CN102737095A

CN102737095A - Recommendation device, recommendation system, recommendation method, and program

Info

Publication number: CN102737095A
Application number: CN2012100837375A
Authority: CN
Inventors: 佐佐木祥; 柳原广昌; 土生由希子
Original assignee: KDDI Corp
Current assignee: KDDI Corp
Priority date: 2011-03-29
Filing date: 2012-03-27
Publication date: 2012-10-17
Anticipated expiration: 2032-03-27
Also published as: WO2012132427A1; CN102737095B; JP2012216175A; JP5806081B2

Abstract

The invention aims to implement a highly precise recommendation without limiting used history information and with reduced computation load required for dissection. A batch processing unit pre-dissects history information and generates information necessary for a recommendation. A realtime processing unit dissects the history information for each occurrence of a recommendation request and generates information necessary for the recommendation. A recommendation result output unit integrates the output of the batch processing unit and/or the output of the realtime processing unit and outputs a recommendation result.

Description

Recommendation apparatus, commending system, recommend method and program

Technical field

The recommendation results that the present invention relates to carry out batch processing and handle also comprehensive two processing is in real time exported recommendation apparatus, commending system, recommend method and the program of final recommendation results.

Background technology

Wait the hobby of inferring the user and recommend to be predicted to be " recommendation service " of the content of having catered to user's hobby etc. automatically in vogue in recent years based on user's historical information.The representative technology of in this recommendation service, using is the technology that is called as " coordinating to filter ".This coordination filtration is to be predicted as the method for user's interest content as recommendation results through defining correlativity and export in the historical information of relatively buying users such as historical or reading history between each user or between each content based on this correlativity.Known have a technology (for example, with reference to non-patent literature 1, non-patent literature 2, non-patent literature 3, non-patent literature 4) of having improved robotization that this coordination filters or high speed etc.

Technical literature formerly

Non-patent literature

Non-patent literature 1:P.Resnick; N.Iacovou; M.Suchak; P.Bergstrom, and J.Riedl. " GroupLens:Open Architecture for Collaborative Filtering of Netnews " In conference on Computer Supported Cooperative Work, pp.175-186 (1994).

Non-patent literature 2:T.Hofmann and J.Puzicha, " Latent Class Models for Collaborative Filtering ", 16 ^ThIJCAI (1999).

Non-Patent Document 3: Polyard letter Sakae Ito Code, "ソ a sleeve boots Shin-ya Hikaru ku ni ku ma an information base づ ku Development see," DEWS? 2008, I1-15, Mar.2008.

Non-patent literature 4:Badrul Sarwar; George Karypis; Joseph Konstan, and John Reidl.2001.Item-based collaborative filtering recommendation algorithms.In Proceedings of the 10 ^ThInternational conference on world Wide Web (www ' 01) .ACM, New York, NY, USA, 285-295.

Summary of the invention

The problem that invention will solve

Here, in the technology shown in non-patent literature 1 and the non-patent literature 4, to each recommendation request; Analysis of history information generates recommendation results; Therefore always can use the recommendation of up-to-date historical information, but in this technology, if historical information is huge more; Analyzing required calculated amount will increase, and therefore exists to be difficult to realize the problem of recommendation in real time.

In non-patent literature 2 described technology; The ex ante analysis historical information defines correlativity, and therefore with non-patent literature 1 described compared with techniques, the calculated amount that when recommendation request is arranged, will analyze is little; Realize easily recommending in real time; But in this technology,, therefore there is the problem of not using up-to-date historical information owing to only use the historical information of resolving time point to define correlativity.

In non-patent literature 3 described technology, the object of analysis of history information is wanted in selection in advance, and only between representative of consumer and recommended object user, defines correlativity; Therefore with non-patent literature 1 described compared with techniques; The calculated amount that when recommendation request is arranged, will analyze is little, realizes easily recommending in real time, but in non-patent literature 3 described technology; Limited the historical information of utilizing, therefore existed and reduced the problem of recommending precision.

Therefore, the present invention makes in view of the above problems, and its purpose is to provide a kind of and does not limit the historical information of utilization and reduce recommendation apparatus, commending system, recommend method and the program that required calculated amount realizes high-precision recommendation of analyzing.

The means that are used to deal with problems

The present invention has proposed the following in order to solve the above problems.For easy understanding, mark the Reference numeral corresponding and describe, but be not limited thereto with embodiment of the present invention.

(1) the present invention proposes a kind of recommendation apparatus; Its use based on the historical information of buying users such as historical or reading history define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results; Said recommendation apparatus is characterised in that; Comprise: batch processing unit (for example, being equivalent to the batch processing portion 1100 of Fig. 1), the said historical information of its ex ante analysis generate the required information of recommending; Real-time processing unit (for example, being equivalent to the real-time handling part 1200 of Fig. 1), it is analyzed historical information and generates the required information of recommending when recommendation request takes place; And recommendation results output unit (for example, being equivalent to the recommendation results efferent 1300 of Fig. 1), the output of its comprehensive said batch processing unit is or/and recommendation results is exported in the output of said real-time processing unit.

According to this invention, batch processing unit ex ante analysis historical information generates the required information of recommending.Processing unit is analyzed historical information and is generated the required information of recommending when recommendation request takes place in real time.The output of the comprehensive batch processing of recommendation results output unit unit is or/and recommendation results is exported in the output of processing unit in real time.Thereby, can not show that the historical information of utilization and the required calculated amount of minimizing analysis realize high-precision recommendation.

(2) the present invention is about the recommendation apparatus of (1); Proposed to have the recommendation apparatus of following characteristic: said batch processing unit comprises: the first historical information collector unit (for example; The historical information collection portion 1101 that is equivalent to Fig. 2), the historical information that its collection analysis is required; The first correlation calculations unit (for example, being equivalent to the correlation calculations portion 1102 of Fig. 2), its historical information of analyzing said collection defines correlativity; The first recommendation degree computing unit (for example, being equivalent to the recommendation degree calculating part 1103 of Fig. 2), it uses the correlativity of said definition to calculate the recommendation degree to each content of user; And storage unit (for example; The storage part 1104 that is equivalent to Fig. 2), it stores said recommendation degree of calculating, and said real-time processing unit comprises: the second historical information collector unit (for example; The historical information collection portion 1201 that is equivalent to Fig. 2), the required historical information of collection analysis in the time of in fact; The second correlation calculations unit (for example, being equivalent to the correlation calculations portion 1202 of Fig. 2), its historical information of analyzing said collection defines correlativity; And second recommendation degree computing unit (for example; The recommendation degree calculating part 1203 that is equivalent to Fig. 2); It uses the said correlativity of calculating to calculate the recommendation degree to each content of user; Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of said storage with by the said second recommendation degree computing unit of said recommendation results output unit is exported recommendation results.

According to this invention, the required historical information of the first historical information collector unit collection analysis of batch processing device.The collected historical information of the first correlation calculations element analysis defines correlativity.The first recommendation degree computing unit uses the correlativity of calculating to calculate the recommendation degree to each content of user.The recommendation degree that memory device stores is calculated.The second historical information collector unit real-time collecting of treating apparatus is analyzed required historical information in real time.The collected historical information of the second correlation calculations device analysis defines correlativity.The second recommendation degree computing unit uses the correlativity of being calculated to calculate the recommendation degree to each content of user.The recommendation degree of each content that the recommendation results output unit is calculated based on the recommendation degree of each content of being stored with through the second recommendation degree computing unit is exported recommendation results.Thereby; Export recommendation results based on recommendation degree that has used the correlativity of in batch processing, calculating and the recommendation degree that has used the correlativity of in real-time the processing, calculating, the historical information that therefore can not limit utilization also reduces the required calculated amount of analysis and realizes high-precision recommendation.

(3) the present invention is about the recommendation apparatus of (1); Proposed to have the recommendation apparatus of following characteristic: said batch processing unit comprises: the first historical information collector unit (for example; The historical information collection portion 1101 that is equivalent to Fig. 5), the historical information that its collection analysis is required; The first correlation calculations unit (for example, being equivalent to the correlation calculations portion 1102 of Fig. 5), its historical information of analyzing said collection defines correlativity; And storage unit (for example; The storage part 1115 that is equivalent to Fig. 5), it stores the said correlativity of calculating, and said real-time processing unit comprises: the second historical information collector unit (for example; The historical information collection portion 1201 that is equivalent to Fig. 5), the required historical information of collection analysis in the time of in fact; The second correlation calculations unit (for example, being equivalent to the correlation calculations portion 1202 of Fig. 5), its historical information of analyzing said collection defines correlativity; The first recommendation degree computing unit (for example, being equivalent to the recommendation degree calculating part 1203 of Fig. 5), it uses the said correlativity of calculating to calculate the recommendation degree to each content of user; Acquiring unit (for example, being equivalent to the portion that obtains 1214 of Fig. 5), it obtains the correlativity of said storage; And second recommendation degree computing unit (for example; The recommendation degree calculating part 1215 that is equivalent to Fig. 5); It uses the said correlativity of obtaining to calculate the recommendation degree to each content of user; Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of being calculated by the said first recommendation degree computing unit with by the said second recommendation degree computing unit of said recommendation results output unit (the recommendation results efferent 1310 that for example, is equivalent to Fig. 5) is exported recommendation results.

