WO2011074113A1 - 協調フィルタリングシステム及び協調フィルタリング方法 - Google Patents
協調フィルタリングシステム及び協調フィルタリング方法 Download PDFInfo
- Publication number
- WO2011074113A1 WO2011074113A1 PCT/JP2009/071141 JP2009071141W WO2011074113A1 WO 2011074113 A1 WO2011074113 A1 WO 2011074113A1 JP 2009071141 W JP2009071141 W JP 2009071141W WO 2011074113 A1 WO2011074113 A1 WO 2011074113A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- content
- evaluation value
- user
- specific content
- similarity
- 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
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/33—Querying
- G06F16/335—Filtering based on additional data, e.g. user or group profiles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/40—Information retrieval; Database structures therefor; File system structures therefor of multimedia data, e.g. slideshows comprising image and additional audio data
- G06F16/43—Querying
- G06F16/435—Filtering based on additional data, e.g. user or group profiles
Definitions
- the present invention relates to a collaborative filtering system and a collaborative filtering method, and more particularly to a collaborative filtering system and a collaborative filtering method that use evaluation values for contents of a plurality of users.
- Patent Document 1 There has been proposed a collaborative filtering technique that accumulates information about preferences for many users and predicts the preferences of other users who have similar preferences to a certain user. Collaborative filtering is used for recommendations and personalization. For example, in Patent Document 1, when an arbitrary user has actually voted for an arbitrary item, the evaluation value of this item is assigned to the corresponding cell of the item-user matrix, and other similar items to this item An information recommendation method is disclosed in which an evaluation value is substituted in a pseudo manner for an item cell. The information recommendation method of Patent Document 1 also makes it possible to recommend items that could not be recommended because there are no evaluation values.
- a car navigation system may recommend content such as a store or a facility using collaborative filtering as described above in consideration of user preferences. For example, if a car navigation system uses a history of restaurants visited by a user in the past and a history of visits of restaurants of other users who visited the restaurant, the user who is likely to meet the user's preferences still visits the user. There may be a service to recommend a restaurant that is not. In such a case, the car navigation system may recommend only content around the user's action range.
- the collaborative filter may not work. That is, as a method of assigning an evaluation value to content, a method in which a user writes an evaluation score via a website, or a method in which the user inputs an evaluation score via a car navigation system after a visit to a store or facility that is the content Etc. are considered. However, since both methods take time and effort for the user, it is expected that user evaluation values are difficult to gather.
- the present invention has been made in view of such circumstances, and the object thereof is a collaborative filtering system and collaborative filtering method capable of further expanding the range of contents for which evaluation values can be predicted by collaborative filtering. Is to provide. It is another object of the present invention to provide a collaborative filtering system and a collaborative filtering method capable of performing prediction by collaborative filtering with a smaller number of user evaluation values.
- the present invention is a collaborative filtering system that uses evaluation values for content of each of a plurality of users, and there is no evaluation value of at least one user who has evaluated both the first content and the second content. Sometimes there is an evaluation value of at least one user who has evaluated both the first content and the third content, and the evaluation is performed on both the second content and the third content.
- a predictive evaluation value calculating unit that calculates a predicted evaluation value of the user not evaluated for any of the content and the second content is a collaborative filtering system with a.
- the similarity calculation unit has an evaluation value of at least one user who has evaluated both the first and third contents, and evaluates both the second and third contents.
- the degree of similarity between the first and second contents is calculated using the evaluation value of the third content in which there is an evaluation value of at least one user who has performed. That is, the similarity calculation unit indirectly calculates the similarity between the first and second contents using the third content that can directly calculate the similarity with each of the first and second contents. . Thereby, the similarity of the 1st and 2nd content which cannot be calculated directly can be calculated.
- the predicted evaluation value calculation unit uses the first content and the second content similarity calculated by the similarity calculation unit, and the evaluation values for the first content and the second content, to calculate the first content.
- a predicted evaluation value of a user who has not evaluated any of the content and the second content is calculated.
- the predicted evaluation values of the first and second contents that cannot be directly calculated can be calculated. Therefore, it is possible to further expand the range of content for which the evaluation value can be predicted by collaborative filtering.
- the similarity calculation unit includes the first content that is the Pearson product-moment correlation coefficient for the i-th third content in which the number of the third content is N and 1 ⁇ i ⁇ N.
- the similarity of the third content is represented by s (C1, C3i)
- the similarity of the second content and the third content, which is Pearson's product-moment correlation coefficient is represented by s (C2, C3i).
- the first and third contents quantitatively based on the similarity between the first and third contents expressed by the Pearson product moment correlation coefficient and the similarity between the second and third contents. It becomes possible to calculate the similarity of the second content.
