JP2012118572A - Content recommendation system, content recommendation device, recommendation mode control method, and recommendation mode control program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a content recommendation system which enables a service providing operator to grasp an effect acquired by recommendation and to easily understand what kind of input is required to have an expected effect, and easily applies intention of the operator to the system.SOLUTION: A content recommendation system includes: weight parameter input means 101 for inputting weight parameters indicating influence degrees of respective recommendation modes; effect index value calculation means 106 for calculating an effect index value for each predetermined effect index indicating recommendation effect, in each recommendation mode, by using history data including recommendation to users and its result by the recommendation mode; effect index composition means 102 for calculating values acquired by composing the weight parameter of each recommendation mode and the effect index value for each effect index, as an estimation value of a total effect index; and display means 103 for displaying the effect index value of each recommendation mode with the estimation value of the total effective index for each effect index.

Description

  The present invention relates to a hybrid type content recommendation system and its operation management method, and in particular, for visualization of effects obtained by recommendation (user satisfaction, profit rate, long tail effect, inventory reduction, etc.) and to obtain expected effects. The present invention relates to a content recommendation system, a content recommendation device, a content recommendation method, and a content recommendation program that have excellent controllability and can easily reflect an operator's intention in the content recommendation system.

  In recent years, the importance of a content recommendation system that automatically grasps content that is of interest to the user from a vast amount of content (electronic books, news, videos, music, etc.) and presents it to the user has increased. .

  The content recommendation system maintains a preference analysis algorithm (hereinafter referred to as a recommendation method) that analyzes user preferences from a large amount of content usage logs and purchase logs, and automatically grasps content that is of interest to users. To do. There are various types of recommendation methods such as collaborative filtering base, content base, and Bayesian network base.

  However, each recommendation method has a trade-off relationship that there are advantages and disadvantages. Also, for example, when purchasing a book, an algorithm like collaborative filtering that extracts information similar to your purchase history is suitable, such as `` people who see this also see this '', but restaurant When searching for information, we would like you to extract information in consideration of the companion “family”, “lover” and “friend” who go to the restaurant together rather than their own history, etc. I.e., the recommendation method is different. Therefore, it is generally said that there is no universal recommendation method in all cases.

  Therefore, a hybrid type content recommendation system that can reinforce the weaknesses of the mutual recommendation methods and extract appropriate information in a wide variety of cases has attracted attention.

  Patent document 1 is a hybrid type content recommendation system, which has the advantages of both a content-based recommendation method and a collaborative filtering type recommendation method, makes recommendations that follow user preferences, and also has the number of users and content. The aim is to be able to process content recommendation at high speed even if the number increases.

  Non-Patent Documents 1 and 2 are hybrid type content recommendation systems that have a plurality of algorithm recommendation methods and recommend information according to the situation of the user as well as the user. is there.

  In this way, the hybrid content recommendation system with multiple algorithms aims to realize information recommendation that follows user preferences in all situations by collaborative filtering and content-based preference analysis algorithms. is there.

  However, the hybrid content recommendation system of the related technology is mainly one that automatically follows the user's preference, and it is difficult to understand or expect the effect of the recommendation for service providers. There is a problem that it is difficult to know how much importance each recommendation method should have in order to obtain an effect.

JP 2010-067175 A

Victoria Bellotti et al. : "Activity-based serendipous recommendations with the Magic mobile mobile guide". Proc. Of the twenty-sixth annual SIGCHI conference on Human factors in computing systems, pp. 1157-1166,2008 Yuta Kanno et al. “Realization of a“ multi-mode recommendation system ”that recommends information in a method suitable for the user situation, 71st National Convention of Information Processing Society, pp. 1-473-1-474 (2009)

  The first problem is that in a hybrid type content recommendation system, it is difficult to understand how important each recommendation method should be for the operator to obtain the expected effect, and easily reflects the operator's intention It was difficult.

  The reason is that the hybrid content recommendation system has multiple recommendation methods with different characteristics, so the overall system effects (user satisfaction, profit margin, long tail effect, inventory reduction, etc.) are the effects of multiple recommendation methods. This is because it has a complicated structure obtained by overlapping, and it has not been possible to obtain an indicator of how much each recommendation method should be emphasized in order to obtain the effect expected by the operator .

(Object of invention)
The purpose of the present invention is to solve the above-mentioned problems, grasp the effect obtained by recommendation, make it easy to understand what input should be given in order to obtain the expected effect, and easily make the intention of the operator content It is to provide a content recommendation system, a content recommendation device, a content recommendation method, and a content recommendation program that can be reflected in the recommendation system.

  A first content recommendation system of the present invention is a content recommendation system including a user terminal and a content recommendation device that recommends content using a plurality of recommendation methods, and a weight parameter indicating the degree of influence of each recommendation method. For each recommendation method, a weight parameter input means to input and an effect indicator indicating a predetermined recommendation effect of the recommendation method, using the history data including the recommendation to the user by the recommendation method and the result thereof, an effect index value is obtained. An effect index value calculating means for calculating, an effect index combining means for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value for each effect index as an estimated value of the overall effect index, and an effect index Each includes display means for displaying the effect index value of each recommendation method and the estimated value of the overall effect index together.

  A first content recommendation device of the present invention is a content recommendation device that receives a predetermined request from a user terminal and returns the result, a weight parameter input means for inputting a weight parameter indicating the degree of influence of each recommendation method, Effect index value calculating means for calculating an effect index value for each recommendation method using history data including recommendation to the user by the recommendation method and the result thereof, for each effect indicator indicating a predetermined recommendation effect of the recommendation method; For each effect index, an effect index composition means for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the overall effect index, and the effect of each recommendation method for each effect index Display means for displaying the index value and the estimated value of the overall effect index together.

  A first recommendation method control method according to the present invention is a recommendation method control method of a content recommendation device that receives a predetermined request from a user terminal and returns the result, and inputs a weight parameter indicating the degree of influence of each recommendation method. For each recommendation method, an effect index value is calculated for each recommendation method using history data including the recommendation to the user and the result of the recommendation for each recommendation method indicating the predetermined recommendation effect of the recommendation method. For each effect index, an effect index value calculating step, a value obtained by combining the weight parameter of each recommended method and the effect index value for each effect index, and an estimated value of the overall effect index, and for each effect index And a display step of displaying the effect index value of each recommendation method and the estimated value of the overall effect index together.

  A first recommendation method control program of the present invention is a recommendation method control program that operates on a computer functioning as a content recommendation device that receives a predetermined request from a user terminal and returns the result, and each recommendation method is transmitted to the computer. For each recommendation method, history data including the recommendation to the user by the recommendation method and the result for each of the effect parameters indicating the predetermined recommendation effect of the recommendation method and the weight parameter input process for inputting the weight parameter indicating the degree of influence of the recommendation method Is used to calculate the effect index value calculation process for calculating the effect index value, and for each effect index, a value obtained by combining the weight parameter of each recommendation method and the effect index value is calculated as the estimated value of the overall effect index. The effect index synthesis process and the display process for displaying the effect index value of each recommended method and the estimated value of the overall effect index for each effect index are executed. .

