearest interest item vector) is calculated for each unrecommended item (508), and similarly, the most recently uninterested distance (recent (509) calculates a signed boundary divergence (the difference between the nearest interest distance and the most recent indifference distance) for each unrecommended item. Calculate (510) and end the step of determining the boundary divergence (511).
- the Euclidean distance is used as the distance between the item vectors.
- ⁇ ⁇ indicates the square root of the value in ⁇
- ⁇ 2 indicates the square.
- the closest distance among the initial positions of the interest data is selected and set as the latest interest distance as described above.
- the closest distance among the initial positions of the indifference data is selected and set as the indifference distance recently. Then, by taking the difference between the two, a signed boundary divergence degree is obtained.
- the sign is a sign of (Recently Indifference Distance-Recently interesting Distance).
- different weights may be assigned to each. Giving a greater weight to the distance on the interested side sets a boundary closer to indifference.
- FIG. 6 is a flowchart showing details of the step of determining the secondary recommended item.
- the flow starts (600) first, data having the smallest absolute value of the signed boundary divergence is selected from the unrecommended items (601). It is determined whether there is only one item (602). If there are a plurality of items, selection is performed in the following order until there is one item (603). That is, only those with a positive sign of the signed boundary divergence degree are selected. (Recent distance of indifference)-(Recent distance of interest)> 0 If there is not one item, random number selection is performed and the selection result is determined as a recommended item (605).
- step 602 If there is only one item in step 602, the item is determined as a secondary recommended item (604), and the process ends (606).
- FIG. 7 is a diagram schematically showing an example of the relationship between the item vector and the interest / indifference boundary in this embodiment.
- 700 ⁇ mark
- 701 x mark
- 702 ⁇ mark
- Reference numerals 703 and 704 respectively indicate the recent interest distance and the latest indifference distance between the interested item 700 and the unrecommended item 702.
- Reference numerals 705 and 706 denote interest / indifference boundaries and signed boundary deviation degrees, respectively.
- the signed boundary divergence degree 706 is schematically illustrated in FIG. 7, but can be calculated as described above.
- FIG. 8 is a diagram showing an example of item / attribute data used in this embodiment. This data is stored in the storage unit as item / attribute information 105 in FIG.
- each row indicates each item, and each column indicates an attribute. Examples of the attribute include the genre, price range, purchase layer, season, and the like described above. Further, as shown in the rightmost column, whether or not it is related to a characteristic keyword can also be used as an attribute.
- FIG. 9 is a diagram showing an example of item recommendation history data in the present embodiment.
- Each row of the table 900 indicates each recommended item, and each column indicates the recommended time, content, and the like of the recommended item. The time, place, and user information can be used as background information for reference when making recommendations.
- the rightmost column shows the user's response to the recommended recommendation item, where 1 indicates a response and 0 indicates no response, which is used as a result of the primary recommendation.
- this table 900 is memorize
- the similarity relation expression of items is a spatial arrangement type using vectors
- the interest / indifference boundary region is calculated in the vector space
- the items existing on or near the boundary are selected.
- the evaluation value fluctuates greatly near the interest / indifference boundary.
- the boundary divergence degree by using the boundary divergence degree, the evaluation is sequentially performed in the vicinity of the boundary. Items can be recommended from near the boundary.
- a graph configuration type that is, a primary recommendation result is used, and an interest / indifference boundary and a boundary divergence degree of each item are calculated in the item graph. It is.
- the response item and the non-reaction item which are the results of the primary recommendation, are used as the starting point to expand each item by following the links on the item graph.
- FIG. 10 is a diagram showing an overall flowchart of the second embodiment. Needless to say, this flow is also executed in the server 104 having the system configuration shown in FIGS. 1 and 3 as in the first embodiment.
- FIG. 10 when the processing flow is started (1000), an item graph is created based on the attributes of all recommended items (1001). Then, primary recommendation using similarity is executed (1002). Then, the interest / indifference boundary and the boundary divergence degree of each item are calculated in the item graph using the primary recommendation result (1003), the secondary recommended item is determined based on the result (1004), and the process ends (1005). ).
- FIG. 11A is a flow showing details of the creation (1001) of the item graph of FIG.
- the similarity between each item uses the normalized correlation between attribute vectors, for example.