- the present invention is a collaborative filtering system that uses evaluation values for the contents of each of a plurality of users, and for one user who has used specific content and did not evaluate specific content, The usage frequency of specific content of one user, the usage frequency of specific content of another user who has used the specific content and evaluated the specific content, and the evaluation value for the specific content It is a collaborative filtering system provided with the prediction evaluation value calculation unit which calculates the prediction evaluation value with respect to the specific content of one user using.
- the predicted evaluation value calculation unit uses one user's specific content for one user who has used the specific content and has not evaluated the specific content.
- One user is identified using the frequency, the use frequency of the specific content of another user who has used the specific content and has evaluated the specific content, and the evaluation value for the specific content. It is preferable to calculate a predicted evaluation value for the content of the content.
- the predicted evaluation value calculation unit uses the specific content of the one user and the specific content of another user who has used the specific content and has evaluated the specific content. Use frequency and an evaluation value for a specific content. It is considered that the usage frequency of content by each user correlates with the evaluation value for the content. Therefore, the predicted evaluation value calculation unit has used the specific content of the one user, the specific content has been used, and the specific content without the evaluation value of the one user himself / herself. It is possible to calculate a predicted evaluation value for a specific content of one user by using the usage frequency and evaluation value of the specific content of another user who has performed the evaluation. Therefore, prediction by collaborative filtering can be performed with a smaller number of user evaluation values.
- the predictive evaluation value calculation unit is configured so that the usage frequency of the specific content of one user is different from the usage frequency of the specific content of the one user within a predetermined threshold. It is preferable to calculate a predicted evaluation value for a specific content of one user by using a use frequency of the specific content of the user and an evaluation value for the specific content.
- the predicted evaluation value calculation unit is different from the usage frequency of the specific content of one user and the usage frequency of the specific content within a predetermined threshold from the usage frequency of the specific content of the one user.
- a predicted evaluation value can be calculated with high accuracy by calculating a predicted evaluation value for a specific content of one user using a use frequency of the specific content of the user and an evaluation value for the specific content.
- the predicted evaluation value calculation unit is configured to use a specific user's specific content usage frequency and an unspecified many content usage frequency within a category to which the specific content belongs. It is preferable to calculate a predicted evaluation value for the specific content of one user using the use frequency of the specific content of the other user and the evaluation value for the specific content that are differences within a predetermined threshold from .
- the predictive evaluation value calculation unit uses the frequency of usage of specific content of one user and the frequency of usage of unspecified number of contents of one user whose usage frequency of unspecified number of contents in the category to which the specific content belongs.
- the predicted evaluation value calculation unit may use any one of a median value and an average value of evaluation values for specific content for each specific content usage frequency of each of a plurality of other users. It is preferable to calculate a predicted evaluation value for a specific content of one user using
- the predicted evaluation value calculation unit can calculate either the median value or the average value of the evaluation values for the specific content for each usage frequency of the specific content of each of the plurality of other users. By using this to calculate the predicted evaluation value for the specific content of one user, the predicted evaluation value can be calculated with higher accuracy.
- the predicted evaluation value calculation unit is based on the use frequency of the specific content of other users who have used the specific content and have evaluated the specific content and the evaluation value for the specific content. Deriving a function of an evaluation value for a specific content with respect to the usage frequency of the specific content, and calculating a predicted evaluation value for the specific content of the one user by using the usage frequency and the function of the specific content of the one user It is preferable to do.
- Frequency of use of specific content derived based on use frequency of specific content of other users who have used specific content and who have evaluated the specific content, and an evaluation value for the specific content It can be estimated that the function of the evaluation value for the specific content with respect to represents the correlation between the usage frequency and the evaluation value with high accuracy. Furthermore, even if the obtained usage frequency and evaluation value are discrete values, interpolation can be performed by the function. Therefore, the predicted evaluation value calculation unit is based on the use frequency of the specific content of another user who has used the specific content and evaluated the specific content and the evaluation value for the specific content.
- the present invention is a collaborative filtering method using evaluation values for the contents of each of a plurality of users, and the evaluation value of at least one user who has evaluated both the first content and the second content is When it does not exist, there is an evaluation value of at least one user who has evaluated both the first content and the third content, and evaluation is performed on both the second content and the third content.
- a collaborative filtering method comprising the predicted evaluation value calculation step of calculating a predicted evaluation value of the user not evaluated for any of the Ceiling and the second content.
- the number of third contents is N
- the similarity of the third content is represented by s (C1, C3i)
- the similarity of the second content and the third content, which is Pearson's product-moment correlation coefficient is represented by s (C2, C3i).
- the present invention is a collaborative filtering method using evaluation values for the contents of each of a plurality of users, for one user who has used specific content and has not performed evaluation on the specific content.