  The first effect is that it is easy to understand what input should be given in order to grasp the effect obtained by recommendation and obtain the expected effect in the hybrid type content recommendation system.

  The reason for this is that the effect of each recommendation method of the hybrid type content recommendation system and the estimated value of the effect of the entire system that changes depending on the input value of the weight parameter are displayed together. This is because the recommended weight parameter can be input while sequentially confirming how the estimated value of the effect of the entire system changes depending on the input value of the weight parameter.

It is a block diagram which shows the structure of the content recommendation system by the 1st Embodiment of this invention. It is a flowchart which shows operation | movement of the weight parameter setting process by 1st Embodiment. It is a flowchart which shows the operation | movement of the recommendation execution process by 1st Embodiment. It is a flowchart which shows the operation | movement of the weight parameter setting process by Example 1 of this invention. 6 is a flowchart illustrating an operation of a recommendation execution process according to the first embodiment. It is the figure which showed the illustration of the recommendation system weight parameter input means in Example 1, and a display means. It is a block diagram which shows the structure of the content recommendation system by the 2nd Embodiment of this invention. It is a flowchart which shows the operation | movement of the effect parameter | index calculation by 2nd Embodiment. It is a flowchart which shows the operation | movement of the effect parameter | index calculation by Example 2 of this invention. It is a block diagram which shows the structure of the content recommendation system by the 3rd Embodiment of this invention. It is a flowchart which shows operation | movement of the weight parameter setting process by 3rd Embodiment. It is a flowchart which shows the operation | movement of the recommendation execution process by 3rd Embodiment. It is a flowchart which shows the operation | movement of the weight parameter setting process by Example 3 of this invention. It is the figure which showed the illustration of the weight parameter input means by Example 3, and a display means. 12 is a flowchart illustrating an operation of a recommendation execution process according to the third embodiment. It is a block diagram which shows the hardware structural example of the content recommendation apparatus of this invention.

  Embodiments of the present invention will be described in detail with reference to the drawings. In all the drawings, the same reference numerals are given to the same components, and the description will be omitted as appropriate.

(First embodiment)
A first embodiment of the present invention will be described in detail with reference to the drawings. In the following drawings, the configuration of parts not related to the essence of the present invention is omitted as appropriate and is not shown.

  FIG. 1 is a block diagram showing a configuration of a content recommendation system 1000 according to the first embodiment of the present invention.

  Referring to FIG. 1, a content recommendation system 1000 according to the present embodiment includes a user terminal 200 and a content recommendation device 100 that recommends content to a user.

  The user terminal 200 transmits a predetermined recommendation request to the content recommendation device and receives the result. The recommendation request includes a user identifier, the number of recommended contents to be acquired, and the like. The content of the recommendation request can be arbitrarily set.

  In addition, in this Embodiment, although it was set as the structure which has one user terminal for convenience of explanation, there is no restriction | limiting in the number of user terminals.

  The content recommendation device 100 includes a transmission / reception unit 111 that transmits / receives data to / from the user terminal 200, a recommended content extraction unit 107 that extracts content recommended to the user, and a weight parameter input unit that inputs a weight parameter of a recommendation method. 101, an effect index storage unit 109 that stores an effect index set for a recommendation method, an effect index value calculation unit 106 that calculates an effect index value for each recommendation method, and an effect index value and a weight parameter are combined. From the user terminal, the effect index composition means 102 for calculating the estimated value of the effect index of the entire content recommendation system 1000, the display means 103 for displaying the effect index value of the recommendation method and the effect index estimated value of the entire system, and the user terminal It is configured.

  The weight parameter input unit 101 receives an input of a weight parameter that controls the degree of influence of the recommendation method.

  Further, the weight parameter input unit 101 records the weight parameter set by the operator in the weight parameter storage unit 110.

  The effect index storage unit 109 stores an effect index for the recommendation method. The effect index is set in advance by an operator or the like.

  The effect index is an index indicating the effect of the recommendation method. As an effect index, a feedback rate, a profit rate, and the like are set. The effect index can be arbitrarily set as an index that can be calculated using the history data.

  The history data is data including a history of the user accessing the content and a history of recommendation executed in the content recommendation system 1000, and includes information that can calculate an effect index such as a feedback rate and a profit rate. .

  The effect index value calculating unit 106 refers to the history data, and calculates an effect index value (effect index value) set in the effect index storage unit 109 for each recommendation method.

  The effect index synthesizing unit 102 calculates an effect index estimated value (effect index estimated value) of the entire content recommendation system 1000 by synthesizing the effect index value of each recommendation method and the weight parameter of each recommendation method. When there are a plurality of effect indices, the effect index synthesizing unit 102 calculates an effect index estimated value for each effect index.

  The display means 103 displays the effect index value of each recommendation method and the estimated value of the effect index of the entire system in the input weight parameter.

  The transmission / reception unit 111 receives a recommendation request from the user terminal 200. The query included in the recommendation request includes a user identifier, the number of contents to be acquired, and the like.

  The recommended content extraction unit 107 has a plurality of recommendation methods therein, and extracts recommended content recommended to the user from the content data using the recommendation methods.

  The recommended content extraction unit 107 determines the ratio of the recommended method to be used based on the weight parameter for each recommendation method recorded in the weight parameter storage unit 110, and extracts the recommended content from the content data.

  For example, when the ratio of the weight parameter is “(recommendation method A) 1: (recommendation method B) 2”, the recommended content extraction unit 107 extracts the recommended content extracted by the recommendation method A and the recommendation method B. The recommended content is extracted so that the ratio of the recommended content is 1: 2.

(Description of the operation of the first embodiment)
Next, the operation of the content recommendation system 1000 according to the present embodiment will be described in detail with reference to the drawings.

(Description of weight parameter setting processing)
First, the operation of the weight parameter setting process recorded in the weight parameter storage unit 110 will be described.

  In the present embodiment, it is assumed that the effect index value calculation unit 106 calculates the effect index stored in the effect index storage unit 109 for each recommendation method in advance.

  FIG. 2 is a flowchart showing the operation of the content recommendation system 1000 according to this embodiment.

  Referring to FIG. 2, first, the weight parameter input unit 101 receives an input of a weight parameter for controlling the degree of influence of each recommendation method from the operator (step A1).

  Next, the effect index synthesizing unit 102 synthesizes the effect index value of each recommendation method and the weight parameter of each recommendation method, thereby obtaining an estimated value (effect index estimate value) of the content recommendation system 1000 as a whole. calculate. (Step A2).

  When there are a plurality of effect indices, the process of step A3 is performed for each effect index.

  Next, the display means 103 displays the effect index value of each recommendation method and the effect index estimated value of the entire system that changes depending on the input value of the weight parameter (step A3).

  The operator can determine whether or not the input of the weight parameter is completed by determining that the balance of the effect index displayed in Step A3 is appropriate. If not, the operator returns to Step A1 to perform the processing. It can be repeated (step A4).