- Cor (x, y) (y1 ⁇ x1 + y2 ⁇ x2 +... + Yd ⁇ xd) / [ ⁇ ⁇ (x1) ⁇ 2+ (x2) ⁇ 2+... (Xd) ⁇ 2 ⁇ ⁇ ⁇ (y1) ⁇ 2+ (y2) ⁇ 2+... (Yd) ⁇ 2 ⁇ ] It becomes.
- ⁇ ⁇ indicates the square root of the value in ⁇
- ⁇ 2 indicates the square
- • indicates multiplication
- / indicates division.
- step 1103 for example, only a part having a similarity equal to or higher than a preset threshold (for example, 0.5) is assumed to be linked.
- FIG. 11B to FIG. 11D schematically show the above-described processing of this embodiment.
- 1 to 5 indicate item i.
- FIG. 11C shows normalized correlation values 0.1 to 0.9, which are similarities between items 1-5.
- FIG. 11D shows what is linked only to the above-described threshold (0.5) or more.
- the boundary divergence degree calculating step 1003 in FIG. 10 will be described in detail.
- a check is made on each reaction data for data that has been reacted by the user as a result of the primary recommendation (1201). First, it is checked whether or not there is an interest reaction within a certain time for each data (1202). If YES, the item is classified as an interested item (1203). If NO, the item is classified as an indifferent item. (1204). Then, it is confirmed whether or not all data has been completed (1205). If not, the next reaction data is checked (1206).
- the node of the interested item and the node of the uninteresting item in the item graph are set as initial values of the interested and uninteresting areas, respectively (1207). Then, the interest / indifference area is expanded to each adjacent node on the graph by the adjacent relationship (1208).
- one extension operation to adjacent nodes is referred to as (one) extension step.
- the node was expanded for the first time in the previous expansion step, or was a node that had not been expanded until the previous time, and expansion from both the interested and indifferent sides overlapped in this expansion step.
- the interest / indifference boundary item that is, the interest / indifference boundary divergence degree is set to zero (1209).
- the interest / indifference boundary item determined above is expanded as an initial node, and the i-step expansion is performed, and the interest / disinterest boundary divergence degree of the first included node is set to i.
- FIGS. 13A to 13D schematically show specific examples of Step 1207 to Step 1210.
- FIG. 13A the nodes of the interested item and the uninterested item are set as initial values of the interested and indifferent areas, respectively.
- the ⁇ mark, the X mark, and the ⁇ mark in the items indicate the primary recommended interest item 1300, the indifferent item 1301, and the unrecommended item 1302, respectively, as in FIG.
- FIG. 13B the interest and indifference areas are expanded to adjacent nodes on the graph by the adjacency relationship.
- the upper stage, middle stage, and lower stage of FIG. 13B show the state after the third expansion, the fourth expansion, and the fifth expansion, respectively.
- items indicated by * and ⁇ at each node represent items 1303 and 1304 after the indifferent item and the item of interest are expanded, respectively.
- FIG. 13C a node that has been expanded for the first time before or once but has not been expanded, and a node in which both expansion of interest and indifference overlap is set as an interest / indifference boundary item. That is, the interest / indifference boundary divergence degree is 0, and three items surrounded by a dotted line in FIG.
- the numerical value other than the divergence degree 0 indicates the interest / indifference boundary divergence degree i of the node included for the first time by expanding i steps from the interest / indifference boundary item.
- the numerical value N for determining the degree of divergence is a parameter given in advance.
- the number of expansion steps may be different between the areas of interest and indifference, and one of them may be increased.
- the number of steps from the unrecommended item to the node and expanded to reach the interested item and the uninterested item for the first time is set as “recent interest distance”, “ As the “recent indifference distance”, the difference may be defined by looking at the difference between them.
- FIG. 14 is a diagram for explaining a modification of the secondary recommended item determination step 1005 of FIG. 10 in the present embodiment. Note that the item graph is the same as in FIG. 13C.
- the recommendation is given priority on the vicinity of the boundary on the graph expression.
- all items for example, three items
- step 1406 the next processing is repeated until there is one item (1406).
- step 1406 selects one with a small degree of interest / indifference boundary divergence.
- step 1407 Selects the item with the maximum average number of distances from all recommended items.