- the usage frequency of specific content of one user, the usage frequency of specific content of another user who has used the specific content and evaluated the specific content, and the evaluation value for the specific content Is a collaborative filtering method including a predicted evaluation value calculation step of calculating a predicted evaluation value for a specific content of one user.
- the predicted evaluation value calculation step uses one user's specific content for one user who has used specific content and has not evaluated the specific content.
- One user is identified using the frequency, the use frequency of the specific content of another user who has used the specific content and has evaluated the specific content, and the evaluation value for the specific content. It is preferable to calculate a predicted evaluation value for the content of the content.
- the predicted evaluation value calculation step includes a usage frequency of one user's specific content and a usage frequency of the specific content that is a difference within a predetermined threshold from the usage frequency of the specific content of the one user. It is preferable to calculate a predicted evaluation value for a specific content of one user by using a use frequency of the specific content of the user and an evaluation value for the specific content.
- the predicted evaluation value calculation step includes the usage frequency of the specific content of one user and the usage frequency of the unspecified number of contents of the user whose usage frequency of the unspecified number of contents in the category to which the specific content belongs. It is preferable to calculate a predicted evaluation value for the specific content of one user using the use frequency of the specific content of the other user and the evaluation value for the specific content that are differences within a predetermined threshold from .
- the predicted evaluation value calculation step includes any one of a median value and an average value of evaluation values for specific content for each specific content usage frequency of each of a plurality of other users. It is preferable to calculate a predicted evaluation value for a specific content of one user using
- the predicted evaluation value calculation step is based on the use frequency of the specific content of another user who has used the specific content and has evaluated the specific content and the evaluation value for the specific content. Deriving a function of an evaluation value for a specific content with respect to the usage frequency of the specific content, and calculating a predicted evaluation value for the specific content of the one user by using the usage frequency and the function of the specific content of the one user It is preferable to do.
- collaborative filtering system and collaborative filtering method of the present invention it is possible to further expand the range of contents for which evaluation values can be predicted by collaborative filtering. Moreover, according to the collaborative filtering system and collaborative filtering method of the present invention, prediction by collaborative filtering can be performed with a smaller number of user evaluation values.
- the collaborative filtering system includes an information processing center 10a that provides information to each of the vehicle navigation devices 40 mounted on a plurality of vehicles.
- the information processing center 10a can be connected to the Internet 50 including a gourmet search site 51 that provides information about restaurants and facilities that are preferred by the user.
- the information processing center 10a predicts each user's preference by collaborative filtering, and recommends content such as restaurants, hotels, and retail stores that the user has not visited to the user via the vehicle navigation device 40. It is a facility for.
- the information processing center 10a includes a user evaluation value storage device 21a, a position information database 22, and an arithmetic processing device 30.
- the user evaluation value storage device 21a is a database that stores and manages evaluation values for the contents of a plurality of users for each content.
- storage device 21a is acquired by transmitting the information which the user input into the vehicle navigation apparatus 40 from the vehicle navigation apparatus 40 to the information processing center 10a.
- each user's evaluation value stored in the user evaluation value storage device 21 a is acquired by collecting information about registered users from various sites on the Internet 50.
- the location information database 22 is a database that stores and manages information related to contents such as restaurants, hotels, and retail stores that a user driving the vehicle may visit in association with the POI (Position Information Position Of Information) of the content. .
- POI Position Information Position Of Information
- the arithmetic processing device 30 is a device that predicts an evaluation value for an unvisited content of a target user by collaborative filtering based on information stored in the user evaluation value storage device 21a and the position information database 22.
- the arithmetic processing device 30 includes a direct similarity calculation unit 31, an indirect similarity calculation unit 32, and a predicted evaluation value calculation unit 33.
- the direct similarity calculation unit 31 is a part that calculates a similarity between two contents based on a user's evaluation value for each of the two contents using a Pearson product-moment correlation coefficient or the like.
- the indirect similarity calculation unit 32 indirectly uses another content that can directly calculate the similarity with each content. This is a part for calculating the similarity between two contents.
- the predicted evaluation value calculation unit 33 determines how the target user operates with respect to unvisited content. It is a part which predicts whether to give a proper evaluation value.
- the user evaluation value storage device 21 a of the information processing center 10 a is based on information transmitted from the vehicle navigation device 40 to the information processing center 10 a and information from a website on the Internet 50.
- An evaluation value for each of the contents is acquired (S11).
- the user's evaluation value can be acquired by acquiring a log of a Web site that evaluates a restaurant or the like on the Internet 50.
- the evaluation of the user is performed again.
- a value is acquired (S11).
- the direct calculation of the arithmetic processing device 30 is performed.
- the unit 31 calculates the similarity between the contents based on the evaluation values of the respective users for the collected contents (S13).
- the similarity between the two contents can be directly calculated by the Pearson product-moment correlation coefficient.