  Finally, when the completion of setting of the weight parameter input by the operator is instructed (step A4 “YES”), the weight parameter input unit 101 records the weight parameter in the weight parameter storage unit 110 (step A5). .

(Explanation of recommendation execution process)
FIG. 3 is a flowchart showing the operation of recommendation execution processing in the content recommendation system 1000 according to the present embodiment.

  Referring to FIG. 3, first, the transmission / reception unit 111 receives a recommendation request from the user terminal 200 (step C1).

  Next, the recommended content extraction unit 107 acquires the weight parameter of each recommendation method from the weight parameter input unit 101 (step C2).

  Next, the recommended content extraction unit 107 performs filtering and scoring of content using one or more recommendation methods having the obtained weight parameter of 0 or more among the recommendation methods held therein, and recommends them. Content is extracted (step C3).

  Then, the recommended content extraction unit 107 selects content from each recommendation method based on the weight parameter ratio, and aggregates the content to be finally recommended (step C4).

  If there is only one recommendation method with a weight parameter of 0 or more, the content extracted by the recommendation method becomes the content to be recommended finally.

  Finally, the transmission / reception unit 111 transmits the obtained content to the user terminal as a final result (step C5).

  Next, the operation of this embodiment will be described using specific examples.

  In this embodiment, it is assumed that the feedback rate and the profit rate are set in advance in the recommended effect index setting unit 109 as an effect index of the recommendation method.

  The feedback rate is [number of times content is used through recommendation of a certain recommendation method / recommendation number of recommendation method], and the profit rate is [income earned through recommendation of a recommendation method / recommendation number of a recommendation method]. It is assumed that it is set as follows. However, it is not limited to this.

  Further, it is assumed that two recommendation methods, a content-based recommendation method (CB) and a collaborative filtering-based recommendation method (CF), are used as recommendation methods of the hybrid content recommendation system.

  In addition, the effect index value calculation unit 106 preliminarily feedbacks {F (CB), F (CF)} of two recommendation methods, a content-based recommendation method (CB) and a collaborative filtering-based recommendation method (CF). , And profit rate {B (CB), B (CF)}. It may be calculated immediately before or after the weight parameter input unit 101 receives an input.

(Description of weight parameter setting processing)
FIG. 4 is a flowchart showing the operation of the weight parameter setting process according to the first embodiment of the present invention.

  Referring to FIG. 4, first, the weight parameter input unit 101 receives an input of a weight parameter for each recommendation method from the operator (step A1 ').

  Referring now to FIG. 6, 10 and 20 in FIG. 6 are diagrams showing examples of input screens of the weight parameter input means 101.

  The operation of step A1 ′ is performed by, for example, using the equalizer 11 and the equalizer 12 by using the equalizer 11 and the equalizer 12, and the equalizer 21 and the equalizer. Change to 22

  Hereinafter, the weight parameter of the CB for which the weight parameter input unit 101 has received the input is referred to as W (CB), and the weight parameter of the CF is referred to as W (CF).

  Note that the weight parameter of each recommendation method is a parameter indicating the degree of influence on how strongly the result from the recommendation method influences the final recommendation result in the recommendation execution of the content recommendation system 1000.

  Next, the effect index synthesizing unit 102 provides feedback rates {F (CB), F (CF)} and profits of the two recommendation methods of the content-based recommendation method (CB) and the collaborative filtering-based recommendation method (CF). The rate {B (CB), B (CF)} is weighted with a predetermined weight based on the weight parameter of each recommendation method input in step A1 ′, and then each value is synthesized, and the feedback rate and profit rate of the entire system are synthesized. An estimated value is calculated (step A2 ′).

  The feedback rate of the entire system can be calculated as F (ALL) = W (CB) * F (CB) + W (CF) * F (CF). However, it is not limited to this.

  The profit rate of the entire system can be calculated as B (ALL) = W (CB) * B (CB) + W (CF) * B (CF). However, it is not limited to this.

  Finally, the display unit 103 calculates the feedback rate {F (CB), F (CF)} and the profit rate {B (CB), B (CF)} for each recommendation method, and the entire system calculated in step A3 ′. The feedback rate F (ALL) and the profit rate B (ALL) are displayed (step A3 ′).

  Referring now to FIG. 12, 30 and 40 in FIG. 12 are examples of the display means 103.

  For example, 31c and 41c are feedback rates of the entire system. 31 c is a value obtained by weighting and synthesizing the content-based feedback rate 31 a and the collaborative filtering feedback rate 31 b based on the content-based weight parameter 11 and the collaborative filtering weight parameter 12.

  The operator can determine whether or not the input of the weight parameter is completed by determining that the balance of the effect index displayed in step A3 ′ is appropriate. If not, return to step A1 ′. The process can be repeated (step A4 ′).

Finally, when the completion of setting of the weight parameter input by the operator is instructed (step A4 '"YES"), the weight parameter input means 101 records the weight parameter in the weight parameter storage means 110 (step A5). ').

  Referring to FIG. 12, when the input value of the weight parameter input unit 101 that changes the weight parameter of each recommendation method is changed from the value shown in 10 to the value shown in 20, the feedback rate that is an index of the recommendation effect and The value of the profit rate changes from 30 to 40. When the weight parameter is changed to 21 and 22, the estimated value of the overall feedback rate also changes from 31c to 41c. The same applies to the profit margin.

  In this way, by changing the weight parameter, it is possible to visualize the effects of the entire system and the balance of each effect index in an easy-to-understand manner. It is possible to input a weight parameter according to the index.

(Explanation of recommendation execution process)
FIG. 5 is a flowchart showing the operation of the recommendation execution process according to the first embodiment of the present invention.

  Referring to FIG. 5, first, the transmission / reception unit 111 receives a recommendation request from the user terminal 200 (step C1 ').

  Next, the recommended content extraction unit 107 acquires from the weight parameter input unit 101 the weight parameters of the content-based recommendation method (CB) and collaborative filtering-based recommendation method (CF) set by the operator. (Step C2 ').

  Here, the weight parameter of CB is W (CB), and the weight parameter of CF is W (CF).

  Next, the recommended content extraction unit 107 performs content filtering and scoring using a recommendation engine of a content-based recommendation method (CB) and a collaborative filtering-based recommendation method (CF) held therein to make a recommendation. Content is extracted (step C3 ′).

  Then, the recommended content extraction unit 107 selects content from each recommendation method based on the weight parameter ratio, and aggregates the content to be finally recommended (step C4 ').

  Since the weight parameter of CB is W (CB) and the weight parameter of CF is W (CF), the recommended content extraction unit 107 has a content-based recommendation method (CB) and a collaborative filtering-based recommendation method (CF). From the recommendation result, content is selected from each recommendation method at a ratio of [CB: CF] = [W (CB): W (CF)], and finally recommended content is aggregated.

  Finally, the transmission / reception unit 111 transmits the obtained content to the user terminal as a final result (step C5 ').

(Effects of the first embodiment)
Next, the effect of this embodiment will be described.