- step 1407 Selects the one with the smallest average number of separations of all reaction items.
- step 1407 selects the selection result is determined as a recommended item (1407), and the process ends (1411).
- the number of steps from the unrecommended item to the node is expanded to reach each of the interested item and the uninterested item for the first time.
- the difference may be defined by looking at the difference between them.
- the evaluation value largely fluctuates in the vicinity of the interest / indifference boundary.
- the vicinity of the boundary can be recommended from the vicinity of the boundary sequentially by continuous evaluation.
- the field context data is data whose basic information is a physiological phenomenon parameter, a human relationship parameter, and a user profile.
- the “physiological phenomenon parameter” is a parameter related to human senses (requests to eat, rest, cold, hot, painful, dark pains, etc.) and emotion (healing pleasure). It means the time of the previous meal, estimated calorie intake, walking distance, and the type of service used recently (eg, movie genre or type).
- the “human relationship parameter” is a parameter related to the human relationship of the user's companion, such as a lover, wife, husband, family, friend, and the like.
- This parameter is estimated from, for example, the communication status (call frequency, mail frequency, common community participation frequency). Whether or not the user is accompanied by a companion is determined by whether or not position information obtained by a GPS or the like mounted on the information terminal held by the user is near a certain time.
- the “user profile” can use information registered in advance such as sex, age, etc. when registering an information terminal.
- FIG. 15 shows an example of the overall system configuration in which the recommendation system according to this embodiment is executed, and the same reference numerals as those in the system of FIG. 1 denote the same items.
- the server 150 stores a field context application program therein, and a field context database 151 that details the contents later is added to the database storing various tables.
- the configuration is the same as that of the system described with reference to FIGS.
- the server 150 estimates the above-described user profile, organizing phenomenon parameter, and human relationship parameter (1601).
- the profile uses information at the time of user registration or the like.
- the physiological phenomenon parameter is estimated from the movement distance of the mobile terminal, the status of various sensors, the staying status in the facility, and the like.
- the status of various sensors refers to physiological values such as pedigree value, body temperature, heart rate, and respiratory rate.
- the human relationship parameter is estimated from the communication status (communication frequency, mail frequency, common community participation frequency). For example, it is estimated by applying a common sense rule such as “There is a friendship with the other party who frequently calls in a private time zone”.
- each parameter obtained by estimation is digitized by a predetermined correspondence relationship to generate a context feature vector (1602).
- a parameter digitization information table 1800 shown in FIG. 18A is used.
- the 30s are “(0, 0, 1, 0, 0, 0)”
- the man is “(1, 0)”
- the physiological phenomenon parameter is the walking distance 10 km is “10”
- human With the related parameters, the wife can be determined for each item such as “(1, 0, 0)”.
- the vector generation method is the same as that described above using the definition.
- FIG. 18B shows a parameter table 1801 of the context AZ.
- the values before the parameters are digitized are shown for easy understanding, but it goes without saying that the corresponding numerical values are actually stored.
- the generated context feature vector is compared with the similarity of each context in the context parameter table of the field context database 151, and the one with the highest similarity is selected and estimated as the current field context. (1603), the flow ends (1604).
- normalized correlation or the like is used for the similarity.
- each item importance is calculated using the weight (attribute importance) corresponding to the previously determined place context (1704).
- the weight (attribute importance) corresponding to the context of this place is stored in advance in the context / attribute table.
- FIG. 18C shows an example of an attribute weight table 1802 for storing context / attribute weights.
- the calculation of the importance of each item in step 1704 is as follows.
- the weight corresponding to the previously estimated field context corresponds to the weight of any row in FIG. 18c.
- w has two subscripts, the left side shows the attribute number, and the right side shows the number assigned to each value that the attribute can take. Therefore, for example, the subscript ⁇ n1 ⁇ indicates the total number of values that the first attribute can take. The same applies to ⁇ n2 ⁇ and thereafter.
- ⁇ _i a sum is taken so that i covers all the attribute numbers 1 to d.
- ⁇ _j ⁇ ⁇ ni ⁇ indicates that j is summed so as to extend over the number numbers 1 to ⁇ ni ⁇ that the i-th attribute can take.