- the evaluation values for the contents X and Y of the i-th user are rX (i) and rY (i), respectively.
- the direct similarity calculation unit 31 calculates the similarity s (X, Y) of the contents X and Y from the following equation (3).
- the similarity calculated by the direct similarity calculation unit 31 is stored in the storage area of the predicted evaluation value calculation unit 33 (S14).
- the indirect similarity calculating unit 32 can determine the degree of similarity that can be indirectly calculated among the degrees of similarity that could not be calculated directly. Is calculated (S16). As shown in FIG. 3, for the content X and the content Z, since there is no evaluation value of the user who has evaluated both the content X and Z, it is impossible to directly calculate the similarity.
- the indirect similarity calculation unit 32 indirectly calculates the similarity between the contents X and Z by using the evaluation value of the content Y.
- the number of the contents Y is not limited to a single number, and by using a plurality of contents Y, the accuracy of calculating the similarity is improved.
- the indirect similarity calculation unit 32 calculates the similarity s (X, Z) of the contents X and Z from the following equation (4).
- the predicted evaluation value calculation unit 33 creates an existing value based on the similarity and the evaluation value calculated in steps S11 to S16.
- the evaluation value for the content that the user has not evaluated is predicted by the collaborative filtering method (S17). For example, as illustrated in FIG. 5, it is possible to predict the evaluation value of the user A who has not evaluated the content Z.
- the indirect similarity calculation unit 32 of the arithmetic processing unit 30 of the information processing center 10a has evaluation values of users who have evaluated both the contents X and Y, and The degree of similarity between the contents X and Z is calculated using the evaluation value of the content Y in which the evaluation value of the user who has evaluated both the contents Y and Z exists.
- the indirect similarity calculation unit 32 indirectly calculates the similarity between the contents X and Z by using the content Y that can directly calculate the similarity with each of the contents X and Z. As a result, the similarity between the contents X and Z that cannot be directly calculated can be calculated.
- the predicted evaluation value calculation unit 33 evaluates either the content X or Z using the similarity of the indirect similarity calculation unit 32 contents X and Z and the evaluation value for the contents X and Z.
- the predicted evaluation value of the user who did not exist is calculated.
- the predicted evaluation values of the contents X and Z that cannot be directly calculated can be calculated. Therefore, it is possible to further expand the range of content for which the evaluation value can be predicted by collaborative filtering. Furthermore, the range of contents that can be recommended is expanded.
- the similarity between the contents X and Z is quantitatively determined based on the similarity between the contents X and Y expressed by the Pearson product moment correlation coefficient and the similarity between the contents Y and Z. It is possible to calculate.
- the information processing center 10b of the present embodiment includes a user evaluation value storage device 21b. Similar to the first embodiment, the user evaluation value storage device 21b stores and manages the visit history (date and time, location) of the content for each user in addition to the evaluation value for the content for each user. Information related to the visit history is transmitted to the information processing center 10b from the vehicle navigation device 40 or the mobile terminal of each user.
- the arithmetic processing device 30 of the information processing center 10b of this embodiment includes a non-input evaluation value prediction unit 34, a similarity calculation unit 35, and a predicted evaluation value calculation unit 36.
- the uninput evaluation value prediction unit 34 is a part that predicts an evaluation value for content for which the user has not input an evaluation value and has a visit history.
- the similarity calculation unit 35 is a part that directly or indirectly obtains the similarity between contents.
- the prediction evaluation value calculation unit 36 is based on the similarity between the two contents calculated by the direct similarity calculation unit 31 and the indirect similarity calculation unit 32 as in the prediction evaluation value calculation unit 33 of the first embodiment. Thus, it is a part that predicts what evaluation value the target user will give to unvisited content.
- the user evaluation value storage device 21b of the information processing center 10b collects the visit history and evaluation value of the target user's content in the same manner as in the first embodiment (S21).
- the uninput evaluation value prediction unit 34 specifies the correlation of the evaluation value with respect to the content visit frequency for each category visit frequency (S25). If the visit frequency of the user A to the category “restaurant” to which the restaurant X belongs is twice a week, the uninput evaluation value predicting unit 34 uses the “restaurant” twice a week as shown in FIG. Statistics are taken on the usage frequency and evaluation value of the restaurant X of each of the other users. In this case, the uninput evaluation value prediction unit 34 performs statistical processing for each use frequency of the restaurant X, and specifies the correlation of the evaluation value with respect to the content visit frequency by taking the median value or the average value of the evaluation values. be able to. Further, the non-input evaluation value prediction unit 34 can specify the correlation between the evaluation value and the visit frequency of the content by obtaining an approximate expression of the function of the evaluation value with respect to the usage frequency of the restaurant X.