  According to this embodiment, the operator who provides the recommendation service can grasp the effect obtained by the recommendation, and how to weight each recommendation method in order to obtain the expected effect. This makes it easier to understand what should be done and can easily reflect the operator's intention in the content recommendation system.

  In addition, according to the present embodiment, the recommended content extraction unit 107 executes the actual recommendation using the weight parameter input by the operator so that the expected effect can be obtained. Can be realized.

  Note that the above-described problem can be solved even with the minimum configuration including the weight parameter input unit 101, the effect index synthesis unit 102, the display unit 103, and the effect index value calculation unit 104.

(Second Embodiment)
A second embodiment of the present invention will be described in detail with reference to the drawings. In the following drawings, the configuration of parts not related to the essence of the present invention is omitted and is not shown.

  FIG. 7 is a block diagram showing the configuration of the recommended operation management apparatus 100 according to the second embodiment of the present invention.

  Referring to FIG. 7, the recommended operation management apparatus 100 according to the present embodiment includes, in the first embodiment of the present invention, filter condition storage means 104 for storing a condition for filtering history data, and history data according to the condition. The configuration further includes history condition filtering means 105 for performing condition filtering.

  Each of these means generally operates as follows.

  The filter condition storage unit 104 records a predetermined condition used by the history condition filter unit 105 for filtering history data. The condition can be arbitrarily set, for example, a user profile condition, a context condition such as position information and time information, a condition of history freshness, and the like.

  The conditions recorded in the filter conditions can be set in advance.

  Further, it may be set at the time of input by the weight parameter input means 101. In this case, the history condition filter unit 105 and the effect index value calculation unit 106 perform processing after setting the conditions.

  The history condition filter unit 105 filters history data using the conditions recorded in the filter condition storage unit 104.

  For example, it is possible to perform condition filtering only on a target history based on a user profile condition, a context condition such as position information and time information, and a newness condition of the history. Further, conditional filtering that changes the degree of influence depending on the history to be used may be used, such as an attenuation function with the passage of time of the history.

  The effect index value calculation means 106 calculates the value of the effect index set in the effect index storage means 109 for each recommendation method, using the history data that has been subjected to the condition filtering by the history condition filter means 105.

  Since other means are the same as those in the first embodiment, description thereof will be omitted.

(Description of operation of second embodiment)
Next, the operation of the recommended operation management apparatus 100 according to the present embodiment will be described in detail with reference to the drawings.

(Explanation of effect index calculation operation)
FIG. 8 is a flowchart showing an effect index calculation operation in the content recommendation system 100 according to the present embodiment.

Referring to FIG. 8, the history condition filter unit 105 first filters history data using a predetermined condition set in the filter condition storage unit 104 (step B1).

  Next, the effect index value calculating means 106 calculates an effect index for each recommendation method using the history data subjected to the conditional filtering (step B2).

  Since the operations of the weight parameter setting process and the recommendation execution process are the same as in the first embodiment, detailed description thereof is omitted.

  Next, the operation of this embodiment will be described using specific examples.

  FIG. 9 is a flowchart showing an effect index calculation operation according to the second embodiment of the present invention.

  As a premise, as in the first embodiment, it is assumed that the feedback rate and the profit rate are set in advance in the recommended effect index setting unit 109 as the effect index of the recommendation method. The feedback rate and profit rate are also defined in the same manner as in the first embodiment.

  Further, it is assumed that a content-based recommendation method (CB) and a collaborative filtering-based recommendation method (CF) are used as a recommendation method of the hybrid content recommendation system.

  In the filter condition storage unit 104, it is assumed that [20s / female] is set as the history condition.

  Referring to FIG. 11, first, the history condition filter unit 105 refers to the history condition set in the history condition 104, and filters only the history that meets the condition from the history data (step B1 ').

  Next, the effect index value calculation means 106 calculates an effect index for each recommendation method using the filtered history data (step B2 ').

  Here, a content-based recommendation method (CB) and a collaborative filtering-based recommendation method (CF) are used as recommendation methods, and the effect index value calculation means 106 calculates a feedback rate and a profit rate for each recommendation method. Calculate each.

  Since the operations of the weight parameter setting process and the recommendation execution process are the same as in the first embodiment, detailed description thereof is omitted.

  Thus, by calculating the effect index of the content recommendation system in the history filtered under a specific condition, it is possible to input a weight parameter for a specific layer.

(Effects of the second embodiment)
Next, the effect of this embodiment will be described.

  According to the present embodiment, it is configured to calculate the effect index of each recommendation method and the effect index of the entire system using the history subjected to the conditional filtering. In view of this, the operator's intention can be reflected in the content recommendation system.

(Third embodiment)
A third embodiment of the present invention will be described in detail with reference to the drawings. In the following drawings, the configuration of parts not related to the essence of the present invention is omitted and is not shown.

  FIG. 10 is a block diagram showing a configuration of a content recommendation system 1000 according to this embodiment.

  Referring to FIG. 10, in addition to the first embodiment, the content recommendation system 1000 according to the present embodiment includes a weight parameter learning unit 112 that learns weight parameters of a recommendation method for each individual or situation from history data. , A weight parameter combining unit 108 for combining the weight parameter recorded in the weight parameter storage unit 110 and the weight parameter learned by the weight parameter learning unit 112 is provided.

  In addition, it is good also as a structure which adds the weight parameter learning means 112 and the weight parameter synthetic | combination means 108 to 2nd Embodiment.

  Each of these means generally operates as follows.

  The weight parameter learning unit 112 learns the importance of each recommendation method as a weight parameter for each predetermined condition from the history data.

  For example, when recommending a restaurant, learn the weight parameter of each recommendation method based on the feedback rate of each recommendation method, and when purchasing a book, when purchasing a book, the weight parameter of each recommendation method based on the profit rate of each recommendation method Can learn. However, the present invention is not limited to this, and the weight parameter learning method can be arbitrarily set.

  The weight parameter combining unit 108 combines the weight parameter of each recommended method learned by the weight parameter learning unit 112 and the weight parameter of each recommended method recorded in the weight parameter storage unit 110. As a result, a synthesized weight parameter is generated for each condition learned by the weight parameter learning unit 112.

  The recommended content extraction unit 107 performs content filtering and scoring using one or more recommendation methods in which the weight parameter synthesized by the weight parameter synthesis unit 108 is 0 or more among the recommendation methods held inside. , Extract content.

  In this case, the weight parameter synthesized by the weight parameter synthesizing unit 108 exists for each predetermined condition, but the weight parameter that matches the recommendation request from the user terminal 200 is used.

  In addition, when there are a plurality of recommended methods having the combined weight parameter, the recommended content extracting unit 107 selects content from each recommended method based on the ratio of the weight parameters, and aggregates the content to be finally recommended.

  Since the transmission / reception means 111 is the same as that of the first embodiment, the description thereof is omitted.

  The effect index synthesizing unit 102 synthesizes the effect index value of each recommendation method and the weight parameter generated by the weight parameter synthesizing unit 108 to thereby estimate the effect index of the entire content recommendation system 1000 (effect index estimated value). ) Is calculated.