- the present embodiment it is possible to make a recommendation from all the recommended items near the boundary by reflecting the priority weight according to the context of the field, and the background situation related to the user behavior is reflected in the recommended content. be able to.
- DESCRIPTION OF SYMBOLS 100 ... User 101 ... Terminal 102 ... Network connection apparatus 103 ... Network 104 ... Server 105 ... Item / attribute information 106 ... User profile 107 ... User action history 206 ... Internal bus 207 ... Arithmetic device 306 ... Internal bus 308 ... Memory 309 ... Interface 310 ... Storage device 311 ... Communication unit 700 ... Interest item 701 ... Indifferent item 702 ... Unrecommended item 703 ... Recent interest distance 704 ... Recently indifference distance 705 ... Interest / indifference distance 706 ... Boundary deviation degree.
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
Description
D(x,y)=√{(y1-x1)∧2+(y2-x2) ∧2+ . . . (yd-xd) ∧2}
ここで、√{ }は、{ }の中の値の平方根を、∧2は、2乗を示す。
(最近無関心距離)-(最近関心距離)>0
アイテムが一つとならない場合、乱数選択を行い、選択結果を推薦アイテムに決定する(605)。
Cor(x,y)=(y1・x1 + y2・x2 + . . . + yd・xd)/
[√{(x1)∧2+(x2) ∧2+ . . . (xd) ∧2}・√{(y1)∧2+(y2) ∧2+ . . . (yd) ∧2}]
となる。ここで、先と同様、√{ }は、{ }の中の値の平方根を、∧2は2乗、・は掛け算、/は割り算を示す。そして、ステップ1103において、例えば、類似度があらかじめ設定した閾値(例えば0.5)以上の部分のみリンクありとする。
(1)関心・無関心境界乖離度が小さいものを選択。
(2)全ての推薦済みアイテムとの隔たり数の平均値が最大のものを選択。
(3)全ての反応アイテムの隔たり数の平均値が最小のものを選択。
(4)乱数選択を行い選択。
そして、選択結果を推薦アイテムに決定し(1407)、終了する(1411)。
1番目の属性の重み … (w11,w12,…,w1{n1})
2番目の属性の重み … (w21,w22,…,w2{n2})
……
d番目の属性の重み … (wd1,wd2,…,wd{nd})
であるとする。ここで、wは2つの添え字を持ち、左側が属性の番号、右側が属性が取りうる値ごとにつけられた番号を示す。そのため例えば、添え字{n1}は、1番目の属性が取りうる値の総数を示す。{n2}以降も同様である。
あるアイテムxのd個の属性の値の番号が、(x1,x2,…,xd)の時の重要度は、
(アイテムの重要度)
=Σ_j∈{n1}δ(x1,j)w1j+Σ_j∈{n2}δ(x2,j)w2j+…
+Σ_j∈{nd}δ(xd,j)wdj
=Σ_i Σ_j∈{ni}δ(xi,j)wij
と算出される。
ここで、δ(x1,j)=“x1 = jの時1、そうでない時0”である。また、Σ_iでは、iが属性の番号1~d全てに渡るように和を取る。また、Σ_j∈{ni}は、jがi番目の属性がとりうる値の番号1~{ni}に渡るように和を取ることを示す。
101…端末
102…ネットワーク接続装置
103…ネットワーク
104…サーバ
105…アイテム・属性情報
106…ユーザプロファイル
107…ユーザ行動履歴
206…内部バス
207…演算装置
306…内部バス
308…メモリ
309…インターフェイス
310…記憶装置
311…通信部
700…関心アイテム
701…無関心アイテム
702…未推薦アイテム
703…最近関心距離
704…最近無関心距離
705…関心・無関心距離
706…境界乖離度。
Claims (15)
- 処理部と記憶部とを備えたサーバを用い、アイテムに関する情報を受け取ったユーザの反応を次に推薦するアイテムの選択基準に反映し、別途推薦を行う情報推薦方法であって、
前記処理部は、
ユーザが反応した情報と、無反応であったという情報を収集し、未推薦アイテムに対して、前記ユーザが反応した関心アイテムと無反応であった無関心アイテム双方と、どちらのアイテムとの類似性が高いかを比較し、その比較結果を用いて前記ユーザへの次の推薦アイテムを決定する、
ことを特徴とする情報推薦方法。 - 請求項1に記載の情報推薦方法であって、
前記処理部は、
前記未推薦アイテムについて、最近関心距離および最近無関心距離を算出し、前記最近関心距離および前記最近無関心距離の差から、前記未推薦アイテムの境界乖離度を算出し、前記境界乖離度が最小の前記未推薦アイテムを次の推薦アイテムに決定する、