- the evaluation value of the user who visits the category “restaurant” twice a week as the user A and visits the restaurant X the same as the user A once a month is “2.5”. It is. Therefore, the uninput evaluation value prediction unit 34 can predict the evaluation value for the restaurant X of the user A as 2.5.
- the category visit frequency and the content visit frequency do not have to coincide completely, and can be considered equal if the difference is within a predetermined threshold range.
- the uninput evaluation value prediction unit 34 substitutes the prediction evaluation value predicted in S26 in the storage area (S27).
- the uninput evaluation value prediction unit 34 performs normal collaborative filtering processing based on the prediction evaluation value (S28).
- step S22 when there is a user visit history and there is no content for which an evaluation value is not input, the uninput evaluation value prediction unit 34 is stored in the storage area in the user evaluation value storage device 21b. An evaluation value is substituted (S29).
- the uninput evaluation value prediction unit 34 of the arithmetic processing unit 30 of the information processing center 10b has used the restaurant X of the user A, the restaurant X, and the restaurant X.
- the usage frequency of the restaurant X of another user who has evaluated X and the evaluation value for the restaurant X are used. It is considered that the usage frequency of the restaurant X by each user correlates with the evaluation value for the restaurant X.
- the uninput evaluation value prediction unit 34 has used the restaurant X of the user A, has used the restaurant X, and has evaluated the restaurant X, even without the evaluation value of the user A himself. It is possible to calculate a predicted evaluation value for user A's restaurant X using the user's usage frequency and evaluation value of restaurant X. Therefore, prediction by collaborative filtering can be performed with a smaller number of user evaluation values.
- the uninput evaluation value prediction unit 34 uses the restaurant X of the user A and the restaurant X of another user whose use frequency of the restaurant X is a difference within a predetermined threshold from the use frequency of the restaurant of the user A.
- the predicted evaluation value can be calculated with high accuracy by calculating the predicted evaluation value for the restaurant X of the user A using the usage frequency and the evaluation value for the restaurant X.
- the uninput evaluation value predicting unit 34 uses the usage frequency of the user A's restaurant X and the usage frequency of the unspecified number of contents in the category “restaurant” to which the restaurant X belongs to the user A's unspecified number of contents.
- the prediction evaluation value for the restaurant X of the user A is calculated, thereby further accurately predicting. An evaluation value can be calculated.
- the uninput evaluation value prediction unit 34 calculates either the median value or the average value of the evaluation values for the restaurant X for each user X's restaurant X usage frequency and the usage frequency of the restaurant X of each of the other users. By using and calculating the prediction evaluation value for the user A's restaurant X, the prediction evaluation value can be calculated with higher accuracy.
- the restaurant X with respect to the usage frequency of the restaurant X derived based on the usage frequency of the restaurant X of other users who have used the restaurant X and evaluated the restaurant X and the evaluation value for the restaurant X It can be estimated that the function of the evaluation value with respect to accurately represents the correlation between the usage frequency and the evaluation value. Furthermore, even if the obtained usage frequency and evaluation value are discrete values, interpolation can be performed by the function. Therefore, the uninput evaluation value prediction unit 34 uses the restaurant X based on the usage frequency of the restaurant X of other users who have used the restaurant X and evaluated the restaurant X and the evaluation value for the restaurant X.
- a function of the evaluation value for the restaurant X with respect to the usage frequency of X is derived, and the predicted evaluation value for the restaurant X of the user A is calculated using the usage frequency and function of the specific content of the user A, thereby further improving the accuracy.
- a predicted evaluation value can be calculated. Even if the usage frequency of other users is different from the usage frequency of user A, a predicted evaluation value can be calculated by interpolation using a function.
- the present invention is not limited to the above-described embodiments, and various modifications can be made.
- the aspect in which the content is a store, a facility, or the like related to the position and the recommendation is made to the user who has boarded the vehicle via the vehicle navigation device 40 has been mainly described.
- an aspect of predicting an evaluation value for other content for which it is difficult to obtain an evaluation value is also included in the scope of the present invention and exhibits its effect.