According to this invention, the required historical information of the first historical information collector unit collection analysis of batch processing unit.The collected historical information of the first correlation calculations element analysis defines correlativity.The correlativity that cell stores is calculated.The second historical information collector unit real-time collecting of processing unit is analyzed required historical information in real time.The collected historical information of the second correlation calculations element analysis defines correlativity.The first recommendation degree computing unit uses the correlativity of being calculated to calculate the recommendation degree to each content of user.Acquiring unit obtains the correlativity of being stored.The second recommendation degree computing unit uses the correlativity of being obtained to calculate the recommendation degree to each content of user.The recommendation degree of each content that the recommendation results output unit is calculated based on the recommendation degree of each content of calculating through the first recommendation degree computing unit with through the second recommendation degree computing unit is exported recommendation results.Thereby; Real-time processing unit is based on the correlativity of calculating in the batch processing and the correlativity of in handling in real time, calculating come the calculated recommendation degree; And export recommendation results based on these recommendation degree, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

(4) the present invention is about the recommendation apparatus of (1); Proposed to have the recommendation apparatus of following characteristic: said batch processing unit comprises: the first historical information collector unit (for example; The historical information collection portion 1101 that is equivalent to Fig. 8), the historical information that its collection analysis is required; The first correlation calculations unit (for example, being equivalent to the correlation calculations portion 1102 of Fig. 8), its historical information of analyzing said collection defines correlativity; The first recommendation degree computing unit (for example, being equivalent to the recommendation degree calculating part 1103 of Fig. 8), it uses the said correlativity of calculating to calculate the recommendation degree to each content of user; And storage unit (for example; The storage part 1126 that is equivalent to Fig. 8); It stores the correlativity of said definition and the recommendation degree of calculating; Said real-time processing unit comprises: the second historical information collector unit (for example, being equivalent to the historical information collection portion 1201 of Fig. 8), the required historical information of collection analysis in the time of in fact; Acquiring unit (for example, being equivalent to the portion that obtains 1227 of Fig. 8), it obtains the correlativity of said definition and the recommendation degree of calculating; And second recommendation degree computing unit (for example; The recommendation degree calculating part 1226 that is equivalent to Fig. 8); It uses the said correlativity of obtaining to calculate the recommendation degree to each content of user; Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of being calculated by the said first recommendation degree computing unit with by the said second recommendation degree computing unit of said recommendation results output unit (the recommendation results efferent 1320 that for example, is equivalent to Fig. 8) is exported recommendation results.

According to this invention, the required historical information of the first historical information collector unit collection analysis of batch processing unit.The collected historical information of the first correlation calculations element analysis defines correlativity.The first recommendation degree computing unit uses the correlativity of being calculated to calculate the recommendation degree to each content of user.Defined correlativity of cell stores and the recommendation degree of calculating.The second historical information collector unit real-time collecting of processing unit is analyzed required historical information in real time.Acquiring unit obtains defined correlativity and the recommendation degree of calculating.The second recommendation degree computing unit uses the correlativity of being obtained to calculate the recommendation degree to each content of user.The recommendation degree of each content that the recommendation results output unit is calculated based on the recommendation degree of each content of calculating through the first recommendation degree computing unit with through the second recommendation degree computing unit is exported recommendation results.Thereby; Use the correlativity of calculating in the batch processing based on the recommendation degree that has used the correlativity of in batch processing, calculating with in handling in real time and the recommendation degree that provides exported recommendation results, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

(5) the present invention proposes a kind of commending system; It comprises batch processing device, real-time treating apparatus and recommendation apparatus; And use based on the historical information of buying users such as historical or reading history define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results; Said commending system is characterised in that; Said batch processing device (for example; The batch processing device 400 that is equivalent to Figure 10) the said historical information of ex ante analysis generates the required information of recommending, and said real-time treating apparatus (the real-time treating apparatus 500 that for example, is equivalent to Figure 10) is analyzed historical information and generated the required information of recommending when recommendation request takes place; The output of the comprehensive said batch processing device of said recommendation apparatus (the recommendation results output unit 600 that for example, is equivalent to Figure 10) is or/and recommendation results is exported in the output of said real-time treating apparatus.

According to this invention; Batch processing device ex ante analysis historical information generates the required information of recommending; Treating apparatus is analyzed historical information and is generated the required information of recommending when recommendation request takes place in real time, and the output of the comprehensive batch processing device of recommendation apparatus is or/and recommendation results is exported in the output of treating apparatus in real time.Thereby the historical information and the required calculated amount of minimizing analysis that can not limit utilization realize high-precision recommendation.

(6) the present invention is about the commending system of (5); Proposed to have the commending system of following characteristic: said batch processing device comprises: the first historical information collector unit (for example; The historical information collection portion 401 that is equivalent to Figure 11), the historical information that its collection analysis is required; The first correlation calculations unit (for example, being equivalent to the correlation calculations portion 402 of Figure 11), its historical information of analyzing said collection defines correlativity; The first recommendation degree computing unit (for example, being equivalent to the recommendation degree calculating part 403 of Figure 11), it uses the said correlativity of calculating to calculate the recommendation degree to each content of user; And storage unit (for example; The storage part 404 that is equivalent to Figure 11), it stores said recommendation degree of calculating, and said real-time treating apparatus comprises: the second historical information collector unit (for example; The historical information collection portion 501 that is equivalent to Figure 11), the required historical information of collection analysis in the time of in fact; The second correlation calculations unit (for example, being equivalent to the correlation calculations portion 502 of Figure 11), its historical information of analyzing said collection defines correlativity; And second recommendation degree computing unit (for example; The recommendation degree calculating part 503 that is equivalent to Figure 11); It uses the said correlativity of calculating to calculate the recommendation degree to each content of user; Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of said storage with by the said second recommendation degree computing unit of said recommendation apparatus is exported recommendation results.

According to this invention, the required historical information of the first historical information collector unit collection analysis of batch processing device.The collected historical information of the first correlation calculations element analysis defines correlativity.The first recommendation degree computing unit uses the correlativity of being calculated to calculate the recommendation degree to each content of user.The recommendation degree that cell stores is calculated.The second historical information collector unit real-time collecting of treating apparatus is analyzed required historical information in real time.The collected historical information of the second correlation calculations device analysis defines correlativity.The second recommendation degree computing unit uses the correlativity of being calculated to calculate the recommendation degree to each content of user.The recommendation degree of each content that recommendation apparatus is calculated based on the recommendation degree of each content of being stored with through the second recommendation degree computing unit is exported recommendation results.Thereby; Export recommendation results based on recommendation degree that has used the correlativity of in batch processing, calculating and the recommendation degree that has used the correlativity of in real-time treating apparatus, calculating, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

(7) the present invention is about the commending system of (5); Proposed to have the commending system of following characteristic: said batch processing device comprises: the first historical information collector unit (for example; The historical information collection portion 401 that is equivalent to Figure 14), the historical information that its collection analysis is required; The first correlation calculations unit (for example, being equivalent to the correlation calculations portion 402 of Figure 14), its historical information of analyzing said collection defines correlativity; And storage unit (for example; The storage part 415 that is equivalent to Figure 14), it stores the said correlativity of calculating, and said real-time treating apparatus comprises: the second historical information collector unit (for example; The historical information collection portion 501 that is equivalent to Figure 14), the required historical information of collection analysis in the time of in fact; The second correlation calculations unit (for example, being equivalent to the correlation calculations portion 502 of Figure 14), its historical information of analyzing said collection defines correlativity; The first recommendation degree computing unit (for example, being equivalent to the recommendation degree calculating part 503 of Figure 14), it uses the said correlativity of calculating to calculate the recommendation degree to each content of user; Acquiring unit (for example, being equivalent to the portion that obtains 514 of Figure 14), it obtains the correlativity of said storage; And second recommendation degree computing unit (for example; The recommendation degree calculating part 515 that is equivalent to Figure 14); It uses the said correlativity of obtaining to calculate the recommendation degree to each content of user; Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of being calculated by the said first recommendation degree computing unit with by the said second recommendation degree computing unit of said recommendation apparatus is exported recommendation results.