ことを特徴とする情報推薦方法。 - 請求項1に記載の情報推薦方法であって、
前記処理部は、
前記関心アイテムと前記無関心アイテムそれぞれに類似した未推薦アイテムを、前記関心アイテムと前記無関心アイテムに拡張することにより、推薦対象である全てのアイテム中に、前記関心アイテムと前記無関心アイテムの境界である無関心・関心境界を推定する、
ことを特徴とする情報推薦方法。 - 請求項3に記載の情報推薦方法であって、
前記処理部は、
前記無関心・関心境界上あるいはその近傍の前記未推薦アイテム、または、前記無関心・関心境界で決められる前記関心アイテムを次の推薦アイテムに決定する、
ことを特徴とする情報推薦方法。 - 請求項1に記載の情報推薦方法であって、
前記処理部は、
前記ユーザの生理現象、および、同伴行動している人との人間関係である場のコンテキストを推定し、前記場のコンテキストを前記アイテムの推薦に反映させる、
ことを特徴とする情報推薦方法。 - ネットワークを介して端末に接続され、一次推薦アイテムに関する情報を受け取ったユーザの反応を、二次推薦アイテムの選択基準に反映し、別途推薦を行うサーバを用いた情報推薦システムであって、
前記サーバは処理部と記憶部とを有し、
前記処理部は、
前記一次推薦アイテムに対し、ユーザが反応した情報と、無反応であったという情報を収集して前記記憶部に蓄積し、未推薦アイテムに対して、前記ユーザが反応した関心アイテムと、無反応であった無関心アイテム双方とのどちらのアイテムとの類似性が高いかを比較し、その比較結果を用いて前記ユーザへの二次推薦アイテムを決定する、
ことを特徴とする情報推薦システム。 - 請求項6に記載の情報推薦システムであって、
前記処理部は、
前記未推薦アイテムについて、最近傍の前記関心アイテムとの距離である最近関心距離、および最近傍の前記無関心アイテムとの距離である最近無関心距離を算出し、前記最近関心距離および前記最近無関心距離の差から、前記未推薦アイテムの境界乖離度を算出し、前記境界乖離度が最小の前記未推薦アイテムを前記二次推薦アイテムに決定する、
ことを特徴とする情報推薦システム。 - 請求項6に記載の情報推薦システムであって、
前記処理部は、
前記関心アイテムと前記無関心アイテムそれぞれに類似した未推薦アイテムを、前記関心アイテムと前記無関心アイテムに拡張することにより、推薦対象である全てのアイテム中に、前記関心アイテムと前記無関心アイテムの境界である無関心・関心境界を決定する、
ことを特徴とする情報推薦システム。 - 請求項8に記載の情報推薦システムであって、
前記処理部は、
前記無関心・関心境界上、またはその近傍の前記未推薦アイテム、または、前記無関心・関心境界で決められる関心エリア側の前記未推薦アイテムを前記二次推薦アイテムに決定する、
ことを特徴とする情報推薦システム。 - 請求項6に記載の情報推薦システムであって、
前記処理部は、
前記ユーザの生理現象、および、同伴行動している人との人間関係である場のコンテキストを推定し、前記場のコンテキストを前記未推薦アイテムの推薦に反映させる、
ことを特徴とする情報推薦システム。 - 一次推薦アイテムに関する情報を受け取ったユーザの反応を、二次推薦アイテムの選択基準に反映して二次推薦を行うサーバであって、
前記サーバは処理部と記憶部とを有し、
前記処理部は、
ユーザが反応した関心アイテムと、無反応であった無関心アイテムとに関する情報を前記記憶部に蓄積し、未推薦アイテムに対して、前記ユーザが反応した関心アイテムと、無反応であった無関心アイテム双方とのどちらのアイテムとの類似性が高いかを比較し、その比較結果を用いて前記ユーザへの前記二次推薦アイテムを決定する、
ことを特徴とするサーバ。 - 請求項11に記載のサーバであって、
前記処理部は、
前記未推薦アイテムについて、最近傍の前記関心アイテムとの距離である最近関心距離、および最近傍の前記無関心アイテムとの距離である最近無関心距離を算出し、前記最近関心距離および前記最近無関心距離の差から、前記未推薦アイテムの境界乖離度を算出し、前記境界乖離度が最小の前記未推薦アイテムを前記二次推薦アイテムに決定する、
ことを特徴とするサーバ。 - 請求項11に記載のサーバであって、
前記処理部は、
前記関心アイテムと前記無関心アイテムそれぞれに類似した未推薦アイテムを、前記関心アイテムと前記無関心アイテムに拡張することにより、推薦対象である全てのアイテム中に、前記関心アイテムと前記無関心アイテムの境界である無関心・関心境界を推定する、
ことを特徴とするサーバ。 - 請求項13に記載のサーバであって、
前記処理部は、
前記無関心・関心境界上、あるいはその近傍の前記未推薦アイテム、または、前記無関心・関心境界で決められる前記関心エリアの前記未推薦アイテムを前記二次推薦アイテムに決定する、
ことを特徴とするサーバ。 - 請求項11に記載のサーバであって、
前記記憶部は、
前記ユーザの生理現象パラメータ、および、同伴行動している人との人間関係パラメータを数値化するためのパラメータ数値化テーブルを記憶し、
前記処理部は、
前記パラメータ数値化テーブルを用いて、前記ユーザの生理現象、および同伴行動している人との人間関係である場のコンテキストを示すコンテキスト特徴ベクトルを算出し、前記コンテキスト特徴ベクトルに基づき、前記場のコンテキストを決定し、
前記場のコンテキストを前記二次推薦アイテムの選定に反映させる、
ことを特徴とするサーバ。
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2011500357A JP5277307B2 (ja) | 2009-02-18 | 2009-02-18 | 情報推薦方法、そのシステム、及びサーバ |
PCT/JP2009/000666 WO2010095169A1 (ja) | 2009-02-18 | 2009-02-18 | 情報推薦方法、そのシステム、及びサーバ |
US13/148,966 US8635241B2 (en) | 2009-02-18 | 2009-02-18 | Method of recommending information, system thereof, and server |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2009/000666 WO2010095169A1 (ja) | 2009-02-18 | 2009-02-18 | 情報推薦方法、そのシステム、及びサーバ |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2010095169A1 true WO2010095169A1 (ja) | 2010-08-26 |
Family
ID=42633475
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/000666 WO2010095169A1 (ja) | 2009-02-18 | 2009-02-18 | 情報推薦方法、そのシステム、及びサーバ |
Country Status (3)
Country | Link |
---|---|
US (1) | US8635241B2 (ja) |
JP (1) | JP5277307B2 (ja) |
WO (1) | WO2010095169A1 (ja) |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120059735A1 (en) * | 2010-09-03 | 2012-03-08 | Alibaba Group Holding Limited | Product recommendations |
JP2015099593A (ja) * | 2013-11-18 | 2015-05-28 | パロ・アルト・リサーチ・センター・インコーポレーテッドPalo Alto Research Center Incorporated | 推薦エンジンに基づく汎用グラフ、ルール及び空間構造 |
JP2016143161A (ja) * | 2015-01-30 | 2016-08-08 | 富士通株式会社 | アイテム推薦プログラム、装置、及び方法 |
JP2017509947A (ja) * | 2014-01-27 | 2017-04-06 | アリババ・グループ・ホールディング・リミテッドAlibaba Group Holding Limited | ネットワークサブジェクトの社会的関係タイプの取得 |
JP2019109739A (ja) * | 2017-12-19 | 2019-07-04 | 富士ゼロックス株式会社 | 情報処理装置及びプログラム |
JP2020149120A (ja) * | 2019-03-11 | 2020-09-17 | 株式会社エヌ・ティ・ティ・データ | 情報処理装置、情報処理方法およびプログラム |
CN112035740A (zh) * | 2020-08-19 | 2020-12-04 | 广州市百果园信息技术有限公司 | 项目使用时长预测方法、装置、设备及存储介质 |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
TWI525460B (zh) * | 2013-10-07 | 2016-03-11 | 財團法人資訊工業策進會 | 電子計算裝置、其個人化資訊推薦方法及其電腦程式產品 |
KR101907008B1 (ko) * | 2017-06-15 | 2018-10-15 | 한양대학교 산학협력단 | 단일 클래스 협업 필터링 환경에서 무관심 아이템 결정 방법 및 장치 |
CN108804492B (zh) * | 2018-03-27 | 2022-04-29 | 阿里巴巴(中国)有限公司 | 用于多媒体对象推荐的方法及装置 |
KR102200340B1 (ko) * | 2018-05-18 | 2021-01-08 | 한양대학교 산학협력단 | 표준화된 평점 기반의 아이템 추천 방법 및 장치 |
EP3598373A1 (en) * | 2018-07-18 | 2020-01-22 | Seulo Palvelut Oy | Determining product relevancy |
US20210065276A1 (en) * | 2019-08-28 | 2021-03-04 | Fuji Xerox Co., Ltd. | Information processing apparatus and non-transitory computer readable medium |
US12111870B2 (en) * | 2021-03-26 | 2024-10-08 | EMC IP Holding Company LLC | Automatic discovery of related data records |
CN113312563B (zh) * | 2021-06-24 | 2022-09-16 | 北京三快在线科技有限公司 | 信息推荐方法、装置、设备及存储介质 |
CN115033798A (zh) * | 2022-07-04 | 2022-09-09 | 贵州电网有限责任公司 | 基于大数据的活动推荐方法及系统 |
CN115618101A (zh) * | 2022-09-21 | 2023-01-17 | 清华大学 | 基于负反馈的流媒体内容推荐方法、装置及电子设备 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001142889A (ja) * | 1999-11-11 | 2001-05-25 | Matsushita Electric Ind Co Ltd | 情報仲介装置および情報仲介方法 |
JP2001256253A (ja) * | 2000-03-13 | 2001-09-21 | Kddi Corp | 文書フィルタリング方法および装置 |
JP2006146630A (ja) * | 2004-11-22 | 2006-06-08 | Sony Corp | コンテンツ選択再生装置、コンテンツ選択再生方法、コンテンツ配信システムおよびコンテンツ検索システム |
JP2006235716A (ja) * | 2005-02-22 | 2006-09-07 | Hitachi Ltd | 文書フィルタリングシステム |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2008204235A (ja) * | 2007-02-21 | 2008-09-04 | Hitachi Ltd | 非類似アイテム推薦方法、推薦装置、及びプログラム |
JP4947709B2 (ja) | 2007-02-23 | 2012-06-06 | Kddi株式会社 | コンテンツ配信システム |
-
2009
- 2009-02-18 JP JP2011500357A patent/JP5277307B2/ja active Active
- 2009-02-18 US US13/148,966 patent/US8635241B2/en active Active
- 2009-02-18 WO PCT/JP2009/000666 patent/WO2010095169A1/ja active Application Filing
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2001142889A (ja) * | 1999-11-11 | 2001-05-25 | Matsushita Electric Ind Co Ltd | 情報仲介装置および情報仲介方法 |
JP2001256253A (ja) * | 2000-03-13 | 2001-09-21 | Kddi Corp | 文書フィルタリング方法および装置 |
JP2006146630A (ja) * | 2004-11-22 | 2006-06-08 | Sony Corp | コンテンツ選択再生装置、コンテンツ選択再生方法、コンテンツ配信システムおよびコンテンツ検索システム |
JP2006235716A (ja) * | 2005-02-22 | 2006-09-07 | Hitachi Ltd | 文書フィルタリングシステム |
Non-Patent Citations (1)
Title |
---|
FUMIHIKO YAMAMOTO ET AL.: "WWW Page Kensaku ni Okeru Riyosha no Kyomi o Jido Chushutsu suru Hoho no Kento", DAI 8 KAI GODO KENKYUKAI, "AI SYMPOSIUM '97", SHIRYO (SIG-J-9701), 5 December 1997 (1997-12-05), pages 43 - 48 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120059735A1 (en) * | 2010-09-03 | 2012-03-08 | Alibaba Group Holding Limited | Product recommendations |
JP2015099593A (ja) * | 2013-11-18 | 2015-05-28 | パロ・アルト・リサーチ・センター・インコーポレーテッドPalo Alto Research Center Incorporated | 推薦エンジンに基づく汎用グラフ、ルール及び空間構造 |
JP2017509947A (ja) * | 2014-01-27 | 2017-04-06 | アリババ・グループ・ホールディング・リミテッドAlibaba Group Holding Limited | ネットワークサブジェクトの社会的関係タイプの取得 |
JP2016143161A (ja) * | 2015-01-30 | 2016-08-08 | 富士通株式会社 | アイテム推薦プログラム、装置、及び方法 |
JP2019109739A (ja) * | 2017-12-19 | 2019-07-04 | 富士ゼロックス株式会社 | 情報処理装置及びプログラム |
JP2020149120A (ja) * | 2019-03-11 | 2020-09-17 | 株式会社エヌ・ティ・ティ・データ | 情報処理装置、情報処理方法およびプログラム |
JP7253410B2 (ja) | 2019-03-11 | 2023-04-06 | 株式会社エヌ・ティ・ティ・データ | 情報処理装置、情報処理方法およびプログラム |
CN112035740A (zh) * | 2020-08-19 | 2020-12-04 | 广州市百果园信息技术有限公司 | 项目使用时长预测方法、装置、设备及存储介质 |
CN112035740B (zh) * | 2020-08-19 | 2024-06-04 | 广州市百果园信息技术有限公司 | 项目使用时长预测方法、装置、设备及存储介质 |
Also Published As
Publication number | Publication date |
---|---|
JP5277307B2 (ja) | 2013-08-28 |
US8635241B2 (en) | 2014-01-21 |
US20110314040A1 (en) | 2011-12-22 |
JPWO2010095169A1 (ja) | 2012-08-16 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP5277307B2 (ja) | 情報推薦方法、そのシステム、及びサーバ | |
US8572017B2 (en) | User-based collaborative filtering recommendation system and method for amending similarity using information entropy | |
US7707068B2 (en) | Method and device for calculating trust values on purchases | |
US20130185294A1 (en) | Recommender system, recommendation method, and program | |
Gogna et al. | DiABlO: Optimization based design for improving diversity in recommender system | |
Xu et al. | Integrated collaborative filtering recommendation in social cyber-physical systems | |
Zhang et al. | Prediction uncertainty in collaborative filtering: Enhancing personalized online product ranking | |
CN102789462A (zh) | 一种项目推荐方法及系统 | |
CN113327151B (zh) | 商品对象推荐方法、装置、计算机设备及存储介质 | |
US20140046804A1 (en) | Customizing online automotive vehicle searches | |
CN109165847A (zh) | 一种基于推荐系统的项目推荐方法、装置及设备 | |
US20180232794A1 (en) | Method for collaboratively filtering information to predict preference given to item by user of the item and computing device using the same | |
KR20170116924A (ko) | 상대 비교를 통한 상품 추천 서비스 방법 및 이를 위한 추천 서비스 장치와 프로그램 | |
US20130097053A1 (en) | Method and system for recommending a combined service by taking into account situation information on a target user and the degree of complementarity of a service | |
Mikeli et al. | A multi-criteria recommendation method for interval scaled ratings | |
KR20140132033A (ko) | 상품 추천 서비스 시스템 및 그 방법, 그리고 이에 적용되는 장치 | |
WO2016046873A1 (ja) | 情報処理装置、情報処理方法及びプログラム | |
Mao et al. | Utilizing multi-source data in popularity prediction for shop-type recommendation | |
Özsoy et al. | Multi-objective optimization based location and social network aware recommendation | |
CN111178951B (zh) | 一种商品推荐方法及装置 | |
CN111046285B (zh) | 一种推荐排序确定方法、装置、服务器及存储介质 | |
US8626772B2 (en) | Double weighted correlation scheme | |
CN114065016A (zh) | 一种推荐方法、装置、设备及计算机可读存储介质 | |
JP2013029896A (ja) | アイテム推薦装置及び方法及びプログラム | |
WO2010084629A1 (ja) | 推薦システム、推薦方法、推薦プログラム及び情報記憶媒体 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09840279 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2011500357 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13148966 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 09840279 Country of ref document: EP Kind code of ref document: A1 |