- the present invention can provide a collaborative filtering system and a collaborative filtering method that can further expand the range of contents for which evaluation values can be predicted by collaborative filtering. Moreover, this invention can provide the collaborative filtering system and the collaborative filtering method which can perform prediction by collaborative filtering with the evaluation value of a smaller number of users.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computational Linguistics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
21a,21b ユーザ評価値記憶装置
22 位置情報データベース
30 演算処理装置
31 直接類似度算出部
32 間接類似度算出部
33 予測評価値算出部
34 未入力評価値予測部
35 類似度算出部
36 予測評価値算出部
40 車両用ナビゲーション装置
50 インターネット
51 グルメ検索サイト
Claims (16)
- 複数のユーザそれぞれのコンテンツに対する評価値を用いる協調フィルタリングシステムであって、
第1のコンテンツ及び第2のコンテンツの両方に対して評価を行った少なくとも一のユーザの評価値が存在しないときに、前記第1のコンテンツ及び第3のコンテンツの両方に対して評価を行った少なくとも一のユーザの評価値が存在し、且つ前記第2のコンテンツ及び前記第3のコンテンツの両方に対して評価を行った少なくとも一のユーザの評価値が存在する前記第3のコンテンツの評価値を用いて、前記第1のコンテンツ及び前記第2のコンテンツの類似度を算出する類似度算出ユニットと、
前記類似度算出ユニットが算出した前記第1のコンテンツ及び前記第2のコンテンツの類似度と、前記第1のコンテンツ及び前記第2のコンテンツに対する評価値とを用いて、前記第1のコンテンツ及び前記第2のコンテンツのいずれかに対して評価を行わなかったユーザの予測評価値を算出する予測評価値算出ユニットと、を備えた協調フィルタリングシステム。 - 複数のユーザそれぞれのコンテンツに対する評価値を用いる協調フィルタリングシステムであって、
特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行わなかった一のユーザについて、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行った他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する予測評価値算出ユニットを備えた協調フィルタリングシステム。 - 前記予測評価値算出ユニットは、特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行わなかった一のユーザについて、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行った他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項1又は2に記載の協調フィルタリングシステム。
- 前記予測評価値算出ユニットは、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツの利用頻度が前記一のユーザの前記特定のコンテンツの利用頻度から所定の閾値以内の差異である他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項3又は4に記載の協調フィルタリングシステム。
- 前記予測評価値算出ユニットは、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツが属するカテゴリー内の不特定多数のコンテンツの利用頻度が前記一のユーザの前記不特定多数のコンテンツの利用頻度から所定の閾値内の差異である他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項3~5のいずれか1項に記載の協調フィルタリングシステム。
- 前記予測評価値算出ユニットは、前記一のユーザの前記特定のコンテンツの利用頻度と、複数の前記他のユーザそれぞれの前記特定のコンテンツの利用頻度ごとの前記特定のコンテンツに対する評価値の中央値及び平均値のいずれかを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項3~6のいずれか1項に記載の協調フィルタリングシステム。
- 前記予測評価値算出ユニットは、特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行った他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とに基づいて、前記特定のコンテンツの利用頻度に対する前記特定のコンテンツに対する評価値の関数を導出し、前記一のユーザの前記特定のコンテンツの利用頻度と前記関数とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項3~7のいずれか1項に記載の協調フィルタリングシステム。
- 複数のユーザそれぞれのコンテンツに対する評価値を用いる協調フィルタリング方法であって、
第1のコンテンツ及び第2のコンテンツの両方に対して評価を行った少なくとも一のユーザの評価値が存在しないときに、前記第1のコンテンツ及び第3のコンテンツの両方に対して評価を行った少なくとも一のユーザの評価値が存在し、且つ前記第2のコンテンツ及び前記第3のコンテンツの両方に対して評価を行った少なくとも一のユーザの評価値が存在する前記第3のコンテンツの評価値を用いて、前記第1のコンテンツ及び前記第2のコンテンツの類似度を算出する類似度算出工程と、
前記類似度算出工程で算出した前記第1のコンテンツ及び前記第2のコンテンツの類似度と、前記第1のコンテンツ及び前記第2のコンテンツに対する評価値とを用いて、前記第1のコンテンツ及び前記第2のコンテンツのいずれかに対して評価を行わなかったユーザの予測評価値を算出する予測評価値算出工程と、を含む協調フィルタリング方法。 - 複数のユーザそれぞれのコンテンツに対する評価値を用いる協調フィルタリング方法であって、
特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行わなかった一のユーザについて、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行った他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する予測評価値算出工程を含む協調フィルタリング方法。 - 前記予測評価値算出工程は、特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行わなかった一のユーザについて、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行った他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項9又は10に記載の協調フィルタリング方法。
- 前記予測評価値算出工程は、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツの利用頻度が前記一のユーザの前記特定のコンテンツの利用頻度から所定の閾値以内の差異である他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項11又は12に記載の協調フィルタリング方法。