According to this invention, the required historical information of the first historical information collector unit collection analysis of batch processing device.The collected historical information of the first correlation calculations element analysis defines correlativity.The correlativity that memory device stores is calculated.The second historical information collector unit real-time collecting of treating apparatus is analyzed required historical information in real time.The collected historical information of the second correlation calculations device analysis defines correlativity.The first recommendation degree computing unit uses the correlativity of being calculated to calculate the recommendation degree to each content of user.Acquiring unit obtains the correlativity of being stored.The second recommendation degree computing unit uses the correlativity of being obtained to calculate the recommendation degree to each content of user.The recommendation degree of each content that recommendation apparatus is calculated based on the recommendation degree of each content of calculating through the first recommendation degree computing unit with through the second recommendation degree computing unit is exported recommendation results.Thereby; Real-time treating apparatus is based on the correlativity of calculating in the batch processing and the correlativity of in handling in real time, calculating come the calculated recommendation degree; And export recommendation results based on these recommendation degree, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

(8) the present invention is about the commending system of (5); Proposed to have the commending system of following characteristic: said batch processing device comprises: the first historical information collector unit (for example; The historical information collection portion 401 that is equivalent to Figure 17), the historical information that its collection analysis is required; The first correlation calculations unit (for example, being equivalent to the correlation calculations portion 402 of Figure 17), its historical information of analyzing said collection defines correlativity; The first recommendation degree computing unit (for example, being equivalent to the recommendation degree calculating part 403 of Figure 17), it uses the said correlativity of calculating to calculate the recommendation degree to each content of user; And storage unit (for example, being equivalent to the storage part 426 of Figure 17), it stores the correlativity of said definition and the recommendation degree of calculating; Said real-time treating apparatus comprises: the second historical information collector unit (for example, being equivalent to the historical information collection portion 501 of Figure 17), the required historical information of collection analysis in the time of in fact; Acquiring unit (for example, being equivalent to the portion that obtains 527 of Figure 17), it obtains the correlativity of said definition and the recommendation degree of calculating; And second recommendation degree computing unit (for example; The recommendation degree calculating part 526 that is equivalent to Figure 17); It uses the said correlativity of obtaining to calculate the recommendation degree to each content of user; Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of being calculated by the said first recommendation degree computing unit with by the said second recommendation degree computing unit of said recommendation apparatus is exported recommendation results.

According to this invention, the required historical information of the first historical information collector unit collection analysis of batch processing device.The collected historical information of the first correlation calculations element analysis defines correlativity.The first recommendation degree computing unit uses the correlativity of being calculated to calculate the recommendation degree to each content of user.Defined correlativity of cell stores and the recommendation degree of calculating.The second historical information collector unit real-time collecting of treating apparatus is analyzed required historical information in real time.Acquiring unit obtains defined correlativity and the recommendation degree of calculating.The second recommendation degree computing unit uses the correlativity of being obtained to calculate the recommendation degree to each content of user.The recommendation degree of each content that recommendation apparatus is calculated based on the recommendation degree of each content of calculating through the first recommendation degree computing unit with through the second recommendation degree computing unit is exported recommendation results.Thereby; Use the correlativity of in batch processing, calculating based on the recommendation degree that has used the correlativity of in batch processing, calculating with in handling in real time and the recommendation degree that provides exported recommendation results, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

(9) the present invention proposes a kind of recommend method; Its use based on the historical information of buying users such as historical or reading history define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results; Said recommend method is characterised in that; Comprise: the batch processing step is used for the said historical information of ex ante analysis and generates the required information of recommending; Treatment step is used for when recommendation request takes place, analyzing historical information and generates the required information of recommending in real time; And recommendation results output step, be used for the output of comprehensive said batch processing step or/and recommendation results is exported in the output of said real-time treatment step.

According to this invention; The ex ante analysis historical information generates the required information of recommending; When recommendation request takes place, analyze historical information and generate the required information of recommending, and the output of comprehensive batch processing step is or/and recommendation results is exported in the output of treatment step in real time.Thereby the historical information and the required calculated amount of minimizing analysis that can not limit utilization realize high-precision recommendation.

(10) the present invention has proposed to have the recommend method of following characteristic about the recommend method of (9): said batch processing step comprises: the first step of the historical information that collection analysis is required (the step S101 that for example, is equivalent to Fig. 3); The historical information of analyzing said collection defines second step of correlativity (the step S102 that for example, is equivalent to Fig. 3); Use the said correlativity of calculating to calculate third step (the step S103 that for example, is equivalent to Fig. 3) to the recommendation degree of each content of user; And the 4th step (for example, being equivalent to the step S104 of Fig. 3) of storing said recommendation degree of calculating, said real-time treatment step comprises: real-time collecting is analyzed the 5th step (the step S105 that for example, is equivalent to Fig. 3) of required historical information; The historical information of analyzing said collection defines the 6th step of correlativity (the step S106 that for example, is equivalent to Fig. 3); And use the said correlativity of calculating (for example to calculate to the 7th step of the recommendation degree of each content of user; The step S107 that is equivalent to Fig. 3); Wherein, said recommendation results output step is based on the recommendation degree of each content of storing in said the 4th step and the recommendation degree of each content of calculating through said the 7th step is exported recommendation results.

According to this invention, the historical information that collection analysis is required, and analyze collected historical information and define correlativity.Then, use the correlativity of being calculated to calculate recommendation degree, and store the recommendation degree of being calculated each content of user.And real-time collecting is analyzed required historical information, and analyzes collected historical information and define correlativity.Then, use the correlativity of being calculated to calculate recommendation degree to each content of user.And, be based on the recommendation degree of each content of storing in the 4th step and the recommendation degree of each content of calculating through the 7th step comes the calculated recommendation result.Thereby; Export recommendation results based on recommendation degree that has used the correlativity of in batch processing, calculating and the recommendation degree that has used the correlativity of in real-time the processing, calculating, the historical information that therefore can not limit utilization also reduces the required calculated amount of analysis and realizes high-precision recommendation.

(11) the present invention has proposed to have the recommend method of following characteristic about the recommend method of (9): said batch processing step comprises: the first step of the historical information that collection analysis is required (the step S201 that for example, is equivalent to Fig. 6); The historical information of analyzing said collection defines second step of correlativity (the step S202 that for example, is equivalent to Fig. 6); And the third step (for example, being equivalent to the step S203 of Fig. 6) of storing the said correlativity of calculating, said real-time treatment step comprises: real-time collecting is analyzed the 4th step (the step S204 that for example, is equivalent to Fig. 6) of required historical information; The historical information of analyzing said collection defines the 5th step of correlativity (the step S205 that for example, is equivalent to Fig. 6); Use the said correlativity of calculating to calculate the 6th step (the step S206 that for example, is equivalent to Fig. 6) to the recommendation degree of each content of user; Obtain the 7th step (the step S207 that for example, is equivalent to Fig. 6) of the correlativity of said storage; And use the said correlativity of obtaining (for example to calculate to the 8th step of the recommendation degree of each content of user; The step S208 that is equivalent to Fig. 6); Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through said the 6th step with through said the 8th step of said recommendation results output step is exported recommendation results.

According to this invention, the historical information that collection analysis is required is analyzed collected historical information and is defined correlativity, and stores the correlativity of being calculated.And real-time collecting is analyzed required historical information, analyzes collected historical information and defines correlativity, uses the correlativity of being calculated to calculate the recommendation degree to each content of user.Then, obtain the correlativity of being stored, use the correlativity of being obtained to calculate recommendation degree each content of user.And the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through the 6th step with through the 8th step is exported recommendation results.Thereby; The correlativity that is based on the correlativity calculated in the batch processing and in handling in real time, calculates is come the calculated recommendation degree; And export recommendation results based on these recommendation degree, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

(12) the present invention has proposed to have the recommend method of following characteristic about the recommend method of (9): said batch processing step comprises: the first step of the historical information that collection analysis is required (the step S301 that for example, is equivalent to Fig. 9); The historical information of analyzing said collection defines second step of correlativity (the step S302 that for example, is equivalent to Fig. 9); And use the said correlativity of calculating to calculate third step (the step S303 that for example, is equivalent to Fig. 9) to the said correlativity of calculating of storage of the recommendation degree of each content of user; And the 4th step of correlativity of storing said definition and the recommendation degree of calculating (for example; The step S304 that is equivalent to Fig. 9); Said real-time treatment step comprises: real-time collecting is analyzed the four or five step (the step S305 that for example, is equivalent to Fig. 9) (the step S304 that for example, is equivalent to Fig. 9) of required historical information; Obtain the correlativity of said definition and the 6th step (the step S306 that for example, is equivalent to Fig. 9) of the recommendation degree of calculating; And use the said correlativity of obtaining (for example to calculate to the 7th step of the recommendation degree of each content of user; The step S307 that is equivalent to Fig. 9); Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through said third step with through said the 7th step of said recommendation results output step is exported recommendation results.