  Since the weight parameter input unit 101 and the display unit 103 are the same as those in the first embodiment, description thereof is omitted.

(Description of the operation of the third embodiment)
Next, the operation of the content recommendation system 1000 according to the present embodiment will be described in detail with reference to the drawings.

  In the present embodiment, as a preliminary step, the weight parameter learning unit 112 learns the weight parameter for each recommendation method for each predetermined case (user, situation, etc.).

  Further, it is assumed that the effect index value calculation unit 106 calculates the effect index value of each recommendation method in advance.

(Description of weight parameter setting processing)
First, the operation of the weight parameter setting process in the content recommendation system 1000 according to the present embodiment will be described with reference to FIG.

  Referring to FIG. 11, first, the weight parameter input unit 101 receives an input of a weight parameter of each recommended method for the effect index from the operator (step D1).

  Next, the weight parameter synthesizing unit 108 synthesizes the weight parameter of the recommended method for each individual or situation learned by the weight parameter learning unit 112 and the weight parameter of the recommended method input from the weight parameter input unit 101 (step D2). ).

  Next, the effect index synthesizing unit 102 acquires the effect index value and the weight parameter for each recommendation method from the effect index value calculating unit 106 and the weight parameter synthesizing unit 108, respectively, and synthesizes them, and the content recommendation system 1000 An estimated value of the overall recommended effect index is calculated (step D3).

  Next, the display means 103 displays the effect index value of each recommendation method and the recommended effect value of the entire content recommendation system 1000 as an index indicating what input should be given in order to obtain the expected effect ( Step D4). Here, in practice, the effect index value of the entire content recommendation system 1000 is a value that differs for each individual or situation, but an average value of the effect may be displayed for the sake of convenience.

  Finally, when the weight parameter determination process is performed by the operator's determination that the balance between the effect index values of the entire content recommendation system 1000 and each effect index is appropriate, the weight parameter input unit 101 displays the determined weights. The parameter is recorded (step D5, step D6).

(Explanation of recommendation execution process)
Next, with reference to FIG. 12, the operation of the recommendation execution process in the content recommendation system 1000 according to the present embodiment will be described.

  Referring to FIG. 12, first, the transmission / reception unit 111 receives a recommendation request from the user terminal 200 (step E1).

  Next, the weight parameter synthesizing unit 108 acquires the weight parameter input from the weight parameter input unit 101 and the weight parameter learned by the weight parameter learning unit 112 for each individual / situation (step E2).

  Then, the weight parameter combining unit 108 combines the acquired two weight parameters (step E3).

  Next, the recommended content extraction unit 107 performs content filtering and scoring using one or more recommendation methods having a combined weight parameter of 0 or more among the recommendation methods held inside, and recommending them. Content to be extracted is extracted (step E4).

Then, the recommended content extraction unit 107 collects the content to be finally recommended from the recommended content extracted by each recommendation method based on the weight parameter ratio (step E5).

  If there is only one recommendation method with a weight parameter of 0 or more, the content extracted by the recommendation method becomes the content to be recommended finally.

  Finally, the transmission / reception unit 111 transmits the obtained content to the user terminal as a final result (step E6).

  Next, the operation of this embodiment will be described using specific examples.

  Also in this embodiment, first, as a premise, it is assumed that a feedback rate and a profit rate are set in advance in the recommended effect index setting unit 109 as a recommended effect index. The feedback rate and profit rate are defined in the same manner as in the example of the first embodiment of the present invention.

  Also, the recommended content extraction unit 107 includes a content-based recommendation method (CB) and a collaborative filtering-based recommendation method (CF) recommendation engine.

  Further, it is assumed that the weight parameter learning unit 112 learns the weight parameter of the recommendation method for each individual or situation from the history data.

(Description of weight parameter setting processing)
FIG. 13 is a flowchart showing the operation of the weight parameter setting process in the content recommendation system 1000 according to this embodiment.

  Referring to FIG. 13, first, the weight parameter input unit 101 receives an input of a weight parameter of each recommendation method for the effect index from the operator (step D1 ').

  At this time, the weight parameter input unit 101 combines the average value of the weight parameters that are different for each individual or situation and the input weight parameter learned by the weight parameter learning unit 112 as shown in 11b ′ and 12b ′ of FIG. It may be presented in the form.

  Referring to FIG. 14, 11 a ′ and 12 a ′ are weight parameters input by the operator, and 11 b ′ and 12 b ′ are average values of the weight parameters learned by the weight parameter learning unit 112.

  When the input value of the weight parameter is changed from 10 ′ to 20 ′, 11a ′ and 12a ′ can be changed to 21a ′ and 22a ′ by an operator input, but 11b ′ and 12b ′ are operators. It does not change depending on the input.

  In FIG. 14, 11b ′, 12b ′, 21b ′, and 22b ′ each start from the end points of 11a ′, 12a ′, 21a ′, and 22a ′, and are always learned even if the equalizer is moved. The value of the weight parameter is visible.

  When the weight parameter input unit 101 accepts the input, the weight parameter synthesis unit 108 then synthesizes the weight parameter learned by the weight parameter learning unit 112 and the weight parameter accepted by the weight parameter input unit 101, and A weight parameter is calculated (step D2 ′).

  Here, the CB weight parameter 21a ′ input by the operator is set to W1 (CB), and the CF weight parameter 22a ′ input by the operator is set to W1 (CF). Assuming that 21b ′ as a parameter is W2 (CB) and 22b ′ as a CF weight parameter obtained by learning is W2 (CF), the CB weight parameter W (CB) of the entire system is W (CB) = W1 (CB) + W2 (CB), and W (CF) is the weight parameter of CF of the entire system, and W (CF) = W1 (CF) + W2 (CF).

  Next, the effect index synthesizing unit 102 uses the feedback rate and profit rate of the two recommendation methods of the content-based recommendation method (CB) and the collaborative filtering-based recommendation method (CF) calculated by the effect index value calculating unit 106; The weight parameters calculated by the weight parameter combining means 108 are combined to calculate the feedback rate and profit rate of the entire content recommendation system 1000 (step D3 ′).

  The feedback rate of the entire content recommendation system 1000 can be calculated as F (ALL) = W (CB) * F (CB) + W (CF) * F (CF).

  The profit rate of the entire content recommendation system 1000 can be calculated as B (ALL) = W (CB) * B (CB) + W (CF) * B (CF).

  Next, the display means 103 is used to obtain the expected effect of the feedback rate and profit rate of each recommendation method and the estimated value of the overall system feedback rate and profit rate calculated in step D3 ′. It is displayed as an index as to whether an input should be given (step D4 ′).

  Finally, when the weight parameter determination process is performed by the operator's determination that the balance between the effect index values of the entire content recommendation system 1000 and each effect index is appropriate, the weight parameter input unit 101 displays the determined weights. The parameters are recorded (step D5 ′, step D6 ′).

(Explanation of recommendation execution process)
FIG. 15 is a flowchart showing the operation of the weight parameter setting process in the content recommendation system 1000 according to this embodiment, and FIG. 15 is a flowchart showing the operation of the recommendation execution process in the content recommendation system 1000 according to this embodiment.

  Referring to FIG. 15, first, the transmission / reception unit 111 receives a recommendation request from the user terminal 200 (step E1 '). The query included in the recommendation request includes a user identifier, the number of contents to be acquired, and the like.

  Next, the weight parameter composition unit 108 inputs the weight parameter of the content-based recommendation method (CB) and the collaborative filtering-based recommendation method (CF) input through the weight parameter input unit 101, and the weight parameter learning unit 112 stores the individual / situation. The content-based recommendation method (CB) and the weighting parameter of the collaborative filtering-based recommendation method (CF) learned each time are acquired (step E2 ′).

  Then, the weight parameter combining unit 108 combines the weight parameters of the content-based recommendation method (CB) and the collaborative filtering-based recommendation method (CF) acquired from the weight parameter learning unit 112 and the weight parameter input unit 101. Perform (step E3 ').

  That is, the weight parameter synthesizing unit 108 synthesizes the weight parameters of each content-based recommendation method (CB) acquired from the weight parameter learning unit 112 and the weight parameter input unit 101. Similarly, the weight parameter learning unit 112 and the weight parameter The weighting parameter of each collaborative filtering-based recommendation method (CF) acquired from the input unit 101 is synthesized.

  As a result, the system-wide weight parameter for the content-based recommendation method (CB) and the system-wide weight parameter for each collaborative filtering-based recommendation method (CF) are calculated.

  Here, the weight parameter of CB inputted by the operator is W1 (CB), the weight parameter of CF is W1 (CF), the weight parameter of CB obtained by learning is W2 (CB), and the weight parameter of CF Is W2 (CF), the CB weight parameter W (CB) of the entire content recommendation system 1000 is W (CB) = W1 (CB) + W2 (CB), and the CF weight parameter of the entire system is W (CF). Is W (CF) = W1 (CF) + W2 (CF).

  Next, the recommended content extraction unit 107 extracts the recommended content by filtering and scoring the content using the content-based recommendation method (CB) and the collaborative filtering-based recommendation method (CF) held therein. (Step E4 ′).

  Further, the recommended content extraction unit 107 aggregates the content to be finally recommended from the recommended content extracted by each recommendation method based on the ratio of the combined weight parameters (step E5 ').

  Here, [CB: CF] = [W (CB): W (CF)], so the recommended content extracting unit 107 recommends each of the content-based recommendation method (CB) and the collaborative filtering-based recommendation method (CF). Then, content is selected at a ratio of [W (CB): W (CF)], and the final recommendation result is obtained.

  Finally, the transmission / reception unit 111 transmits the obtained content to the user terminal as a final result (step E6 ').

  In this way, by combining the weight parameter learned for each individual or situation and the weight parameter input as the operator's intention, and executing the recommendation using the combined weight parameter, It is possible to make a recommendation reflecting both the preference that changes according to the situation and the intention of the operator.

(Effects of the third embodiment)
Next, the effect of this embodiment will be described.

  According to the present embodiment, the weight parameter learned for each individual or situation is synthesized with the weight parameter input as the operator's intention, and the recommendation is executed using the synthesized weight parameter. There is a merit that the recommendation reflecting both the preference changing according to the individual person and the situation and the intention of the operator can be easily realized.

  Next, a hardware configuration example of the content recommendation device 100 of the present invention will be described with reference to FIG. FIG. 16 is a block diagram illustrating a hardware configuration example of the content recommendation device 100.

  Referring to FIG. 16, the content recommendation device 100 has a hardware configuration similar to that of a general computer device, and includes a data work area including a memory such as a CPU (Central Processing Unit) 801 and a RAM (Random Access Memory). And a main storage unit 802 used for a temporary data saving area, a communication unit 803 that transmits and receives data via a network, an input / output interface that transmits and receives data by connecting to the input device 805, the output device 806, and the storage device 807 A unit 804 and a system bus 808 for interconnecting the above components. The storage device 807 is realized by, for example, a hard disk device including a non-volatile memory such as a ROM (Read Only Memory), a magnetic disk, and a semiconductor memory.

  Each means of the content recommendation device 100 of the present invention implements its operation in hardware by mounting circuit components which are hardware components such as LSI (Large Scale Integration) incorporating a program. As another example, a program that provides the function may be stored in the storage device 807, loaded into the main storage unit 802, and executed by the CPU 801 to be realized in software.

  Although the present invention has been described with reference to the preferred embodiments, the present invention is not necessarily limited to the above embodiments, and various modifications can be made within the scope of the technical idea. .

  It should be noted that any combination of the above-described constituent elements and a conversion of the expression of the present invention between a method, an apparatus, a system, a recording medium, a computer program, etc. are also effective as an aspect of the present invention.

  The various components of the present invention do not necessarily have to be independent of each other. A plurality of components are formed as a single member, and a single component is formed of a plurality of members. It may be that a certain component is a part of another component, a part of a certain component overlaps with a part of another component, or the like.

  Moreover, although the several procedure is described in order in the method and computer program of this invention, the order of the description does not limit the order which performs a several procedure. For this reason, when implementing the method and computer program of this invention, the order of the several procedure can be changed in the range which does not interfere in content.

  The plurality of procedures of the method and the computer program of the present invention are not limited to being executed at different timings. For this reason, another procedure may occur during the execution of a certain procedure, or some or all of the execution timing of a certain procedure and the execution timing of another procedure may overlap.

  Further, a part or all of the above-described embodiment can be described as in the following supplementary notes, but is not limited thereto.

(Appendix 1)
A content recommendation system comprising a user terminal and a content recommendation device that recommends content using a plurality of recommendation methods,
A weight parameter input means for inputting a weight parameter indicating the degree of influence of each recommendation method;
For each effect index indicating a predetermined recommended effect of the recommendation method, an effect index value calculation unit that calculates an effect index value for each recommendation method using history data including recommendation to the user by the recommendation method and the result thereof When,
For each effect index, an effect index combining means for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A content recommendation system comprising: a display unit that displays the effect index value of each recommendation method and the estimated value of the entire effect index for each effect index.

(Appendix 2)
The effect index synthesis means includes
For each of the recommendation methods, performing a predetermined weighting on the effect index value based on the weight parameter, and adding up the weighted effect index values to calculate an estimated value of the entire effect index The content recommendation system according to supplementary note 1, which is characterized.

(Appendix 3)
The effect index is
The content recommendation system according to Supplementary Note 1 or Supplementary Note 2, wherein the content recommendation system is a feedback rate calculated from the number of times of recommendation by the recommendation method and the number of times the content has been used through recommendation of the recommendation method.

(Appendix 4)
The effect index is
The content recommendation system according to Supplementary Note 1 or Supplementary Note 2, wherein the content recommendation system is a profit rate calculated from the number of times of recommendation by the recommendation method and the amount of income generated through the recommendation of the recommendation method.

(Appendix 5)
Including history condition filtering means for filtering the history data according to a predetermined condition set in advance,
The effect index calculating means is
The content recommendation system according to any one of appendix 1 to appendix 4, wherein the effect index value is calculated using the filtered history data.

(Appendix 6)
Any one of appendix 1 to appendix 5 including recommended content extraction means for extracting recommended content to be recommended using each recommendation method based on the weight parameter input from the weight parameter input means. The content recommendation system described in 1.

(Appendix 7)
Weight parameter learning means for learning the importance of each recommendation method as a weight parameter for each predetermined condition from the history data;
A weight parameter combining unit that combines the weight parameter input from the weight parameter input unit and each weight parameter learned by the weight parameter learning unit;
The recommended content extraction means includes
The content recommendation system according to appendix 6, wherein recommended content is extracted using each of the recommendation methods based on a predetermined weight parameter synthesized by the weight parameter synthesis means.

(Appendix 8)
The weight parameter synthesis means includes:
The content recommendation system according to appendix 7, wherein, for each weight parameter learned by the weight parameter learning means, the weight parameter and the weight parameter input from the weight parameter input means are added together.

(Appendix 9)
The recommended content extraction means includes
The content recommendation system according to appendix 7 or appendix 8, wherein a weight parameter corresponding to the recommendation request from the user terminal is acquired from the weight parameter combining unit.

(Appendix 10)
A content recommendation device that receives a predetermined request from a user terminal and returns the result,
A weight parameter input means for inputting a weight parameter indicating the degree of influence of each recommendation method;
For each effect index indicating a predetermined recommended effect of the recommendation method, an effect index value calculation unit that calculates an effect index value for each recommendation method using history data including recommendation to the user by the recommendation method and the result thereof When,
For each effect index, an effect index combining means for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A content recommendation apparatus comprising: a display unit that displays, for each effect index, the effect index value of each recommendation method and the estimated value of the entire effect index.

(Appendix 11)
The effect index synthesis means includes
For each of the recommendation methods, performing a predetermined weighting on the effect index value based on the weight parameter, and adding up the weighted effect index values to calculate an estimated value of the entire effect index The content recommendation device according to Supplementary Note 10, which is a feature.

(Appendix 12)
The effect index is
12. The content recommendation device according to appendix 10 or appendix 11, wherein the content recommendation device is a feedback rate calculated from the number of times recommended by the recommendation method and the number of times the content is used through recommendation of the recommendation method.

(Appendix 13)
The effect index is
12. The content recommendation device according to appendix 10 or appendix 11, wherein the content recommendation device is a profit rate calculated from the number of times of recommendation by the recommendation method and the amount of income generated through the recommendation of the recommendation method.

(Appendix 14)
Including history condition filtering means for filtering the history data according to a predetermined condition set in advance,
The effect index calculating means is
14. The content recommendation device according to any one of supplementary note 10 to supplementary note 13, wherein the effect index value is calculated using the filtered history data.

(Appendix 15)
Any one of Supplementary Note 10 to Supplementary Note 14, including recommended content extracting means for extracting recommended content recommended using each recommendation method based on the weight parameter input from the weight parameter input means. The content recommendation device described in 1.

(Appendix 16)
Weight parameter learning means for learning the importance of each recommendation method as a weight parameter for each predetermined condition from the history data;
A weight parameter combining unit that combines the weight parameter input from the weight parameter input unit and each weight parameter learned by the weight parameter learning unit;
The recommended content extraction means includes
16. The content recommendation device according to appendix 15, wherein recommended content to be recommended is extracted using each recommendation method based on a predetermined weight parameter synthesized by the weight parameter synthesis means.

(Appendix 17)
The weight parameter synthesis means includes:
The content recommendation device according to appendix 16, wherein, for each weight parameter learned by the weight parameter learning unit, the weight parameter and the weight parameter input from the weight parameter input unit are added together.

(Appendix 18)
The recommended content extraction means includes
18. The content recommendation device according to appendix 16 or appendix 17, wherein a weight parameter corresponding to the recommendation request from the user terminal is acquired from the weight parameter combining unit.

(Appendix 19)
A recommendation method control method for a content recommendation device that receives a predetermined request from a user terminal and returns the result,
A weight parameter input step for inputting a weight parameter indicating the degree of influence of each recommendation method;
Effect index value calculating step for calculating an effect index value for each recommendation method using history data including recommendation to the user by the recommendation method and the result thereof for each effect index indicating a predetermined recommendation effect of the recommendation method When,
For each effect index, an effect index combining step for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A recommendation method control method, comprising: a display step for displaying, for each effect indicator, the effect indicator value of each recommendation method and the estimated value of the entire effect indicator.

(Appendix 20)
In the effect index synthesis step,
For each of the recommendation methods, performing a predetermined weighting on the effect index value based on the weight parameter, and adding up the weighted effect index values to calculate an estimated value of the entire effect index The recommendation method control method according to Supplementary Note 19, which is a feature.

(Appendix 21)
The effect index is
The recommendation method control method according to appendix 19 or appendix 20, characterized in that the feedback rate is calculated from the number of times of recommendation by the recommendation method and the number of times content is used through recommendation of the recommendation method.

(Appendix 22)
The effect index is
The recommendation method control method according to appendix 19 or appendix 20, wherein the recommendation rate is a profit rate calculated from the number of recommendations by the recommendation scheme and the amount of income generated through the recommendation scheme recommendation.

(Appendix 23)
A history condition filtering step of filtering the history data according to a predetermined condition set in advance;
In the effect index calculating step,
The recommendation method control method according to any one of appendix 19 to appendix 22, wherein the effect index value is calculated using the filtered history data.

(Appendix 24)
Any one of appendix 19 to appendix 23, further comprising a recommended content extraction step for extracting recommended content to be recommended using each recommendation method based on the weight parameter input from the weight parameter input step. Recommendation method control method described in 1.

(Appendix 25)
From the history data, for each predetermined condition, a weight parameter learning step for learning the importance of each recommendation method as a weight parameter;
A weight parameter combining step of combining the weight parameter input from the weight parameter input step and each weight parameter learned by the weight parameter learning step;
In the recommended content extraction step,
25. The recommendation method control method according to appendix 24, wherein recommended content recommended by each recommendation method is extracted based on the predetermined weight parameter synthesized by the weight parameter synthesis step.

(Appendix 26)
In the weight parameter synthesis step,
The recommendation method control method according to appendix 25, wherein, for each weight parameter learned in the weight parameter learning step, the weight parameter and the weight parameter input from the weight parameter input step are added together.

(Appendix 27)
In the recommended content extraction step,
27. The recommendation method control method according to appendix 25 or appendix 26, wherein a weight parameter corresponding to the recommendation request from the user terminal is acquired from the weight parameter synthesis step.

(Appendix 28)
A recommendation method control program that operates on a computer that functions as a content recommendation device that receives a predetermined request from a user terminal and returns the result,
A weight parameter input process for inputting a weight parameter indicating the degree of influence of each recommendation method to the computer;
An effect index value calculation process for calculating an effect index value for each recommendation method using history data including a recommendation to the user by the recommendation method and a result thereof for each recommendation index indicating a predetermined recommendation effect of the recommendation method When,
For each effect index, an effect index combining process for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A recommendation method control program for executing, for each effect indicator, display processing for displaying the effect indicator value of each recommendation method and the estimated value of the whole effect indicator.

(Appendix 29)
In the effect index synthesis process,
For each of the recommendation methods, performing a predetermined weighting on the effect index value based on the weight parameter, and adding up the weighted effect index values to calculate an estimated value of the entire effect index The recommendation method control program according to appendix 28, which is characterized by the following.

(Appendix 30)
The effect index is
The recommendation method control program according to appendix 28 or appendix 29, which is a feedback rate calculated from the number of times recommended by the recommendation method and the number of times the content has been used through recommendation of the recommendation method.

(Appendix 31)
The effect index is
30. The recommendation method control program according to appendix 28 or appendix 29, wherein the program is a profit rate calculated from the number of recommendations by the recommendation method and the amount of income generated through the recommendation by the recommendation method.

(Appendix 32)
In the computer,
A history condition filtering process for filtering the history data according to a predetermined condition set in advance;
In the effect index calculation process,
32. The recommendation method control program according to any one of appendix 28 to appendix 31, wherein the effect index value is calculated using the filtered history data.

(Appendix 33)
In the computer,
Any one of appendix 28 to appendix 32, wherein a recommended content extraction process for extracting recommended content using each recommendation method is executed based on the weight parameter input from the weight parameter input process. The recommendation method control program according to the item.

(Appendix 34)
In the computer,
From the history data, for each predetermined condition, a weight parameter learning process for learning the importance of each recommendation method as a weight parameter;
Causing the weight parameter input process from the weight parameter input process and a weight parameter synthesis process to synthesize each weight parameter learned by the weight parameter learning process;
In the recommended content extraction process,
34. The recommendation method control program according to appendix 33, wherein recommended content to be recommended is extracted using each recommendation method based on a predetermined weight parameter synthesized by the weight parameter synthesis process.

(Appendix 35)
In the weight parameter synthesis process,
The recommendation method control program according to attachment 34, wherein, for each weight parameter learned by the weight parameter learning process, the weight parameter and the weight parameter input from the weight parameter input process are added together.

(Appendix 36)
In the recommended content extraction process,
36. The recommendation method control program according to appendix 34 or appendix 35, wherein a weight parameter corresponding to the recommendation request from the user terminal is acquired from the weight parameter synthesis process.

The present invention can be applied to uses such as marketing analysis for analyzing a propagation pattern of a certain content.
It can also be applied to uses such as information recommendation using the correlation of propagation patterns of certain contents.

DESCRIPTION OF SYMBOLS 100: Content recommendation apparatus 101: Weight parameter input means 102: Effect index synthesis means 103: Display means 104: Filter condition storage means 105: History condition filter means 106: Effect index value calculation means 107: Recommended content extraction means 108: Weight parameter Combining means 109: Effect index storage means 110: Weight parameter storage means 111: Transmission / reception means 200: User terminal 1000: Content recommendation system

Claims (10)

A content recommendation system comprising a user terminal and a content recommendation device that recommends content using a plurality of recommendation methods,
A weight parameter input means for inputting a weight parameter indicating the degree of influence of each recommendation method;
For each effect index indicating a predetermined recommended effect of the recommendation method, an effect index value calculation unit that calculates an effect index value for each recommendation method using history data including recommendation to the user by the recommendation method and the result thereof When,
For each effect index, an effect index combining means for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A content recommendation system comprising: a display unit that displays the effect index value of each recommendation method and the estimated value of the entire effect index for each effect index. The effect index is
The content recommendation system according to claim 1, wherein the content recommendation system is a feedback rate calculated from the number of times of recommendation by the recommendation method and the number of times content is used through the recommendation of the recommendation method. The effect index is
The content recommendation system according to claim 1, wherein the content recommendation system is a profit rate calculated from the number of recommendations by the recommendation method and the amount of income generated through the recommendation by the recommendation method. Including history condition filtering means for filtering the history data according to a predetermined condition set in advance,
The effect index calculating means is
The content recommendation system according to any one of claims 1 to 3, wherein the effect index value is calculated using the filtered history data.   5. The recommended content extracting means for extracting recommended content to be recommended using each recommendation method based on the weight parameter input from the weight parameter input means. The content recommendation system according to item 1. Weight parameter learning means for learning the importance of each recommendation method as a weight parameter for each predetermined condition from the history data;
A weight parameter combining unit that combines the weight parameter input from the weight parameter input unit and each weight parameter learned by the weight parameter learning unit;
The recommended content extraction means includes
6. The content recommendation system according to claim 5, wherein recommended content to be recommended is extracted using each recommendation method based on a predetermined weight parameter synthesized by the weight parameter synthesis means. The recommended content extraction means includes
The content recommendation system according to claim 6, wherein a weight parameter corresponding to the recommendation request from the user terminal is acquired from the weight parameter synthesis unit. A content recommendation device that receives a predetermined request from a user terminal and returns the result,
A weight parameter input means for inputting a weight parameter indicating the degree of influence of each recommendation method;
For each effect index indicating a predetermined recommended effect of the recommendation method, an effect index value calculation unit that calculates an effect index value for each recommendation method using history data including recommendation to the user by the recommendation method and the result thereof When,
For each effect index, an effect index combining means for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A content recommendation apparatus comprising: a display unit that displays, for each effect index, the effect index value of each recommendation method and the estimated value of the entire effect index. A recommendation method control method for a content recommendation device that receives a predetermined request from a user terminal and returns the result,
A weight parameter input step for inputting a weight parameter indicating the degree of influence of each recommendation method;
Effect index value calculating step for calculating an effect index value for each recommendation method using history data including recommendation to the user by the recommendation method and the result thereof for each effect index indicating a predetermined recommendation effect of the recommendation method When,
For each effect index, an effect index combining step for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A recommendation method control method, comprising: a display step for displaying, for each effect indicator, the effect indicator value of each recommendation method and the estimated value of the entire effect indicator. A recommendation method control program that operates on a computer that functions as a content recommendation device that receives a predetermined request from a user terminal and returns the result,
In the computer,
A weight parameter input process for inputting a weight parameter indicating the degree of influence of each recommendation method;
An effect index value calculation process for calculating an effect index value for each recommendation method using history data including a recommendation to the user by the recommendation method and a result thereof for each recommendation index indicating a predetermined recommendation effect of the recommendation method When,
For each effect index, an effect index combining process for calculating a value obtained by combining the weight parameter of each recommendation method and the effect index value as an estimated value of the entire effect index;
A recommendation method control program for executing, for each effect indicator, display processing for displaying the effect indicator value of each recommendation method and the estimated value of the whole effect indicator.

Images