- 前記予測評価値算出工程は、前記一のユーザの前記特定のコンテンツの利用頻度と、前記特定のコンテンツが属するカテゴリー内の不特定多数のコンテンツの利用頻度が前記一のユーザの前記不特定多数のコンテンツの利用頻度から所定の閾値内の差異である他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項11~13のいずれか1項に記載の協調フィルタリング方法。
- 前記予測評価値算出工程は、前記一のユーザの前記特定のコンテンツの利用頻度と、複数の前記他のユーザそれぞれの前記特定のコンテンツの利用頻度ごとの前記特定のコンテンツに対する評価値の中央値及び平均値のいずれかを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項11~14のいずれか1項に記載の協調フィルタリング方法。
- 前記予測評価値算出工程は、特定のコンテンツを利用したことがあり且つ前記特定のコンテンツに対して評価を行った他のユーザの前記特定のコンテンツの利用頻度と前記特定のコンテンツに対する評価値とに基づいて、前記特定のコンテンツの利用頻度に対する前記特定のコンテンツに対する評価値の関数を導出し、前記一のユーザの前記特定のコンテンツの利用頻度と前記関数とを用いて、前記一のユーザの前記特定のコンテンツに対する予測評価値を算出する、請求項11~15のいずれか1項に記載の協調フィルタリング方法。
Priority Applications (5)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2011545913A JP5403069B2 (ja) | 2009-12-18 | 2009-12-18 | 協調フィルタリングシステム及び協調フィルタリング方法 |
US13/512,368 US9087123B2 (en) | 2009-12-18 | 2009-12-18 | Collaborative filtering using evaluation values of contents from users |
CN200980162673.4A CN102640141B (zh) | 2009-12-18 | 2009-12-18 | 协同过滤系统以及协同过滤方法 |
PCT/JP2009/071141 WO2011074113A1 (ja) | 2009-12-18 | 2009-12-18 | 協調フィルタリングシステム及び協調フィルタリング方法 |
EP09852305.3A EP2515244A4 (en) | 2009-12-18 | 2009-12-18 | COLLABORATIVE FILTRATION SYSTEM AND COLLABORATIVE FILTERING PROCESS |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
PCT/JP2009/071141 WO2011074113A1 (ja) | 2009-12-18 | 2009-12-18 | 協調フィルタリングシステム及び協調フィルタリング方法 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2011074113A1 true WO2011074113A1 (ja) | 2011-06-23 |
Family
ID=44166902
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2009/071141 WO2011074113A1 (ja) | 2009-12-18 | 2009-12-18 | 協調フィルタリングシステム及び協調フィルタリング方法 |
Country Status (5)
Country | Link |
---|---|
US (1) | US9087123B2 (ja) |
EP (1) | EP2515244A4 (ja) |
JP (1) | JP5403069B2 (ja) |
CN (1) | CN102640141B (ja) |
WO (1) | WO2011074113A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2018505456A (ja) * | 2014-12-22 | 2018-02-22 | ロヴィ ガイズ, インコーポレイテッド | メタデータおよび使用データ解析を用いるフィルタリング技術のためのシステムおよび方法 |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10360758B2 (en) | 2011-02-17 | 2019-07-23 | Aristocrat Technologies Australia Pty Limited | Gaming tracking and recommendation system |
US10957152B2 (en) | 2011-02-17 | 2021-03-23 | Aristocrat Technologies Australia Pty Limited | Gaming tracking and recommendation system |
US9387392B1 (en) * | 2011-02-17 | 2016-07-12 | Aristocrat Technologies Australia Pty Limited | Gaming tracking and recommendation system |
CN105718488A (zh) * | 2014-12-04 | 2016-06-29 | 阿里巴巴集团控股有限公司 | 基于计算机系统的推荐方法及其装置 |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2005202540A (ja) | 2004-01-14 | 2005-07-28 | Neo Jsk:Kk | 情報推薦方法 |
JP2007501975A (ja) * | 2003-08-08 | 2007-02-01 | コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ | データ処理システム及びその方法 |
JP2009266096A (ja) * | 2008-04-28 | 2009-11-12 | Sony Corp | 情報処理装置、および関連アイテムの提供方法 |
Family Cites Families (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2006127660A2 (en) * | 2005-05-23 | 2006-11-30 | Picateers, Inc. | System and method for collaborative image selection |
JP2007286817A (ja) * | 2006-04-14 | 2007-11-01 | Lealcom Kk | 情報配信システム、情報配信装置及び情報配信方法 |
US7574422B2 (en) * | 2006-11-17 | 2009-08-11 | Yahoo! Inc. | Collaborative-filtering contextual model optimized for an objective function for recommending items |
US7584171B2 (en) * | 2006-11-17 | 2009-09-01 | Yahoo! Inc. | Collaborative-filtering content model for recommending items |
JP5092575B2 (ja) * | 2007-06-22 | 2012-12-05 | ソニー株式会社 | 情報信号処理装置、情報信号処理方法、並びにその処理方法を実行させるためのプログラム |
JP4524709B2 (ja) * | 2007-12-03 | 2010-08-18 | ソニー株式会社 | 情報処理装置および方法、並びに、プログラム |
CN100581227C (zh) * | 2008-05-16 | 2010-01-13 | 华东师范大学 | 一种引入节目热门度权重的协作过滤推荐方法 |
CN101320461A (zh) * | 2008-07-01 | 2008-12-10 | 浙江大学 | 基于电阻网络和稀疏数据预测的协同过滤方法 |