According to this invention, the historical information that collection analysis is required is analyzed collected historical information and is defined correlativity, uses the correlativity of being calculated to calculate the recommendation degree to each content of user, and stores defined correlativity and the recommendation degree of calculating.And real-time collecting is analyzed required historical information, obtains defined correlativity and the recommendation degree of calculating, and uses the correlativity of being obtained to calculate the recommendation degree to each content of user.And the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through third step with through the 7th step is exported recommendation results.Thereby; Use the correlativity of in batch processing, calculating based on the recommendation degree that has used the correlativity of in batch processing, calculating with in handling in real time and the recommendation degree that provides exported recommendation results, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

(13) the present invention proposes a kind of program that is used to make computing machine execution recommend method; Wherein, Said recommend method use based on the historical information of buying users such as historical or reading history define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results; Said program is carried out computing machine: the batch processing step is used for the said historical information of ex ante analysis and generates the required information of recommending; Treatment step is used for the said historical information of ex ante analysis and generates the required information of recommending in real time; Treatment step is used for when recommendation request takes place, analyzing historical information and generates the required information of recommending in real time; And recommendation results output step, be used for the output of comprehensive said batch processing step or/and recommendation results is exported in the output of said real-time treatment step.

(14) the present invention proposes a kind of program that is used for making computing machine to carry out the recommend method of commending system; Wherein, Said commending system comprises batch processing device, real-time treating apparatus and recommendation apparatus; And use based on the historical information of buying users such as historical or reading history define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results, said program is carried out computing machine: said historical information generated the first step of recommending required information about said batch processing device divided in advance; Said real-time treating apparatus is analyzed historical information and is generated second step of recommending required information when recommendation request takes place; And the output of the comprehensive said first step of said recommendation apparatus is or/and the third step of recommendation results is exported in the output of said second step.

According to this invention; Batch processing device ex ante analysis historical information generates the required information of recommending; Treating apparatus is analyzed historical information and is generated the required information of recommending when recommendation request takes place in real time; And the output of the comprehensive first step of recommendation apparatus is or/and recommendation results is exported in the output of second step.Thereby the historical information and the required calculated amount of minimizing analysis that can not limit utilization realize high-precision recommendation.

The invention effect

According to the present invention, has following effect: can recommendation results that utilize whole historical informations be provided through batch processing portion, and can the recommendation results that reflect up-to-date historical information collected when request takes place be provided through real-time handling part.

Description of drawings

Fig. 1 is the pie graph that the recommendation apparatus that first embodiment of the invention relates to is shown.

Fig. 2 is the detailed pie graph that the recommendation apparatus that first embodiment of the invention relates to is shown.

Fig. 3 is the processing figure that the recommendation apparatus that first embodiment of the invention relates to is shown.

Fig. 4 is the pie graph that the recommendation apparatus that second embodiment of the invention relates to is shown.

Fig. 5 is the detailed pie graph that the recommendation apparatus that second embodiment of the invention relates to is shown.

Fig. 6 is the processing figure that the recommendation apparatus that second embodiment of the invention relates to is shown.

Fig. 7 is the pie graph that the recommendation apparatus that third embodiment of the invention relates to is shown.

Fig. 8 is the detailed pie graph that the recommendation apparatus that third embodiment of the invention relates to is shown.

Fig. 9 is the processing figure that the recommendation apparatus that third embodiment of the invention relates to is shown.

Figure 10 is the pie graph that the commending system that four embodiment of the invention relates to is shown.

Figure 11 is the detailed pie graph that the commending system that four embodiment of the invention relates to is shown.

Figure 12 is the processing figure that the commending system that four embodiment of the invention relates to is shown.

Figure 13 is the pie graph that the commending system that fifth embodiment of the invention relates to is shown.

Figure 14 is the detailed pie graph that the commending system that fifth embodiment of the invention relates to is shown.

Figure 15 is the processing figure that the commending system that fifth embodiment of the invention relates to is shown.

Figure 16 is the pie graph that the commending system that sixth embodiment of the invention relates to is shown.

Figure 17 is the detailed pie graph that the commending system that sixth embodiment of the invention relates to is shown.

Figure 18 is the processing figure that the commending system that sixth embodiment of the invention relates to is shown.

Figure 19 is the pie graph that the recommendation apparatus that the embodiment 1 of first embodiment of the invention relates to is shown.

Figure 20 is the pie graph that the recommendation apparatus that the embodiment 1 of first embodiment of the invention relates to is shown.

Figure 21 is user's the processing figure of the recommendation process of the recommendation apparatus that relates to of the schematically illustrated embodiment 1 that is used for first embodiment of the invention.

Figure 22 is the pie graph that the recommendation apparatus that the embodiment 2 of third embodiment of the invention relates to is shown.

Figure 23 is the processing figure that the recommendation apparatus that the embodiment 2 of third embodiment of the invention relates to is shown.

Embodiment

Below, with reference to accompanying drawing embodiment of the present invention is elaborated.

Inscape in this embodiment can be suitably replaced with existing inscape etc., and can comprise with being combined in of other existing inscapes in various changes.Thereby, be not that the record based on following embodiment limits the summary of the invention of putting down in writing in claims.

< first embodiment >

Use Fig. 1 and Fig. 3, first embodiment of the present invention is described.

< formation of recommendation apparatus >

Use Fig. 1, the formation of the recommendation apparatus that this embodiment is related to describes.

As shown in Figure 1, the recommendation apparatus 1000 that this embodiment relates to comprises batch processing portion 1100, real-time handling part 1200 and recommendation results efferent 1300, and batch processing portion 1100 is connected with database 200, and handling part 1200 is connected with database 300 in real time.

For example preserve 100 user's data in the database 200, for example preserve 10 user's data picking out with predetermined rule in the database 300, to be used for real-time processing.The recommendation apparatus 1000 that this embodiment relates to disperses computational load through recommendation process being divided batch processing with handling in real time, and the result of comprehensive batch processing portion 1100 exports with the result of handling part 1200 in real time in recommendation results efferent 1300.As comprehensive method, both can also can directly export exporting recommendation results after two results added according to the result of time period any one or in handling in real time with batch processing as recommendation results.In addition, also can two results' mean value or weighted mean be exported as recommendation results.

< the detailed formation of recommendation apparatus >

As shown in Figure 2, the recommendation apparatus that this embodiment relates to comprises: the historical information collection portion 1101, correlation calculations portion 1102, recommendation degree calculating part 1103, the storage part 1104 that constitute batch processing portion 1100; Constitute historical information collection portion 1201, correlation calculations portion 1202, the recommendation degree calculating part 1203 of real-time handling part 1200; And recommendation results efferent 1300.

Historical information collection portion 1101 is from the required historical information of database 200 collection analysis.Correlation calculations portion 1102 analyzes collected historical information and defines correlativity.Recommendation degree calculating part 1103 uses the correlativity of calculating to calculate the recommendation degree to each content of user.The recommendation degree that storage part 1104 storages are calculated.

Historical information collection portion 1201 is from the required historical information of database 300 collection analysis.Correlation calculations portion 1202 analyzes collected historical information and defines correlativity.Recommendation degree calculating part 1203 uses the correlativity of being calculated to calculate the recommendation degree to each content of user.

Recommendation results efferent 1300 is exported recommendation results based on the recommendation degree of the recommendation degree of each content of storage in the storage part 1104 and each content of calculating through recommendation degree calculating part 1203.

< processing of recommendation apparatus >

Use Fig. 3 describes the processing of the recommendation apparatus that this embodiment relates to.

At first; The historical information collection portion 1101 of batch processing portion 1100 is from the required historical information (step S101) of database 200 collection analysis; Correlation calculations portion 1102 analyzes collected historical information and defines correlativity; For example, compare the historical of two users and define correlativity, perhaps compare the historical of two contents and define correlativity (step S102) based on the content number in the history that is included in two users simultaneously based on having historical content number simultaneously.Recommendation degree calculating part 1103 uses the correlativity of being calculated to calculate the recommendation degree to each content of user; For example; Calculate this recommendation degree (step S103) based on the correlativity between existing content and the other guide in the correlativity between this user and other users or this user's the history, the recommendation degree (step S104) that storage part 1104 storages are calculated with content utilization identical with this user.

The historical information collection portion 1201 of handling part 1200 is from the required historical information (step S105) of database 300 collection analysis in real time, and correlation calculations portion 1202 analyzes collected historical information and defines correlativity (step S106) with the mode identical with aforesaid way.Recommendation degree calculating part 1203 uses the correlativity of being calculated to calculate the recommendation degree (step S107) to each content of user with the mode identical with aforesaid way.

Then, recommendation results efferent 1300 is exported recommendation results (step S108) based on the recommendation degree of the recommendation degree of each content of storage in the storage part 1104 and each content of calculating through recommendation degree calculating part 1203.

That kind as described above; According to this embodiment; Owing to the recommendation degree of the correlativity of in real-time processing, calculating based on recommendation degree that uses the correlativity of in batch processing, calculating and use is exported recommendation results, the historical information and the required calculated amount of minimizing analysis that therefore can not limit utilization realize high-precision recommendation.

< second embodiment >

Use Fig. 4 to Fig. 6, second embodiment of the present invention is described.

< formation of recommendation apparatus >

Use Fig. 4, the formation of the recommendation apparatus that this embodiment is related to describes.

As shown in Figure 4, the recommendation apparatus 1000 that this embodiment relates to comprises batch processing portion 1110, real-time handling part 1210 and recommendation results efferent 1310, and batch processing portion 1110 is connected with database 200, and handling part 1210 is connected with database 300 in real time.

The recommendation apparatus 1000 that this embodiment relates to disperses computational load through recommendation process being divided batch processing with handling in real time, and comprehensive two results from real-time handling part 1210 export in recommendation results efferent 1310.As comprehensive method, both can also can directly export exporting recommendation results after two results added according to the result of time period any one or in handling in real time with batch processing as recommendation results.In addition, also can two results' mean value or weighted mean be exported as recommendation results.

< the detailed formation of recommendation apparatus >

As shown in Figure 5, the recommendation apparatus that this embodiment relates to comprises: the historical information collection portion 1101, correlation calculations portion 1102, the storage part 1115 that constitute batch processing portion 1110; Constitute real-time handling part 1210 historical information collection portion 1201, correlation calculations portion 1202, recommendation degree calculating part 1203, obtain portion 1214, recommendation degree calculating part 1215; And recommendation results efferent 1310.About having been marked the inscape of the Reference numeral identical,, therefore omit its detailed description owing to have identical functions with first embodiment.

The correlativity that storage part 1115 storages are calculated.Obtain the correlativity that portion 1214 obtains storage in the storage part 1115.Recommendation degree calculating part 1215 uses the correlativity of being obtained to calculate the recommendation degree to each content of user.

Recommendation results efferent 1310 is exported recommendation results based on the recommendation degree of each content that the recommendation degree and the recommendation degree calculating part 1215 of each content of calculating through recommendation degree calculating part 1203 are calculated.

< processing of recommendation apparatus >

Use Fig. 6 describes the processing of the recommendation apparatus that this embodiment relates to.

At first; The historical information collection portion 1101 of batch processing portion 1110 is from the required historical information (step S201) of database 200 collection analysis; Correlation calculations portion 1102 analyzes collected historical information and defines correlativity (step S202), the correlativity (step S203) that storage part 1115 storages are calculated.

The historical information collection portion 1201 of handling part 1210 is from the required historical information (step S204) of database 300 collection analysis in real time, and correlation calculations portion 1202 analyzes collected historical information and defines correlativity (step S205).Recommendation degree calculating part 1203 uses the correlativity of being calculated to calculate the recommendation degree (step S206) to each content of user.In addition, obtain the correlativity (step S207) that portion 1214 obtains storage in the storage part 1115, recommendation degree calculating part 1215 uses the correlativity of being obtained to calculate the recommendation degree (step S208) to each content of user.

Then, the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through recommendation degree calculating part 1203 with through recommendation degree calculating part 1215 of recommendation results efferent 1310 is exported recommendation results (step S209).

That kind as described above; According to this embodiment; Real-time handling part is based on the correlativity of calculating in the batch processing and the correlativity of in handling in real time, calculating come the calculated recommendation degree; And based on these recommendation degree output recommendation results, the historical information and the required calculated amount of minimizing analysis that therefore can not limit utilization realize high-precision recommendation.

< the 3rd embodiment >

Use Fig. 7 and Fig. 9, the 3rd embodiment of the present invention is described.

< formation of recommendation apparatus >

Use Fig. 7, the formation of the recommendation apparatus that this embodiment is related to describes.

As shown in Figure 7, the recommendation apparatus 1000 that this embodiment relates to comprises batch processing portion 1120, real-time handling part 1220 and recommendation results efferent 1320, and batch processing portion 1120 is connected with database 200, and handling part 1220 is connected with database 300 in real time.

The recommendation apparatus 1000 that this embodiment relates to disperses computational load through recommendation process being divided batch processing with handling in real time; And batch processing portion 1120 sends to real-time handling part 1220 with the result, and in recommendation results efferent 1320, comprehensively exports from batch processing portion 1120 and two results of real-time handling part 1220.As comprehensive method, both can also can directly export exporting recommendation results after two results added according to the result of time period any one or in handling in real time with batch processing as recommendation results.In addition, also can two results' mean value or weighted mean be exported as recommendation results.

< the detailed formation of recommendation apparatus >

As shown in Figure 8, the recommendation apparatus that this embodiment relates to comprises: the historical information collection portion 1101, correlation calculations portion 1102, recommendation degree calculating part 1103, the storage part 1126 that constitute batch processing portion 1120; Constitute real-time handling part 1220 historical information collection portion 1201, recommendation degree calculating part 1226, obtain portion 1227; And recommendation results efferent 1320.About having been marked the inscape of the Reference numeral identical,, therefore omit its detailed description owing to have identical functions with first embodiment.

Storage part 1126 is stored in the correlativity of definition in the correlation calculations portion 1102 and the recommendation degree of in recommendation degree calculating part 1103, calculating.Obtain portion 1227 and obtain the correlativity of the definition in correlation calculations portion 1102 of storage in the storage part 1126 and the recommendation degree of in recommendation degree calculating part 1103, calculating.Recommendation degree calculating part 1226 uses the correlativity of being obtained to calculate the recommendation degree to each content of user.

The recommendation degree of each content that recommendation results efferent 1320 is calculated based on the recommendation degree of each content of calculating through recommendation degree calculating part 1103 with through recommendation degree calculating part 1226 is exported recommendation results.

< processing of recommendation apparatus >

Use Fig. 9, the processing of the recommendation apparatus that this embodiment is related to describes.

At first, the historical information collection portion 1101 of batch processing portion 1120 is from the required historical information (step S301) of database 200 collection analysis, and correlation calculations portion 1102 analyzes collected historical information and defines correlativity (step S302).Recommendation degree calculating part 1103 uses the correlativity of being calculated to calculate the recommendation degree (step S303) to each content of user, and storage part 1126 is stored in the correlativity of definition in the correlation calculations portion 1102 and the recommendation degree (step S304) of in recommendation degree calculating part 1103, calculating.

The historical information collection portion 1201 of handling part 1220 is from the required historical information (step S305) of database 300 collection analysis in real time.Obtain portion 1227 and obtain the correlativity of the definition in correlation calculations portion 1102 of storage in the storage part 1126 and the recommendation degree (step S306) of in recommendation degree calculating part 1103, calculating, recommendation degree calculating part 1226 uses the correlativity of being calculated to calculate the recommendation degree (step S307) to each content of user.

Then, the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through recommendation degree calculating part 1103 with through recommendation degree calculating part 1226 of recommendation results efferent 1320 is exported recommendation results (step S308).

That kind as described above; According to this embodiment; Export recommendation results based on the recommendation degree that uses the correlativity in batch processing, calculate with through the recommendation degree that real-time processing uses the correlativity in batch processing, calculated to provide, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

< the 4th embodiment >

Use Figure 10 to Figure 12, the 4th embodiment of the present invention is described.

< formation of commending system >

Use Figure 10, the formation of the commending system that this embodiment is related to describes.

Shown in figure 10, the commending system that this embodiment relates to comprises batch processing device 400, real-time treating apparatus 500 and recommendation results output unit 600, and batch processing device 400 is connected with database 200, and treating apparatus 500 is connected with database 300 in real time.

The commending system that this embodiment relates to disperses computational load through recommendation process being divided batch processing with handling in real time, and the result of comprehensive batch processing device 400 exports with the result of treating apparatus 500 in real time in recommendation results output unit 600.As batch processing device 400; Can enumerate the example of implementing through the high Cloud Server of calculation process ability; As real-time treating apparatus 500, except that likewise implementing, also can enumerate through calculation process ability lower portable terminal device or STB (Set Top Box through the high Cloud Server of calculation process ability with batch processing device 400; STB) etc. the example of implementing, but be not limited thereto.

< the detailed formation of commending system >

Shown in figure 11, the commending system that this embodiment relates to comprises: the historical information collection portion 401, correlation calculations portion 402, recommendation degree calculating part 403, the storage part 404 that constitute batch processing device 400; Constitute historical information collection portion 501, correlation calculations portion 502, the recommendation degree calculating part 503 of real-time treating apparatus 500; And recommendation results output unit 600.

Historical information collection portion 401 is from the required historical information of database 200 collection analysis.Correlation calculations portion 402 analyzes collected historical information and defines correlativity.Recommendation degree calculating part 403 uses the correlativity of being calculated to calculate the recommendation degree to each content of user.The recommendation degree that storage part 404 storages are calculated.

Historical information collection portion 501 is from the required historical information of database 300 collection analysis.Correlation calculations portion 502 analyzes collected historical information and defines correlativity.Recommendation degree calculating part 503 uses the correlativity of being calculated to calculate the recommendation degree to each content of user.

Recommendation results output unit 600 is exported recommendation results based on the recommendation degree of the recommendation degree of each content of storage in the storage part 404 and each content of calculating through recommendation degree calculating part 503.

< processing of commending system >

Use Figure 12, the processing of the commending system that this embodiment is related to describes.

At first, the historical information collection portion 401 of batch processing device 400 is from the required historical information (step S401) of database 200 collection analysis, and correlation calculations portion 402 analyzes collected historical information and defines correlativity (step S402).Recommendation degree calculating part 403 uses the correlativity of being calculated to calculate the recommendation degree (step S403) to each content of user, the recommendation degree (step S404) that storage part 404 storages are calculated.

The historical information collection portion 501 of treating apparatus 500 is from the required historical information (step S405) of database 300 collection analysis in real time, and correlation calculations portion 502 analyzes collected historical information and defines correlativity (step S406).Recommendation degree calculating part 503 uses the correlativity of being calculated to calculate the recommendation degree (step S407) to each content of user.

Then, recommendation results output unit 600 is exported recommendation results (step S408) based on the recommendation degree of the recommendation degree of each content of storage in the storage part 404 and each content of calculating through recommendation degree calculating part 503.

That kind as described above; According to this embodiment; Recommendation apparatus is based on recommendation degree that uses the correlativity of in the batch processing device, calculating and the recommendation degree that uses the correlativity of in real-time treating apparatus, calculating; Export recommendation results, the historical information and the required calculated amount of minimizing analysis that therefore can not limit utilization realize high-precision recommendation.

< the 5th embodiment >

Use Figure 13 to Figure 15, the 5th embodiment of the present invention is described.

< formation of commending system >

Use Figure 13, the formation of the commending system that this embodiment is related to describes.

Shown in figure 13, the commending system that this embodiment relates to comprises batch processing device 410, real-time treating apparatus 510 and recommendation results output unit 610, and batch processing device 410 is connected with database 200, and treating apparatus 510 is connected with database 300 in real time.

The commending system that this embodiment relates to disperses computational load through recommendation process being divided batch processing with handling in real time, and comprehensive two results from real-time treating apparatus 410 export in recommendation results output unit 610.As batch processing device 410; Can enumerate the example of implementing through the high Cloud Server of calculation process ability; As real-time treating apparatus 510; Except that likewise implementing, also can enumerate through calculation process ability lower portable terminal device or STB (Set Top Box) and wait the example of enforcement, but be not limited thereto through the high Cloud Server of calculation process ability with batch processing device 410.

< the detailed formation of commending system >

Shown in figure 14, the commending system that this embodiment relates to comprises: the historical information collection portion 401, correlation calculations portion 402, the storage part 415 that constitute batch processing device 410; Constitute real-time treating apparatus 510 historical information collection portion 501, correlation calculations portion 502, recommendation degree calculating part 503, obtain portion 514, recommendation degree calculating part 515; And recommendation results output unit 610.About having been marked the inscape of the Reference numeral identical,, therefore omit its detailed description owing to have identical functions with the 4th embodiment.

The correlativity that storage part 415 storages are calculated.Obtain the correlativity that portion 514 obtains storage in the storage part 415.Recommendation degree calculating part 515 uses the correlativity of being obtained to calculate the recommendation degree to each content of user.

The recommendation degree of each content that recommendation results output unit 610 is calculated based on the recommendation degree of each content of calculating through recommendation degree calculating part 503 with through recommendation degree calculating part 515 is exported recommendation results.

< processing of commending system >

Use Figure 15, the processing of the commending system that this embodiment is related to describes.

At first; The historical information collection portion 401 of batch processing device 410 is from the required historical information (step S501) of database 200 collection analysis; Correlation calculations portion 402 analyzes collected historical information and defines correlativity (step S502), the correlativity (step S503) that storage part 415 storages are calculated.

The historical information collection portion 501 of treating apparatus 510 is from the required historical information (step S504) of database 300 collection analysis in real time, and correlation calculations portion 502 analyzes collected historical information and defines correlativity (step S505).Recommendation degree calculating part 503 uses the correlativity of being calculated to calculate the recommendation degree (step S506) to each content of user.In addition, obtain the correlativity (step S507) that portion 514 obtains storage in the storage part 415, recommendation degree calculating part 515 uses the correlativity of being obtained to calculate the recommendation degree (step S508) to each content of user.

Then, the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through recommendation degree calculating part 503 with through recommendation degree calculating part 515 of recommendation results output unit 610 is exported recommendation results (step S509).

That kind as described above; According to this embodiment; Real-time treating apparatus is based on the correlativity of calculating in the batch processing device and the correlativity of in real-time treating apparatus, calculating come the calculated recommendation degree; And based on these recommendation degree output recommendation results, the historical information and the required calculated amount of minimizing analysis that therefore can not limit utilization realize high-precision recommendation.

< the 6th embodiment >

Use Figure 16 to Figure 18, the 6th embodiment of the present invention is described.

< formation of commending system >

Use Figure 16, the formation of the commending system that this embodiment is related to describes.

Shown in figure 16, the commending system that this embodiment relates to comprises batch processing device 420, real-time treating apparatus 520 and recommendation results output unit 620, and batch processing device 420 is connected with database 200, and treating apparatus 520 is connected with database 300 in real time.

The commending system that this embodiment relates to disperses computational load through recommendation process being divided batch processing with handling in real time; And batch processing device 420 sends to real-time treating apparatus 520 with the result, and in recommendation results output unit 620, comprehensively exports from two results of batch processing device 420 with real-time treating apparatus 520.As batch processing device 420; Can enumerate the example of implementing through the high Cloud Server of calculation process ability; As real-time treating apparatus 520; Except that likewise implementing, also can enumerate through calculation process ability lower portable terminal device or STB (Set Top Box) and wait the example of enforcement, but be not limited thereto through the high Cloud Server of calculation process ability with batch processing device 420.

< the detailed formation of commending system >

Shown in figure 17, the commending system that this embodiment relates to comprises: the historical information collection portion 401, correlation calculations portion 402, recommendation degree calculating part 403, the storage part 426 that constitute batch processing device 420; Constitute real-time treating apparatus 520 historical information collection portion 501, recommendation degree calculating part 526, obtain portion 527; And recommendation results output unit 620.About having been marked the inscape of the Reference numeral identical,, therefore omit its detailed description owing to have identical functions with the 4th embodiment.

Storage part 426 is stored in the correlativity of definition in the correlation calculations portion 402 and the recommendation degree of in recommendation degree calculating part 403, calculating.Obtain portion 527 and obtain the correlativity of the definition in correlation calculations portion 402 of storage in the storage part 426 and the recommendation degree of in recommendation degree calculating part 403, calculating.Recommendation degree calculating part 526 uses the correlativity of being obtained to calculate the recommendation degree to each content of user.

The recommendation degree of each content that recommendation results output unit 620 is calculated based on the recommendation degree of each content of calculating through recommendation degree calculating part 403 with through recommendation degree calculating part 526 is exported recommendation results.

< processing of commending system >

Use Figure 18, the processing of the commending system that this embodiment is related to describes.

At first, the historical information collection portion 401 of batch processing device 420 is from the required historical information (step S601) of database 200 collection analysis, and correlation calculations portion 402 analyzes collected historical information and defines correlativity (step S602).Recommendation degree calculating part 403 uses the correlativity of being calculated to calculate the recommendation degree (step S603) to each content of user, and storage part 426 is stored in the correlativity of definition in the correlation calculations portion 402 and the recommendation degree (step S604) of in recommendation degree calculating part 403, calculating.

The historical information collection portion 501 of treating apparatus 520 is from the required historical information (step S605) of database 300 collection analysis in real time.Obtain portion 527 and obtain the correlativity of the definition in correlation calculations portion 402 of storage in the storage part 426 and the recommendation degree (step S606) of in recommendation degree calculating part 403, calculating, recommendation degree calculating part 526 uses the correlativity of being calculated to calculate the recommendation degree (step S607) to each content of user.

Then, the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through recommendation degree calculating part 403 with through recommendation degree calculating part 526 of recommendation results output unit 620 is exported recommendation results (step S608).

That kind as described above; According to this embodiment; Use the correlativity of in the batch processing device, calculating and the recommendation degree that provides exported recommendation results based on the recommendation degree that uses the correlativity in the batch processing device, calculate with by real-time treating apparatus, therefore can not limit the historical information of utilization and reduce and analyze required calculated amount and realize high-precision recommendation.

< embodiment 1 >

Use Figure 19 to Figure 21, embodiment 1 is described.Present embodiment is the example that the recommendation apparatus that first embodiment relates to is specialized more.

< formation of recommendation apparatus >

Shown in figure 19, the recommendation apparatus that present embodiment relates to comprises batch processing portion 1100, real-time handling part 1200, recommendation results efferent 1300, database 1410, reduction portion 1420 and database 1430.

In addition; Batch processing portion 1100 comprises historical information collection portion 1101, correlation calculations portion 1102, recommendation degree calculating part 1103 and storage part 1104, and handling part 1200 comprises historical information collection portion 1201, correlation calculations portion 1202 and recommendation degree calculating part 1203 in real time.About having been marked the inscape of the Reference numeral identical,, therefore omit its detailed description owing to have identical functions with first embodiment.

Here, database 1410 is the databases that keep whole users' historical information etc., the optimal user that reduction portion 1420 is suitable for recommending from the extracting data in the database 1410.The historical information of the optimal user that database 1430 keeps being extracted by reduction portion 1420 etc.

< processing of recommendation apparatus >

Use Figure 20, the processing of the recommendation apparatus that present embodiment is related to describes.

At first, based on the data in the database 1410 (being expressed as " evaluation history T " in the drawings), extract good user from its effectiveness (Utility) value.This is the processing of from whole historical datas, extracting the user data that stays abundant evaluating data.Usually, a lot of users almost do not have history in the real data, if use this user's data, calculated amount need not increase on ground, and precision also improves not quite, so will carry out the screening of data.

Then, stochastic sampling optimal user.This is that data volume is tapered to the processing of being able to catch up with handling in real time the degree in required processing time.At this moment; Pull out between preferably carrying out, in order to avoid data generation bias, can consider " with user data be divided into N bunch → select the processing of N representative of consumer with each bunch people's mode " etc.; In order to simplify this processing, (when carrying out batch processing) selected N people at random when sampling at every turn.

In batch processing, similarity is used as correlativity between the user calculating between good user and all users, and to each user's calculated recommendation degree.Here, suppose that similarity between the user is such in user's similarity that the common point of the evaluation of content is defined the similar scale between the user and the user's that accepts to recommend unregistered content is estimated relatively between the user shown in non-patent literature 1.And similarity between this user is remained among the middle T1 (being equivalent to storage part 1104).When recommending, the ID of recommending the object user as key word, is obtained the recommendation degree from middle T1, and obtain the prediction and evaluation value.

On the other hand, in handling in real time, the stochastic sampling optimal user, and in the middle of the historical information of relative users is kept in advance among the T2 (being equivalent to database (reduction) 1430).When handling in real time, obtain the historical information of optimal user from middle T2, and on the other hand, obtain the historical information of recommending the object user.Then, based on this information of obtaining, the similarity of calculated recommendation object user and optimal user is calculated the prediction and evaluation value.Obtain the prediction and evaluation value then.

That is, in handling in real time, operation is handled at every turn when needing recommendation process.And, when handling, when obtaining the data of optimal user, from evaluation history T, obtain the user's who accepts recommendation history in addition.Therefore, can use the user's who accepts recommendation up-to-date history to come real-time implementation to recommend.In addition, because each correlativity of calculating, thereby can find similar user this moment, thus for the historical not enough user before registration just and batch processing after the batch processing, can reflect the history on the same day at once.

Figure 21 exemplarily show the user that uses in the recommendation process and batch processing and handle in real time between relation.According to this figure; Based on all user's data that are kept at for example 1,000,000 people in the evaluation history T; From its utility value, extract for example 10000 people's good user, and the stochastic sampling optimal user, for example extract 100 people's optimal user thus and it is saved in the middle of among the T2.

The information of evaluation history T and good user's information are used to batch processing, and calculate correlativity based on all users and good user's information.On the other hand, in handling in real time, calculate correlativity based on the recommendation object user's who from evaluation history T, obtains the historical information and the historical information of optimal user.

< embodiment 2 >

Use Figure 22 and Figure 23, embodiment 2 is described.Present embodiment is the example that the recommendation apparatus that the 3rd embodiment relates to is specialized more.

< formation of recommendation apparatus >

Shown in figure 22, the recommendation apparatus that present embodiment relates to comprises batch processing portion 1120, real-time handling part 1220, recommendation results efferent 1320 and database 1410.

In addition; Batch processing portion 1120 comprises historical information collection portion 1101, correlation calculations portion 1102, recommendation degree calculating part 1103 and storage part 1126, and handling part 1220 comprises historical information collection portion 1201, recommendation degree calculating part 1226 and obtains portion 1227 in real time.About having been marked the inscape of the Reference numeral identical,, therefore omit its detailed description owing to have identical functions with the 3rd embodiment.

Here, database 1410 is the databases that keep whole users' historical information etc.

< processing of recommendation apparatus >

Use Figure 23, the processing of the recommendation apparatus that present embodiment is related to describes.

At first, based on the data in the database 1410 (being expressed as " evaluation history T " in the drawings), extract good user from its utility value.This is the processing of from whole historical datas, extracting the user data that stays abundant evaluating data.Usually, a lot of users almost do not have history in the real data, if use this user's data, calculated amount need not increase on ground, and precision also improves not quite, so carry out the screening of data.

In batch processing, use that good evaluation of user is historical calculates that similarity is used as correlativity between content, and to each user's calculated recommendation degree.And, should be kept among the middle T1 by the recommendation degree.Here, suppose that similarity between content is such in user's similarity that the common point of the evaluation of content is defined the similar scale between content and the user's that accepts to recommend unregistered content is estimated relatively between the content shown in non-patent literature 4.The ID that when recommending, will recommend the object user obtains the recommendation degree as key word from middle T1, and obtains the prediction and evaluation value.The similarity data are saved among the middle T2 between the content that will in the process of calculating the prediction and evaluation value, obtain on the other hand.

On the other hand, in handling in real time, obtain the historical information of recommending the object user, from middle T2, obtain with historical information in similarity between the relevant content of the content that comprises.Then, obtain the prediction and evaluation value based on this information of obtaining.

That is, in handling in real time, the value of the similarity of calculating when being utilized in batch processing again.In addition, when obtaining the historical information of recommending the object user, also can only use the historical information of " up-to-date X incident ".Particularly, create the calculating that the blank user data of up-to-date X incident historical information is in the past carried out the prediction and evaluation value.Thus, can recommend based on the history of upgrading.

On the other hand, be included in the middle of correlativity among the T2 be that similarity is a foundation between the content during with batch processing.Therefore, though can not reflect the history on the same day, similarity is not easy to change than similarity between the user between content, so it recommends precision to be not easy to receive big influence.

Be recorded in the computer-readable recording medium through processing recommendation apparatus or commending system; And the program in this recording medium of will being recorded in is read in recommendation apparatus, batch processing device, treating apparatus or recommendation results output unit are carried out in real time, can realize recommendation apparatus of the present invention or commending system.Here said computer system comprises the hardware of OS or peripheral unit etc.

In addition, " computer system " comprises also that then homepage provides environment (or display environment) if utilize the occasion of WWW (World Wide Web, WWW) system.In addition, said procedure also can from the computer system this program being remained on memory storage etc. via transmission medium, or be transferred in other computer systems through the transmission ripple in the transmission medium.Here, " transmission medium " of transmission procedure is meant the medium that as communication lines (order wire) such as networks such as internet (communication network) or telephone line, has information transfer capability.

In addition, said procedure also can be the program that is used to realize the part of above-mentioned functions.In addition, also can be can with above-mentioned computer system in program recorded make up the so-called differential file (difference program) of realizing above-mentioned functions.

More than, with reference to accompanying drawing embodiment of the present invention is specified, but concrete constitute be not limited to this embodiment, also be included in design in the scope that does not break away from main idea of the present invention etc.For example, in the present invention, supposed that storage mediums such as using database manages concentratedly the historical information of utilizing, and used historical information collection portion to obtain these historical informations, but and nonessentially be defined in this.

In addition, the tabulation to the recommendation degree of this user's content one by one that provides method to suppose will to have write down according to recommendation results of the present invention offers outside recommendation reminding method, but must not be limited to this method of application.

Each processing of recommendation of the present invention also can be the implementation method of implementing through a server, perhaps implements, or has used implementation method that the cooperation of a plurality of servers of load dispersing function implements etc. through the cooperation by a plurality of servers of function customizations.

Description of reference numerals

1100; Batch processing portion

1110; Batch processing portion

1120; Batch processing portion

1101; Historical information collection portion

1102; Correlation calculations portion

1103; Recommendation degree calculating part

1104; Storage part

1115; Storage part

1126; Storage part

1200; Real-time handling part

1210; Real-time handling part

1220; Real-time handling part

1201; Historical information collection portion

1202; Correlation calculations portion

1203; Recommendation degree calculating part

1214; Obtain portion

1215; Recommendation degree calculating part

1226; Recommendation degree calculating part

1237; Obtain portion

1300; The recommendation results efferent

1310; The recommendation results efferent

1320; The recommendation results efferent

1410; Database

1420; Reduction portion

200; Database

300; Database

400; The batch processing device

410; The batch processing device

420; The batch processing device

401; Historical information collection portion

402; Correlation calculations portion

403; Recommendation degree calculating part

404; Storage part

415; Storage part

426; Storage part

500; Real-time treating apparatus

510; Real-time treating apparatus

520; Real-time treating apparatus

501; Historical information collection portion

502; Correlation calculations portion

503; Recommendation degree calculating part

514; Obtain portion

515; Recommendation degree calculating part

526; Recommendation degree calculating part

527; Obtain portion

600; The recommendation results output unit

610; The recommendation results output unit

620; The recommendation results output unit.

Claims

1. recommendation apparatus; Its use based on buy the historical information historical or user that reading is historical define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results; Said recommendation apparatus is characterised in that, comprising:

Batch processing unit, the said historical information of its ex ante analysis generate the required information of recommending;

Real-time processing unit, it is analyzed historical information and generates the required information of recommending when recommendation request takes place; And

The recommendation results output unit, the output of its comprehensive said batch processing unit is or/and recommendation results is exported in the output of said real-time processing unit.

2. recommendation apparatus according to claim 1 is characterized in that,

Said batch processing unit comprises:

The first historical information collector unit, the historical information that its collection analysis is required;

The first correlation calculations unit, its historical information of analyzing said collection defines correlativity;

The first recommendation degree computing unit, it uses the said correlativity of calculating to calculate the recommendation degree to each content of user; And

Storage unit, it stores said recommendation degree of calculating,

Said real-time processing unit comprises:

The second historical information collector unit, the required historical information of collection analysis in the time of in fact;

The second correlation calculations unit, its historical information of analyzing said collection defines correlativity; And

The second recommendation degree computing unit, it uses the said correlativity of calculating to calculate the recommendation degree to each content of user,

Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of said storage with by the said second recommendation degree computing unit of said recommendation results output unit is exported recommendation results.

3. recommendation apparatus according to claim 1 is characterized in that,

Said batch processing unit comprises:

The first correlation calculations unit, its historical information of analyzing said collection defines correlativity; And

Storage unit, it stores the said correlativity of calculating,

Said real-time processing unit comprises:

The second correlation calculations unit, its historical information of analyzing said collection defines correlativity;

The first recommendation degree computing unit, it uses the said correlativity of calculating to calculate the recommendation degree to each content of user;

Acquiring unit, it obtains the correlativity of said storage; And

The second recommendation degree computing unit, it uses the said correlativity of obtaining to calculate the recommendation degree to each content of user,

Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of being calculated by the said first recommendation degree computing unit with by the said second recommendation degree computing unit of said recommendation results output unit is exported recommendation results.

4. recommendation apparatus according to claim 1 is characterized in that,

Said batch processing unit comprises:

Storage unit, it stores the correlativity of said definition and the recommendation degree of calculating,

Said real-time processing unit comprises:

Acquiring unit, it obtains the correlativity of said definition and the recommendation degree of calculating; And

5. commending system; It comprises batch processing device, real-time treating apparatus and recommendation apparatus; And use based on buy the historical information historical or user that reading is historical define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results; Said commending system is characterised in that

The said historical information of said batch processing device ex ante analysis generates the required information of recommending; Said real-time treating apparatus is analyzed historical information and is generated the required information of recommendation when recommendation request takes place, the output of the comprehensive said batch processing device of said recommendation apparatus is or/and recommendation results is exported in the output of said real-time treating apparatus.

6. commending system according to claim 5 is characterized in that,

Said batch processing device comprises:

Storage unit, it stores said recommendation degree of calculating,

Said real-time treating apparatus comprises:

Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of said storage with by the said second recommendation degree computing unit of said recommendation apparatus is exported recommendation results.

7. commending system according to claim 5 is characterized in that,

Said batch processing device comprises:

Storage unit, it stores the said correlativity of calculating,

Said real-time treating apparatus comprises:

Acquiring unit, it obtains the correlativity of said storage; And

Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of being calculated by the said first recommendation degree computing unit with by the said second recommendation degree computing unit of said recommendation apparatus is exported recommendation results.

8. commending system according to claim 5 is characterized in that,

Said batch processing device comprises:

Storage unit, it stores the correlativity of said definition and the recommendation degree of calculating;

Said real-time treating apparatus comprises:

9. recommend method; Its use based on buy the historical information historical or user that reading is historical define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results; Said recommend method is characterised in that, comprising:

The batch processing step is used for the said historical information of ex ante analysis and generates the required information of recommending;

Treatment step is used for when recommendation request takes place, analyzing historical information and generates the required information of recommending in real time; And

Recommendation results output step is used for the output of comprehensive said batch processing step or/and recommendation results is exported in the output of said real-time treatment step.

10. recommend method according to claim 9 is characterized in that,

Said batch processing step comprises:

The first step of the historical information that collection analysis is required;

The historical information of analyzing said collection defines second step of correlativity;

Use the said correlativity of calculating to calculate third step to the recommendation degree of each content of user; And

Store the 4th step of said recommendation degree of calculating,

Said real-time treatment step comprises:

Real-time collecting is analyzed the 5th step of required historical information;

The historical information of analyzing said collection defines the 6th step of correlativity; And

Use the said correlativity of calculating to calculate the 7th step to the recommendation degree of each content of user,

Wherein, said recommendation results output step is based on the recommendation degree of each content of storing in said the 4th step and the recommendation degree of each content of calculating through said the 7th step is exported recommendation results.

11. recommend method according to claim 9 is characterized in that,

Said batch processing step comprises:

The historical information of analyzing said collection defines second step of correlativity; And

Store the third step of the said correlativity of calculating,

Said real-time treatment step comprises:

Real-time collecting is analyzed the 4th step of required historical information;

The historical information of analyzing said collection defines the 5th step of correlativity;

Use the said correlativity of calculating to calculate the 6th step to the recommendation degree of each content of user;

Obtain the 7th step of the correlativity of said storage; And

Use the said correlativity of obtaining to calculate the 8th step to the recommendation degree of each content of user,

Wherein, the recommendation degree of each content of calculating based on the recommendation degree of each content of calculating through said the 6th step with through said the 8th step of said recommendation results output step is exported recommendation results.

12. recommend method according to claim 9 is characterized in that,

Said batch processing step comprises:

Store the correlativity of said definition and the 4th step of the recommendation degree of calculating,

Said real-time treatment step comprises:

Obtain the correlativity of said definition and the 6th step of the recommendation degree of calculating; And

Use the said correlativity of obtaining to calculate the 7th step to the recommendation degree of each content of user,

The recommendation degree of each content that said recommendation results output step is calculated based on the recommendation degree of each content of calculating through said third step with through said the 7th step is exported recommendation results.

13. one kind is used to make computing machine to carry out the program of recommend method; Wherein, Said recommend method use based on buy the historical information historical or user that reading is historical define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results, said program is carried out computing machine:

14. one kind is used for making computing machine to carry out the program of the recommend method of commending system; Wherein, Said commending system comprises batch processing device, real-time treating apparatus and recommendation apparatus; And use based on buy the historical information historical or user that reading is historical define correlativity, and output be predicted to be the coordination filter method of user's interest content based on the correlativity of said definition as recommendation results, said program is carried out computing machine:

Said historical information generated the first step of recommending required information about said batch processing device divided in advance;

Said real-time treating apparatus is analyzed historical information and is generated second step of recommending required information when recommendation request takes place; And

The output of the comprehensive said first step of said recommendation apparatus is or/and the third step of recommendation results is exported in the output of said second step.