CN101329683A (zh) * | 2008-07-25 | 2008-12-24 | 华为技术有限公司 | 推荐系统及方法 |
KR101330589B1 (ko) | 2010-12-24 | 2013-11-18 | 김선희 | 이동형 고속 진공 발효 교반기 |
-
2009
- 2009-12-18 WO PCT/JP2009/071141 patent/WO2011074113A1/ja active Application Filing
- 2009-12-18 CN CN200980162673.4A patent/CN102640141B/zh not_active Expired - Fee Related
- 2009-12-18 EP EP09852305.3A patent/EP2515244A4/en not_active Ceased
- 2009-12-18 JP JP2011545913A patent/JP5403069B2/ja not_active Expired - Fee Related
- 2009-12-18 US US13/512,368 patent/US9087123B2/en not_active Expired - Fee Related
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2007501975A (ja) * | 2003-08-08 | 2007-02-01 | コーニンクレッカ フィリップス エレクトロニクス エヌ ヴィ | データ処理システム及びその方法 |
JP2005202540A (ja) | 2004-01-14 | 2005-07-28 | Neo Jsk:Kk | 情報推薦方法 |
JP2009266096A (ja) * | 2008-04-28 | 2009-11-12 | Sony Corp | 情報処理装置、および関連アイテムの提供方法 |
Non-Patent Citations (1)
Title |
---|
See also references of EP2515244A4 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JP2018505456A (ja) * | 2014-12-22 | 2018-02-22 | ロヴィ ガイズ, インコーポレイテッド | メタデータおよび使用データ解析を用いるフィルタリング技術のためのシステムおよび方法 |
Also Published As
Publication number | Publication date |
---|---|
EP2515244A4 (en) | 2015-06-17 |
JPWO2011074113A1 (ja) | 2013-04-25 |
CN102640141B (zh) | 2015-05-13 |
CN102640141A (zh) | 2012-08-15 |
EP2515244A1 (en) | 2012-10-24 |
JP5403069B2 (ja) | 2014-01-29 |
US20120239604A1 (en) | 2012-09-20 |
US9087123B2 (en) | 2015-07-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20110282798A1 (en) | Making Friend and Location Recommendations Based on Location Similarities | |
JP5277307B2 (ja) | 情報推薦方法、そのシステム、及びサーバ | |
US8612134B2 (en) | Mining correlation between locations using location history | |
US20120327119A1 (en) | User adaptive augmented reality mobile communication device, server and method thereof | |
US20120278320A1 (en) | Recommendation information providing system, recommendation information providing apparatus, recommendation information service method, and recommendation information distribution program | |
JP5403069B2 (ja) | 協調フィルタリングシステム及び協調フィルタリング方法 | |
JP7302634B2 (ja) | 経路処理プログラム、経路処理装置および経路処理方法 | |
JP6171388B2 (ja) | ナビゲーションシステム、ナビゲーション方法、及びナビゲーションプログラム | |
US11783215B2 (en) | Information processing apparatus and recommendation control method | |
JP2011003151A (ja) | 類似度算出装置、推薦poi決定装置、poi推薦システム、類似度算出方法およびプログラム | |
Forsati et al. | Web page personalization based on weighted association rules | |
JP7365465B1 (ja) | 推薦装置、推薦方法、およびプログラム | |
JP2012256239A (ja) | 目的地予測装置及びプログラム | |
KR101693015B1 (ko) | 개인 질병 예측 방법, 개인 질병 예측 시스템 및 개인 질병 예측을 위한 프로그램을 저장하는 저장매체 | |
JP3933127B2 (ja) | 旅行時間予測方法、装置及びプログラム | |
JP2020190964A (ja) | 情報処理装置、情報処理方法および情報処理プログラム | |
US11199417B2 (en) | Distributed system for dynamic sensor-based trip estimation | |
US8606793B1 (en) | Business metric score for web pages | |
JP6276722B2 (ja) | 行動予測装置、行動予測方法及び行動予測プログラム | |
JP3674603B2 (ja) | 旅行時間提供方法、システム及びプログラム | |
JPWO2010084629A1 (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 | |
JP6923348B2 (ja) | 情報処理装置、情報処理方法及びプログラム | |
US20250027783A1 (en) | Recommendation method | |
Hawas et al. | Location–Based in recommendation system using naive Bayesian algorithm | |
KR102416686B1 (ko) | 자율주행차량 사용자정보 수집을 통한 유동인구 및 지역 가치 예측 방법 |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
WWE | Wipo information: entry into national phase |
Ref document number: 200980162673.4 Country of ref document: CN |
|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 09852305 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2011545913 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2009852305 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13512368 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |