CN105005582A - Recommendation method and device for multimedia information - Google Patents

Abstract

The invention discloses a recommendation method and device for multimedia information and belongs to the technical field of Internet. The recommendation method comprises: determining first multimedia classes interested by a first user according to a historical click action of the multimedia information; determining second multimedia classes which are potentially interested and are not historically clicked by the first user according to each first multimedia class of the first user and a plurality of historical click actions on the multimedia information by the second user; determining a target multimedia class according to the first multimedia classes interested by the first user and the second multimedia classes which are potentially interested by the first user; and recommending the multimedia information of the target multimedia class. When the information is recommended, the determined target multimedia class is more accurate by realizing the multimedia classes of the multimedia information historically clicked by the first user and realizing the multimedia classes which are potentially interested by the first user, so that the clicking rate of the recommended multimedia information can be improved and resources and financial resources can be saved.

Description

The recommend method of multimedia messages and device

Technical field

The present invention relates to Internet technical field, particularly a kind of recommend method of multimedia messages and device.

Background technology

Along with the swift and violent growth of market economy, market competition is more and more fierce.Each businessman, in order to promote oneself product or service better, usually can make to the product of oneself or serve relevant multimedia messages, and recommending these multimedia messagess by various common platform to user.When the multimedia messages that common platform is recommended just in time is met consumers' demand, the probability of transaction of product or service can be increased, thus the income of businessman can be increased.Under this kind of scene, how its interested or relevant to its demand multimedia messages is recommended to seem very important to user.

In prior art, when recommending multimedia messages, common platform can be analyzed the multimedia messages clicked in user's nearest a period of time, obtains the multimedia classification belonging to the interested multimedia messages of user, and recommends these multimedia class object multimedia messagess.

Realizing in process of the present invention, inventor finds that correlation technique at least exists following problem:

In the prior art, directly determine the interested multimedia classification of user according to the multimedia classification clicked in user's nearest a period of time, and the reasons such as contingency maybe may be there is because data volume is little in the multimedia classification clicked in a period of time recently, the interested multimedia classification of user may not be contained, thus there is certain limitation, result through user's interested multimedia classification that this mode determines not accurate enough, and then make when recommending these multimedia class object multimedia messagess, the multimedia messages many times recommended can not cause user to note, and then cause the multimedia messages specific aim of recommendation not strong, and cause the clicking rate of the multimedia messages recommended not high, thus cause waste resource and financial resources.

Summary of the invention

In order to solve the problem of prior art, embodiments provide a kind of recommend method and device of multimedia messages.Described technical scheme is as follows:

First aspect, provides a kind of recommend method of multimedia messages, and described method comprises:

According to first user, behavior is clicked to the history of multimedia messages, determine the interested first multimedia classification of described first user, classification belonging to the multimedia messages that described first multimedia classification is clicked for user's history;

Behavior is clicked according to each the first multimedia classification of described first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of described first user, classification belonging to the multimedia messages that described second multimedia classification is not clicked for user's history;

According to the interested first multimedia classification of described first user and the potential interested second multimedia classification of described first user, determine destination multimedia classification;

The multimedia messages of described destination multimedia classification is recommended to described first user.

Second aspect, provides a kind of recommendation apparatus of multimedia messages, and described device comprises:

First determination module, for clicking behavior according to first user to the history of multimedia messages, determines the interested first multimedia classification of described first user, classification belonging to the multimedia messages that described first multimedia classification is clicked for user's history;

Second determination module, for clicking behavior according to each the first multimedia classification of described first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of described first user, classification belonging to the multimedia messages that described second multimedia classification is not clicked for user's history;

3rd determination module, for according to the interested first multimedia classification of described first user and the potential interested second multimedia classification of described first user, determines destination multimedia classification;

Recommending module, for recommending the multimedia messages of described destination multimedia classification to described first user.

The beneficial effect that the technical scheme that the embodiment of the present invention provides is brought is:

By determining the interested first multimedia classification of first user, and determine the potential interested second multimedia classification that first user history is not clicked, thus according to the interested first multimedia classification of first user and the potential interested second classification determination destination multimedia classification of first user, make when recommending multimedia messages, except the multimedia classification belonging to the multimedia messages clicked according to first user history realizes, also realize according to the potential interested multimedia classification of first user, the destination multimedia classification determined is more accurate, thus make the multimedia messages of recommended destination multimedia classification fully to cause user to note, the clicking rate of the multimedia messages of recommendation can not only be improved, the multimedia messages recommended is made to have more specific aim, and can saving resource and financial resources.In addition, when destination multimedia classification comprises the potential interested arbitrary second multimedia classification of first user, because first user history does not click this second multimedia classification, therefore, the multimedia messages of recommendation has certain freshness to first user.

Accompanying drawing explanation

In order to be illustrated more clearly in the technical scheme in the embodiment of the present invention, below the accompanying drawing used required in describing embodiment is briefly described, apparently, accompanying drawing in the following describes is only some embodiments of the present invention, for those of ordinary skill in the art, under the prerequisite not paying creative work, other accompanying drawing can also be obtained according to these accompanying drawings.

Fig. 1 is the process flow diagram of the recommend method of a kind of multimedia messages that another embodiment of the present invention provides;

Fig. 2 is the process flow diagram of the recommend method of a kind of multimedia messages that another embodiment of the present invention provides;

Fig. 3 is a kind of process schematic determining destination multimedia classification that another embodiment of the present invention provides;

Fig. 4 is the structural representation of the recommendation apparatus of a kind of multimedia messages that another embodiment of the present invention provides;

Fig. 5 is the structural representation of a kind of server that another embodiment of the present invention provides.

Embodiment

For making the object, technical solutions and advantages of the present invention clearly, below in conjunction with accompanying drawing, embodiment of the present invention is described further in detail.

Fig. 1 is the process flow diagram of the recommend method of a kind of multimedia messages provided according to an exemplary embodiment.See Fig. 1, the method flow that the embodiment of the present invention provides comprises:

101, according to first user, behavior is clicked to the history of multimedia messages, determine the interested first multimedia classification of first user, wherein, classification belonging to the multimedia messages that the first multimedia classification is clicked for user's history.

102, behavior is clicked according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, wherein, classification belonging to the multimedia messages do not clicked for user's history of the second multimedia classification.

103, according to the interested first multimedia classification of first user and the potential interested second multimedia classification of first user, destination multimedia classification is determined.

104, the multimedia messages of destination multimedia classification is recommended to first user.

The method that the embodiment of the present invention provides, by determining the interested first multimedia classification of first user, and determine the potential interested second multimedia classification that first user history is not clicked, thus according to the interested first multimedia classification of first user and the potential interested second classification determination destination multimedia classification of first user, make when recommending multimedia messages, except the multimedia classification belonging to the multimedia messages clicked according to first user history realizes, also realize according to the potential interested multimedia classification of first user, the destination multimedia classification determined is more accurate, thus make the multimedia messages of recommended destination multimedia classification fully to cause user to note, the clicking rate of the multimedia messages of recommendation can not only be improved, the multimedia messages recommended is made to have more specific aim, and can saving resource and financial resources.In addition, when destination multimedia classification comprises the potential interested arbitrary second multimedia classification of first user, because first user history does not click this second multimedia classification, therefore, the multimedia messages of recommendation has certain freshness to first user.

In another embodiment, according to first user, behavior is clicked to the history of multimedia messages, determines the interested first multimedia classification of first user, comprising:

According to first user, behavior is clicked to the history of multimedia messages, determine each the first multimedia class object interest score value of first user;

According to each the first multimedia class object interest score value of first user, determine the interested first multimedia classification of first user;

Click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, comprising:

Click behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of first user;

According to each the second multimedia class object interest score value of first user, determine the potential interested second multimedia classification of first user;

According to the interested first multimedia classification of first user and the potential interested second multimedia classification of first user, determine destination multimedia classification, comprising:

According to the interested first multimedia class object interest score value of first user and the potential interested second multimedia class object interest score value of first user, determine the destination multimedia classification recommended to first user.

In another embodiment, according to first user, behavior is clicked to the history of multimedia messages, determines each the first multimedia class object interest score value of first user, comprising:

When using arbitrary first multimedia classification of first user as appointment first multimedia classification time, for appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number;

Click behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number;

Obtain and click appointment first multimedia class object number of users and all numbers of users;

According to first number, second number, number of users and all numbers of users, determine appointment first multimedia class object interest score value.

In another embodiment, according to first number, second number, number of users and all numbers of users, determine appointment first multimedia class object interest score value, comprising:

According to first number, second number, number of users and all numbers of users, determine appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_1=\frac{{\mathrm{CK}}_{i,j}}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{CK}}_{i,j}}$

${\mathrm{IDF}}_1=\log \frac{N}{N_j}$

S ₁＝TF ₁×IDF ₁

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, S ₁for specifying the interest score value of the first multimedia classification j.

In another embodiment, according to first user, behavior is clicked to the history of multimedia messages, determines each the first multimedia class object interest score value of first user, comprising:

When using arbitrary for first user the first multimedia classification as appointment first multimedia classification time, for appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number;

Click behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number;

Obtain and click appointment first multimedia class object number of users and all numbers of users;

Click the appointment first multimedia class object moment according to first user, determine appointment first multimedia class object weight;

According to first number, second number, number of users, all numbers of users and appointment the first multimedia class object weight, determine appointment first multimedia class object interest score value.

In another embodiment, click the appointment first multimedia class object moment according to first user, determine appointment first multimedia class object weight, comprising:

Click the appointment first multimedia class object moment according to first user, determine appointment first multimedia class object weight by following formula:

$w_j=\frac{1}{1+e^{-t}}$

In formula, t represents that first user clicks the difference between the moment of appointment first multimedia classification j and current time;

According to first number, second number, number of users, all numbers of users and appointment the first multimedia class object weight, determine appointment first multimedia class object interest score value, comprising:

According to first number, second number, number of users, all numbers of users and appointment the first multimedia class object weight, determine appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_2=\frac{{\mathrm{CK}}_{i,j}\×w_j}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{CK}}_{i,j}\×w_j)}$

${\mathrm{IDF}}_2=\log \frac{N}{N_j}$

S ₂＝TF ₂×IDF ₂

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, w _jfor specifying the weight of the first multimedia classification j, S ₂for specifying the interest score value of the first multimedia classification j.

In another embodiment, click behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of first user, comprising:

For arbitrary second user, the history according to the second user on multimedia information clicks behavior, determines each the first multimedia class object interest score value of the second user;

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, determine first user and the second user first degree of correlation when clicking multimedia messages;

From all second users, select target second user of the first numerical value, wherein, arbitrary target second user and first user first degree of correlation when clicking multimedia messages meets the first specified requirements;

When using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, according to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value.

In another embodiment, according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, determining first user and the second user first degree of correlation when clicking multimedia messages, comprising:

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, first degree of correlation when clicking multimedia messages by following formula determination first user and the second user:

$\mathrm{sim}(i,y)=\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}(r_{i,k}-\stackrel{\‾}{r_i})(r_{y,k}-\stackrel{\‾}{r_y})}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{i,k}-\stackrel{\‾}{r_i})}^2\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{y,k}-\stackrel{\‾}{r_y})}^2}}$

In formula, i is first user, y is arbitrary second user, m has multimedia class object quantity in the first multimedia classification of first user i with the first multimedia classification of the second user y, k is arbitrary multimedia classification total in the first multimedia classification of first user and the first multimedia classification of the second user, r _{i, k}for the interest score value of the first multimedia classification k of first user i, r _{y, k}be the interest score value of the first multimedia classification k of the second user y, for all first multimedia class object interest score value mean values of first user i, be all first multimedia class object interest score value mean values of the second user y;

According to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value, comprising:

According to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value by following formula:

$r_{i,h}=\stackrel{\‾}{r_i}+\frac{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})(r_{u^{\′},h}-\stackrel{\‾}{r_{u^{\′}}})}{\sqrt{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})}}$

In formula, U is the set of target second user, and u' is arbitrary target second user, and h is appointment second multimedia classification, r _i,hfor specifying the interest score value of the second multimedia classification h, sim (i, u') is first degree of correlation between first user i and target second user u', r _{u', h}for the interest score value of the first multimedia classification h of target second user u', for all first multimedia class object interest score value mean values of target second user u'.

In another embodiment, click behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of first user, comprising:

History according to each second user on multimedia information clicks behavior, determines each the first multimedia class object interest score value of each second user;

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications;

When using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, the target first multimedia classification of second value is selected from all first multimedia classifications of first user, wherein, second degree of correlation between arbitrary target first multimedia classification and appointment the second multimedia classification meets the second specified requirements;

According to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value.

In another embodiment, according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications, comprising:

For any two multimedia classifications, according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation in all multimedia classifications between any two multimedia classifications by following formula:

$\mathrm{sim}(a,b)=\frac{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}(r_{a,d}-\stackrel{\‾}{r_a})(r_{b,d}-\stackrel{\‾}{r_b})}{\sqrt{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{a,d}-\stackrel{\‾}{r_a})}^2\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{b,d}-\stackrel{\‾}{r_b})}^2}}$

In formula, a is arbitrary multimedia classification, and b is other arbitrary multimedia classification different from a, and N is all numbers of users, and d is arbitrary user, r _{a, d}for the interest score value of the multimedia classification a of user d, for the interest score value mean value of the multimedia classification a of all users, r _{b, d}for the interest score value of the multimedia classification b of user d, for the interest score value mean value of the multimedia classification b of all users;

According to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value, comprising:

According to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value by following formula:

$r_{i,f}=\stackrel{\‾}{r_f}+\frac{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})(r_{g^{\′},i}-\stackrel{\‾}{r_{g^{\′}}})}{\sqrt{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})}}$

In formula, f is appointment second multimedia classification, r _i,ffor specifying the interest score value of the second multimedia classification f, for all second users are to the interest score value mean value of appointment second multimedia classification f, G is target first multimedia class destination aggregation (mda), g' is arbitrary target first multimedia classification, sim (f, g') be second degree of correlation between target first multimedia classification g' and appointment the second multimedia classification f, r _{g', i}for the interest score value of the target first multimedia classification g' of first user i, for all second users are to the interest score value mean value of target first multimedia classification g'.

In another embodiment, according to first user, behavior is clicked to the history of multimedia messages, determines the interested first multimedia classification of first user, comprising:

According to first user, behavior is clicked to the history of multimedia messages, determine first user each the first multimedia class object number of clicks to first user;

According to first user each the first multimedia class object number of clicks to first user, determine the interested first multimedia classification of first user;

Click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, comprising:

Click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the third phase pass degree between any two multimedia classifications;

According to the third phase pass degree between any two multimedia classifications and the interested first multimedia classification of first user, determine the potential interested second multimedia classification of first user.

The content of embodiment corresponding to composition graphs 1, Fig. 2 is the process flow diagram of the recommend method of a kind of multimedia messages provided according to an exemplary embodiment.See Fig. 2, the method flow that the embodiment of the present invention provides comprises:

201, according to first user, behavior is clicked to the history of multimedia messages, determine the interested first multimedia classification of first user, wherein, classification belonging to the multimedia messages that the first multimedia classification is clicked for user's history.

First user is the object recommending multimedia messages, needs the multimedia messages recommended to be sent to the terminal of first user.Multimedia messages can be any type of information, if multimedia messages can be the combination etc. of a kind of in text message, image information, audio-frequency information, graph text information or video information or at least two kinds.Particularly, the form of multimedia messages includes but not limited to as advertisement etc.

The embodiment of the present invention is when recommending multimedia messages to first user, the history being excavated each user on multimedia information by the degree of depth clicks behavior, to determine the interested first multimedia classification of first user and the potential interested second multimedia classification of first user.Wherein, classification belonging to the multimedia messages that first multimedia classification of first user is clicked for first user history, therefore, when recommending multimedia messages, first according to first user, the interested first multimedia classification of behavior determination first user is clicked to the history of multimedia messages.For convenience of explanation, in embodiments of the present invention, the multimedia classification belonging to a certain bar multimedia messages first user history clicked is defined as a first multimedia classification of first user.Such as, if the multimedia classification belonging to a certain bar multimedia messages a that first user history has been clicked is multimedia classification A, then using the first multimedia classification of multimedia classification A as first user.

History is clicked behavior and is comprised user to the dominant feedback click behavior of the multimedia messages that history is shown and explicit feedback click behavior.Dominant feedback click behavior refers to the multimedia messages that first user is shown history, directly clicks and likes or do not like codominance to identify, carry out the whether interested feedback of multimedia messages.Stealthy feedback click behavior refers to the multimedia messages that first user is shown history, and by click or the behavior such as not click, recessiveness is carried out the whether interested feedback of multimedia messages.

Particularly, according to first user, behavior is clicked to the history of multimedia messages, determines the interested first multimedia classification of first user, included but not limited to following first kind of way and the second way two kinds of implementations:

First kind of way: by giving a mark to each the first multimedia classification of first user, to determine each first multimedia class object interest score value of first user, and according to all first multimedia class object interest score values of first user, determine the interested first multimedia classification of first user.Particularly, this first kind of way can as follows 2011 and step 2012 realize:

2011, according to first user, behavior is clicked to the history of multimedia messages, determine each the first multimedia class object interest score value of first user.

Particularly, according to first user, behavior is clicked to the history of multimedia messages, determines each the first multimedia class object interest score value of first user, included but not limited to following 1st kind and the 2nd kind of two kinds of implementations:

1st kind: realize in conjunction with TF-IDF (Term Frequency-Inverse Document Frequency, word frequency-reverse document-frequency) thought.For convenience of explanation, in embodiments of the present invention, the arbitrary first multimedia classification in each the first multimedia classification of first user is defined as appointment first multimedia classification.Under the 1st kind of mode, during arbitrary first multimedia class object interest score value in each the first multimedia classification determining first user, can realize by 2011A to step 2011D as follows:

2011A, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number.

Wherein, before first user recommendation information, the multimedia messages that first user history is clicked can be added up, and click each first multimedia class object first number based on this statistics determination first user.Particularly, for the multimedia messages that history is recommended, first user often pair multimedia messages carries out one click behavior, then determine once the multimedia classification belonging to this multimedia messages, and by number of clicks increase corresponding for this multimedia classification once.On this basis, according to historical statistical data, can determine that first user clicks this appointment first multimedia class object first number.

Such as, this appointment first multimedia classification is multimedia classification A, if before cut-off current time, added up obtaining first user and click the multimedia messages that 1000 belong to multimedia classification A, then first user clicks first number of this multimedia classification A is 1000 times.

2011B, click behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number.

Particularly, before recommendation information, all multimedia classifications can be pre-determined.In conjunction with the content in above-mentioned steps 2011A, for each first multimedia classification of first user, all perform statistics first user and click this multimedia class object first number, each first multimedia class object first number that first user history is clicked can be obtained.Each first multimedia class object first number that user clicks is superposed, first user can be obtained and click all multimedia class object number of times sums, i.e. second number.

Such as, if all multimedia classifications comprise multimedia classification A, multimedia classification B, multimedia classification C, multimedia classification D and multimedia classification E, and Corpus--based Method data obtain, first user clicks multimedia classification A1000 time, first user clicks multimedia classification B1500 time, first user clicks multimedia classification C1300 time, first user clicks multimedia classification E1200 time, then can determine that first user clicks all multimedia class objects second number is 5000.

Appointment first multimedia class object number of users and all numbers of users are clicked in 2011C, acquisition.

Clicking appointment first multimedia class object number of users is clicked the number of users sum belonging to this appointment first multimedia class object multimedia messages in history.Such as, if in history, have 50 users to click and belong to this appointment first multimedia class object multimedia messages, then clicking appointment first multimedia class object number of users is 50.Wherein, when obtaining click appointment first multimedia class object number of users, also can Corpus--based Method data realize.

All numbers of users refer to the quantity of the object of added up all recommendation informations.Wherein, when obtaining all numbers of users, include but not limited to determine according to registered users quantity.Such as, if existing 10000 users register at present, then all numbers of users are 10000.

2011D, according to first number, second number, number of users and all numbers of users, determine appointment first multimedia class object interest score value.

Particularly, according to first number, second number, number of users and all numbers of users, appointment first multimedia class object interest score value can be determined by following formula:

${\mathrm{TF}}_1=\frac{{\mathrm{CK}}_{i,j}}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{CK}}_{i,j}}$

${\mathrm{IDF}}_1=\log \frac{N}{N_j}$

S ₁＝TF ₁×IDF ₁

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, S ₁for specifying the interest score value of the first multimedia classification j.Wherein, the truth of a matter of log can be 2,10 or e in any one, the embodiment of the present invention does not do concrete restriction to this.

Above-mentioned steps 2011A is all carried out to step 2011D to each multimedia classification of first user, all first multimedia class object interest score values of first user can be determined.

2nd kind: the first multimedia classification that first user is clicked at different time, distribute different weights, and realized by TF-IDF mode.For convenience of explanation, in embodiments of the present invention, the arbitrary first multimedia classification in each the first multimedia classification of first user is defined as appointment first multimedia classification.Under the 2nd kind of mode, during arbitrary first multimedia class object interest score value in each the first multimedia classification determining first user, can realize by 2011a to step 2011e as follows:

2011a, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number.

2011b, click behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number.

Appointment first multimedia class object number of users and all numbers of users are clicked in 2011c, acquisition.

Wherein, the principle of step 2011a to step 2011c is consistent with the principle of above-mentioned steps 2011A to step 2011C, specifically see above-mentioned steps 2011A to the content in step 2011C, can repeat no more herein.

2011d, according to first user click the appointment first multimedia class object moment, determine appointment first multimedia class object weight.

Particularly, the appointment first multimedia class object moment can be clicked according to first user, determine appointment first multimedia class object weight by following formula:

$w_j=\frac{1}{1+e^{-t}}$

In formula, t represents that first user clicks the difference between the moment of appointment first multimedia classification j and current time.

It should be noted that, when determining the difference t that first user is clicked between appointment first multimedia class object moment and current time, needing the unit in conjunction with Preset Time window T to realize.

Such as, if the unit of Preset Time window T is month, then T=12, now, the difference that first user is clicked between appointment first multimedia class object moment and current time is: first user clicks the difference of specifying between multimedia class object month and current month.As, it is in January, 2015 that first user clicks the month of specifying multimedia classification j, and current month is in May, 2015, then t=4.

Again such as, if the unit of Preset Time window T is sky, then T=365, now, the difference that first user is clicked between appointment first multimedia class object moment and current time is: first user clicks the difference of specifying between multimedia class target date and current date.As, it is on January 1st, 2015 that first user clicks the date of specifying multimedia classification j, and current date is on March 1st, 2015, then t=59.

2011e, according to first number, second number, number of users, all numbers of users and specify the first multimedia class object weight, determine appointment first multimedia class object interest score value.

About according to first number, second number, number of users, all numbers of users and specify the first multimedia class object weight, determine the mode of appointment first multimedia class object interest score value, the embodiment of the present invention does not do concrete restriction.During concrete enforcement, include but not limited to be realized by following formula:

${\mathrm{TF}}_2=\frac{{\mathrm{CK}}_{i,j}\×w_j}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{CK}}_{i,j}\×w_j)}$

${\mathrm{IDF}}_2=\log \frac{N}{N_j}$

S ₂＝TF ₂×IDF ₂

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, w _jfor specifying the weight of the first multimedia classification j, S ₂for specifying the interest score value of the first multimedia classification j.

Above-mentioned steps 2011a is all carried out to step 2011e to each first multimedia classification of first user, all first multimedia class object interest score values of first user can be determined.

During each the first multimedia class object interest score value by the 2nd kind of mode determination first user, can click for the first multimedia class object moment according to first user, the first multimedia classification clicked for the different moment distributes different weights.Therefore, during each the first multimedia class object interest score value by the 2nd kind of first user that mode is determined, the change that the interested first multimedia classification of first user moves in time can be analyzed, make the interested first multimedia classification of the follow-up first user determined more accurate.

2012, according to each the first multimedia class object interest score value of first user, the interested first multimedia classification of first user is determined.

Particularly, an interested first multimedia classification of first user can be determined, also can determine the interested first multimedia classification of multiple first user.About the interested first multimedia class object quantity of the first user determined, the embodiment of the present invention does not do concrete restriction.

Wherein, at each the first multimedia class object interest score value according to first user, when determining the interested first multimedia classification of first user, can according to each the first multimedia class object interest score value of first user, all first multimedia class object interest score values of first user are sorted, and according to the interested first multimedia classification of ranking results determination first user.

When first multimedia classification interested according to ranking results determination first user, interested first multimedia class object quantity can realize in conjunction with default first user.Particularly, according to ranking results, from all first multimedia classifications of first user, the multimedia classification of the high predetermined number of interest score value can be filtered out as the interested first multimedia classification of first user.

The second way: according to first user to the interested first multimedia classification of multimedia class object number of clicks determination first user.

Under this second way, behavior can be clicked according to first user to the history of multimedia messages, determine first user each the first multimedia class object number of clicks to first user; According to first user each the first multimedia class object number of clicks to first user, determine the interested first multimedia classification of first user.

Particularly, first user often clicks a multimedia messages, then determine once the multimedia classification belonging to this multimedia messages, and by number of clicks increase corresponding for this multimedia classification once.All first multimedia classifications for first user all perform this operation, get final product Corpus--based Method data, determine first user each the first multimedia class object number of clicks to first user.

In addition, according to first user each the first multimedia class object number of clicks to first user, when determining the interested first multimedia classification of first user, can sort to all first multimedia class object numbers of clicks of first user, and according to the interested first multimedia classification of ranking results determination first user.

Particularly, when first multimedia classification interested according to ranking results determination first user, interested first multimedia class object quantity can realize in conjunction with default first user.Particularly, according to ranking results, from all first multimedia classifications of first user, the multimedia classification of the many predetermined numbers of number of clicks can be filtered out as the interested first multimedia classification of first user.

202, behavior is clicked according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, wherein, classification belonging to the multimedia messages do not clicked for user's history of the second multimedia classification.

Wherein, the second user is other user beyond first user.Second multimedia classification of first user is not clicked for first user history and the classification belonging to multimedia messages clicked at least one second user's history.

In order to realize accurately to first user recommendation information, the embodiment of the present invention, except determined the interested first multimedia classification of first user by above-mentioned steps 201 except, also continues to determine the potential interested second multimedia classification of first user.Wherein, the potential interested second multimedia classification of first user may be clicked relevant with the history of each the first multimedia classification of first user and multiple second user on multimedia information, therefore, need to click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user.

Particularly, behavior is being clicked according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, when determining the potential interested second multimedia classification of first user, include but not limited to following first kind of way and the second way two kinds of implementations:

First kind of way: by giving a mark to each the second multimedia classification of first user, to determine each second multimedia class object interest score value of first user, and according to all second multimedia class object interest score values of first user, determine the potential interested second multimedia classification of first user.Particularly, this first kind of way can as follows 2021 and step 2022 realize:

2021, click behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of first user.

Particularly, behavior is being clicked according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, when determining each the second multimedia class object interest score value of first user, include but not limited to following 1st kind and the 2nd kind of two kinds of implementations:

1st kind: realized by the degree of correlation of different user when clicking multimedia messages.Particularly, when clicking multimedia messages, may there is certain degree of correlation between multimedia classification belonging to the multimedia messages clicked in different user.As shown in table 1, it illustrates a kind of different user and click multimedia class object signal table.

Table 1

	A	B	C	D
					U1	4	2	6	3
U2	3	？	2	1
					U3	2	1	2	2

Wherein, the row in table 1 represents each multimedia classification, and each user is shown in list, and each element can represent number of clicks, also can represent interest score value, "? " represent element to be inferred.Associative list 1, if U2 and U3 is when clicking multimedia classification, number of times or the interest score value of clicking multimedia classification A are more close, and number of times or the interest score value of clicking multimedia classification C are more close, then can infer that user U2 and U3 also has certain similarity when clicking multimedia classification B.In such cases, if known users U3 clicks number of times or the interest score value of multimedia classification B, then can infer that user U2 clicks number of times or the interest score value of multimedia classification B.

In conjunction with foregoing, under the 1st kind of mode, each the second multimedia class object interest score value of first user can be determined by 2021A to step 2021D as follows:

2021A, for arbitrary second user, click behavior according to the history of the second user on multimedia information, determine each the first multimedia class object interest score value of the second user.

Wherein, history according to the second user on multimedia information clicks behavior, determine each the first multimedia class object interest score value of the second user mode can with above-mentioned steps 2011 according to first user, behavior is clicked to the history of multimedia messages, determine that the mode of each the first multimedia class object interest score value of first user is identical, during concrete enforcement, behavior can be clicked according to first user to the history of multimedia messages see in above-mentioned steps 2011, determine the content in each the first multimedia class object interest score value of first user, repeat no more herein.

For all second with performing this step 2021A per family, each the first multimedia class object interest score value of all second users can be determined.

2021B, according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, determine first user and the second user first degree of correlation when clicking multimedia messages.

Particularly, first can determine the first multimedia classification common between each the first multimedia classification of first user and each the first multimedia classification of each second user, and based on this first total multimedia classification, determine first user and the second user first degree of correlation when clicking multimedia messages.Particularly, can according to each the first multimedia class object interest score value of each the first multimedia class object interest score value of first user and the second user, first degree of correlation when clicking multimedia messages by following formula determination first user and the second user:

$\mathrm{sim}(i,y)=\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}(r_{i,k}-\stackrel{\‾}{r_i})(r_{y,k}-\stackrel{\‾}{r_y})}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{i,k}-\stackrel{\‾}{r_i})}^2\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{y,k}-\stackrel{\‾}{r_y})}^2}}$

In formula, i is first user, y is arbitrary second user, m has multimedia class object quantity in the first multimedia classification of first user i with the first multimedia classification of the second user y, k is arbitrary multimedia classification total in the first multimedia classification of first user and the first multimedia classification of the second user, r _{i, k}for the interest score value of the first multimedia classification k of first user i, r _{y, k}be the interest score value of the first multimedia classification k of the second user y, for all first multimedia class object interest score value mean values of first user i, be all first multimedia class object interest score value mean values of the second user y.

It should be noted that, only list a kind of mode determining first user and the second user first degree of correlation when clicking multimedia messages herein, but, in the specific implementation, can also by alternate manner determination first user and the second user first degree of correlation when clicking multimedia messages, as, can adopt cos (included angle cosine) formula etc., the embodiment of the present invention is to determining that the mode of first user and the second user first degree of correlation when clicking multimedia messages limits.

2021C, from all second users, select target second user of the first numerical value, wherein, arbitrary target second user and first user first degree of correlation when clicking multimedia messages meets the first specified requirements.

About the concrete numerical value of the first numerical value, the embodiment of the present invention does not do concrete restriction.During concrete enforcement.Can set as required.In addition, the embodiment of the present invention does not limit the content of the first specified requirements equally, and particularly, the content of the first specified requirements can be relevant to the first numerical value.Such as, when the first numerical value is 10, target second user can be in all second users, second user of 10 before coming with first degree of correlation of first user when clicking multimedia messages.

2021D, when using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, according to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value.

For convenience of explanation, arbitrary second multimedia classification of first user is defined as appointment second multimedia classification by the embodiment of the present invention.For arbitrary appointment second multimedia classification, according to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, this appointment second multimedia class object interest score value can be determined by following formula:

$r_{i,h}=\stackrel{\‾}{r_i}+\frac{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})(r_{u^{\′},h}-\stackrel{\‾}{r_{u^{\′}}})}{\sqrt{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})}}$

In formula, U is the set of target second user, and u' is arbitrary target second user, and h is appointment second multimedia classification, r _i,hfor specifying the interest score value of the second multimedia classification h, sim (i, u') is first degree of correlation between first user i and target second user u', r _{u', h}for the interest score value of the first multimedia classification h of target second user u', for all first multimedia class object interest score value mean values of target second user u'.

All above-mentioned steps 2021D is carried out to each multimedia classification of first user, all second multimedia class object interest score values of first user can be determined.

2nd kind: realized by the degree of correlation between multimedia classification.Particularly, between different multimedia classification, certain degree of correlation may be there is, between mobile phone classification and Cellphone Accessories classification, there is certain degree of correlation, finishing classification may and household electrical appliance between there is certain degree of correlation.Therefore, each the second multimedia class object interest score value of first user can be determined by the 2nd kind of mode.

Particularly, under the 2nd kind of mode, each the second multimedia class object interest score value of first user can be determined by 2021a to step 2021d as follows:

2021a, click behavior according to the history of each second user on multimedia information, determine each the first multimedia class object interest score value of each second user.

Wherein, history according to the second user on multimedia information clicks behavior, determine each the first multimedia class object interest score value of the second user mode can with above-mentioned steps 2011 according to first user, behavior is clicked to the history of multimedia messages, determine that the mode of each the first multimedia class object interest score value of first user is identical, during concrete enforcement, behavior can be clicked according to first user to the history of multimedia messages see in above-mentioned steps 2011, determine the content in each the first multimedia class object interest score value of first user, repeat no more herein.

2021b, according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications.

That is, this step, by each the first multimedia classification of all users, determines second degree of correlation between any two multimedia classifications.

Particularly, according to each the first multimedia class object interest score value of each the first multimedia class object interest score value of first user and all second users, second degree of correlation in all multimedia classifications between any two multimedia classifications can be determined by following formula:

$\mathrm{sim}(a,b)=\frac{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}(r_{a,d}-\stackrel{\‾}{r_a})(r_{b,d}-\stackrel{\‾}{r_b})}{\sqrt{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{a,d}-\stackrel{\‾}{r_a})}^2\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{b,d}-\stackrel{\‾}{r_b})}^2}}$

In formula, a is arbitrary multimedia classification, and b is other arbitrary multimedia classification different from a, and N is all numbers of users, and d is arbitrary user, r _{a, d}for the interest score value of the multimedia classification a of user d, for the interest score value mean value of the multimedia classification a of all users, r _{b, d}for the interest score value of the multimedia classification b of user d, for the interest score value mean value of the multimedia classification b of all users.

It should be noted that, only list a kind of mode determining second degree of correlation between any two multimedia classifications herein, but, in the specific implementation, second degree of correlation can also determining between any two multimedia classifications by alternate manner, as, can adopt cos (included angle cosine) formula etc., the embodiment of the present invention does not limit the mode of second degree of correlation determined between any two multimedia classifications.

This step 2021b is all performed for all multimedia classifications, second degree of correlation between any two multimedia classifications can be obtained.

2021c, when using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, the target first multimedia classification of second value is selected from all first multimedia classifications of first user, wherein, second degree of correlation between arbitrary target first multimedia classification and appointment the second multimedia classification meets the second specified requirements.

About the concrete numerical value of second value, the embodiment of the present invention does not do concrete restriction.During concrete enforcement.Can set as required.In addition, the embodiment of the present invention does not limit the content of the second specified requirements equally, and particularly, the content of the second specified requirements can be relevant to second value.Such as, when second value is 5, target first multimedia classification can be in all multimedia classifications, comes the first multimedia classification of the first user of front 5 with this appointment second multimedia class object second degree of correlation.

2021d, according to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value.

Particularly, and second degree of correlation between the second multimedia classification and all second users can be specified appointment second multimedia class object interest score value according to each target first multimedia class object interest score value, each target first multimedia classification, determine appointment second multimedia class object interest score value by following formula:

$r_{i,f}=\stackrel{\‾}{r_f}+\frac{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})(r_{g^{\′},i}-\stackrel{\‾}{r_{g^{\′}}})}{\sqrt{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})}}$

In formula, f is appointment second multimedia classification, r _i,ffor specifying the interest score value of the second multimedia classification f, for all second users are to the interest score value mean value of appointment second multimedia classification f, G is target first multimedia class destination aggregation (mda), g' is arbitrary target first multimedia classification, sim (f, g') be second degree of correlation between target first multimedia classification g' and appointment the second multimedia classification f, r _{g', i}for the interest score value of the target first multimedia classification g' of first user i, for all second users are to the interest score value mean value of target first multimedia classification g'.

2022, according to each the second multimedia class object interest score value of first user, the potential interested second multimedia classification of first user is determined.

Particularly, a potential interested first multimedia classification of first user can be determined, also can determine the potential interested first multimedia classification of multiple first user.About the potential interested first multimedia class object quantity of the first user determined, the embodiment of the present invention does not do concrete restriction.

Wherein, at each the first multimedia class object interest score value according to first user, when determining the potential interested first multimedia classification of first user, can according to each the second multimedia class object interest score value of first user, all second multimedia class object interest score values of first user are sorted, and according to the potential interested second multimedia classification of ranking results determination first user.

When interested second multimedia classification potential in ranking results determination first user, potential interested second multimedia class object quantity can realize in conjunction with default first user.Particularly, according to ranking results, from all second multimedia classifications of first user, the multimedia classification of the high predetermined number of interest score value can be filtered out as the potential interested second multimedia classification of first user.

The second way: click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the third phase pass degree between any two multimedia classifications; According to the third phase pass degree between any two multimedia classifications and the interested first multimedia classification of first user, determine the potential interested second multimedia classification of first user.

Wherein, behavior is clicked according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the principle of the third phase pass degree between any two multimedia classifications, can with above-mentioned steps 2021a and step 2021b according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine that the principle of second degree of correlation in all multimedia classifications between any two multimedia classifications is consistent, during concrete enforcement, can see the content in above-mentioned steps 2021a and step 2021b, repeat no more herein.

In addition, according to the third phase pass degree between any two multimedia classifications and the interested first multimedia classification of first user, when determining the potential interested second multimedia classification of first user, can after determining the interested first multimedia classification of first user, search the multimedia classification of the maximum third value of third phase pass degree between the interested first multimedia classification of this first user, using these multimedia classifications as the potential interested second multimedia classification of first user.

203, according to the interested first multimedia classification of first user and the potential interested second multimedia classification of first user, destination multimedia classification is determined.

Wherein, destination multimedia classification can be the first multimedia classification of certain numerical value, also can be the first multimedia classification of certain numerical value, can also comprise the first multimedia classification and the second multimedia classification simultaneously.

Determine the first kind of way in the interested first multimedia classification of first user in conjunction with above-mentioned and determine the first kind of way in the potential interested second multimedia classification of first user, according to the interested first multimedia classification of first user and the potential interested second multimedia classification of first user, when determining destination multimedia classification, according to the interested first multimedia class object interest score value of first user and the potential interested second multimedia class object interest score value of first user, the destination multimedia classification recommended to first user can be determined.

In conjunction with above-mentioned steps 201 and the step 202 pair embodiment of the present invention when to first user recommendation information, determine that the interested first multimedia classification of first user and potential interested second multimedia class object process have carried out detailed explanation.As shown in Figure 3, it illustrates a kind of process schematic determining destination multimedia classification.

204, when detecting that first user logs in, the multimedia messages of destination multimedia classification is recommended to first user.

In order to recommend the multimedia messages of destination multimedia classification in time to first user, when detecting that first user logs in, the multimedia messages of this destination multimedia classification can be recommended to first user.Particularly, when detecting that first user logs in, push the multimedia messages of this destination multimedia classification to first user, and the multimedia messages of this recommendation is shown in the appointed area of first user terminal screen.Wherein, this appointed area can be the top, right etc. of screen.In addition, the multimedia messages of destination multimedia classification forever can be shown in this appointed area, also can every the multimedia messages of specifying duration to upgrade once shown destination multimedia classification.In addition, the multimedia messages etc. of this destination multimedia classification can also be shown by playing window mode.

Wherein, when detecting first user and whether logging in, whether can have input account and password at login interface by detecting first user, and the account submitted to when registering with first user of the account inputted and password and password mate and realize.When detecting that first user have input account and password at login interface, and the account submitted to when the account of input and password are registered with first user and password mate, then determine that first user logs in.

Certainly, herein only to be illustrated to its recommendation information when first user logs in, but, in the specific implementation, other scene can also be had to first user recommendation information.Such as, when detecting that first user browses webpage, or when the relevant information etc. of first user search destination multimedia classification be detected, the scene of the embodiment of the present invention not subtend first user recommendation information limits.

The method that the embodiment of the present invention provides, by determining the interested first multimedia classification of first user, and determine the potential interested second multimedia classification that first user history is not clicked, thus according to the interested first multimedia classification of first user and the potential interested second classification determination destination multimedia classification of first user, make when recommending multimedia messages, except the multimedia classification belonging to the multimedia messages clicked according to first user history realizes, also realize according to the potential interested multimedia classification of first user, the destination multimedia classification determined is more accurate, thus make the multimedia messages of recommended destination multimedia classification fully to cause user to note, the clicking rate of the multimedia messages of recommendation can not only be improved, the multimedia messages recommended is made to have more specific aim, and can saving resource and financial resources.In addition, when destination multimedia classification comprises the potential interested arbitrary second multimedia classification of first user, because first user history does not click this second multimedia classification, therefore, the multimedia messages of recommendation has certain freshness to first user.

Fig. 4 is the structural representation of the recommendation apparatus of a kind of multimedia messages provided according to an exemplary embodiment, and the recommendation apparatus of this multimedia messages may be used for the recommend method performing the multimedia messages that embodiment provides corresponding to above-mentioned Fig. 1 or Fig. 2.See Fig. 4, the recommendation apparatus of this multimedia messages comprises:

First determination module 401, for clicking behavior according to first user to the history of multimedia messages, determines the interested first multimedia classification of first user, wherein, and classification belonging to the multimedia messages that the first multimedia classification is clicked for user's history;

Second determination module 402, for clicking behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, wherein, classification belonging to the multimedia messages do not clicked for user's history of the second multimedia classification;

3rd determination module 403, for according to the interested first multimedia classification of first user and the potential interested second multimedia classification of first user, determines destination multimedia classification;

Recommending module 404, for recommending the multimedia messages of destination multimedia classification to first user.

The device that the embodiment of the present invention provides, by determining the interested first multimedia classification of first user, and determine the potential interested second multimedia classification that first user history is not clicked, thus according to the interested first multimedia classification of first user and the potential interested second classification determination destination multimedia classification of first user, make when recommending multimedia messages, except the multimedia classification belonging to the multimedia messages clicked according to first user history realizes, also realize according to the potential interested multimedia classification of first user, the destination multimedia classification determined is more accurate, thus make the multimedia messages of recommended destination multimedia classification fully to cause user to note, the clicking rate of the multimedia messages of recommendation can not only be improved, the multimedia messages recommended is made to have more specific aim, and can saving resource and financial resources.In addition, when destination multimedia classification comprises the potential interested arbitrary second multimedia classification of first user, because first user history does not click this second multimedia classification, therefore, the multimedia messages of recommendation has certain freshness to first user.

In another embodiment, the first determination module 401 comprises:

First determining unit, for clicking behavior according to first user to the history of multimedia messages, determines each the first multimedia class object interest score value of first user;

Second determining unit, for each the first multimedia class object interest score value according to first user, determines the interested first multimedia classification of first user;

Second determination module 402 comprises:

3rd determining unit, for clicking behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determines each the second multimedia class object interest score value of first user;

4th determining unit, for each the second multimedia class object interest score value according to first user, determines the potential interested second multimedia classification of first user;

3rd determination module 403, for according to the interested first multimedia class object interest score value of first user and the potential interested second multimedia class object interest score value of first user, determines the destination multimedia classification recommended to first user.

In another embodiment, the first determining unit comprises:

First determines subelement, for when using arbitrary first multimedia classification of first user as appointment first multimedia classification time, for appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number;

Second determines subelement, and for clicking behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number;

First obtains subelement, clicks appointment first multimedia class object number of users and all numbers of users for obtaining;

3rd determines subelement, for according to first number, second number, number of users and all numbers of users, determines appointment first multimedia class object interest score value.

In another embodiment, the 3rd determines subelement, also for according to first number, second number, number of users and all numbers of users, determines appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_1=\frac{{\mathrm{CK}}_{i,j}}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{CK}}_{i,j}}$

${\mathrm{IDF}}_1=\log \frac{N}{N_j}$

S ₁＝TF ₁×IDF ₁

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, S ₁for specifying the interest score value of the first multimedia classification j.

In another embodiment, the first determining unit comprises:

4th determines subelement, for when using arbitrary for first user the first multimedia classification as appointment first multimedia classification time, for appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number;

5th determines subelement, and for clicking behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number;

Second obtains subelement, clicks appointment first multimedia class object number of users and all numbers of users for obtaining;

6th determines subelement, for clicking the appointment first multimedia class object moment according to first user, determines appointment first multimedia class object weight;

7th determines subelement, for according to first number, second number, number of users, all numbers of users and appointment the first multimedia class object weight, determines appointment first multimedia class object interest score value.

In another embodiment, the 6th determines subelement, also for clicking the appointment first multimedia class object moment according to first user, determines appointment first multimedia class object weight by following formula:

$w_j=\frac{1}{1+e^{-t}}$

In formula, t represents that first user clicks the difference between the moment of appointment first multimedia classification j and current time;

7th determines subelement, also for according to first number, second number, number of users, all numbers of users and specify the first multimedia class object weight, determines appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_2=\frac{{\mathrm{CK}}_{i,j}\×w_j}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{CK}}_{i,j}\×w_j)}$

${\mathrm{IDF}}_2=\log \frac{N}{N_j}$

S ₂＝TF ₂×IDF ₂

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, w _jfor specifying the weight of the first multimedia classification j, S ₂for specifying the interest score value of the first multimedia classification j.

In another embodiment, the 3rd determining unit comprises:

8th determines subelement, for for arbitrary second user, according to the history click behavior of the second user on multimedia information, determines each the first multimedia class object interest score value of the second user;

9th determines subelement, for according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, determines first user and the second user first degree of correlation when clicking multimedia messages;

First chooser unit, for selecting target second user of the first numerical value from all second users, wherein, arbitrary target second user and first user first degree of correlation when clicking multimedia messages meets the first specified requirements;

Tenth determines subelement, for when using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, according to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value.

In another embodiment, 9th determines subelement, also for according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, first degree of correlation when clicking multimedia messages by following formula determination first user and the second user:

$\mathrm{sim}(i,y)=\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}(r_{i,k}-\stackrel{\‾}{r_i})(r_{y,k}-\stackrel{\‾}{r_y})}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{i,k}-\stackrel{\‾}{r_i})}^2\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{y,k}-\stackrel{\‾}{r_y})}^2}}$

In formula, i is first user, y is arbitrary second user, m has multimedia class object quantity in the first multimedia classification of first user i with the first multimedia classification of the second user y, k is arbitrary multimedia classification total in the first multimedia classification of first user and the first multimedia classification of the second user, r _{i, k}for the interest score value of the first multimedia classification k of first user i, r _{y, k}be the interest score value of the first multimedia classification k of the second user y, for all first multimedia class object interest score value mean values of first user i, be all first multimedia class object interest score value mean values of the second user y;

Tenth determines subelement, also for each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user according to first user, determine appointment second multimedia class object interest score value by following formula:

$r_{i,h}=\stackrel{\‾}{r_i}+\frac{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})(r_{u^{\′},h}-\stackrel{\‾}{r_{u^{\′}}})}{\sqrt{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})}}$

In formula, U is the set of target second user, and u' is arbitrary target second user, and h is appointment second multimedia classification, r _i,hfor specifying the interest score value of the second multimedia classification h, sim (i, u') is first degree of correlation between first user i and target second user u', r _{u', h}for the interest score value of the first multimedia classification h of target second user u', for all first multimedia class object interest score value mean values of target second user u'.

In another embodiment, the 3rd determining unit comprises:

11 determines subelement, clicks behavior, determine each the first multimedia class object interest score value of each second user for the history according to each second user on multimedia information;

12 determines subelement, for according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications;

Second chooser unit, for when using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, the target first multimedia classification of second value is selected from all first multimedia classifications of first user, wherein, second degree of correlation between arbitrary target first multimedia classification and appointment the second multimedia classification meets the second specified requirements;

13 determines subelement, for according to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value.

In another embodiment, 12 determines subelement, also for for any two multimedia classifications, according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation in all multimedia classifications between any two multimedia classifications by following formula:

$\mathrm{sim}(a,b)=\frac{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}(r_{a,d}-\stackrel{\‾}{r_a})(r_{b,d}-\stackrel{\‾}{r_b})}{\sqrt{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{a,d}-\stackrel{\‾}{r_a})}^2\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{b,d}-\stackrel{\‾}{r_b})}^2}}$

In formula, a is arbitrary multimedia classification, and b is other arbitrary multimedia classification different from a, and N is all numbers of users, and d is arbitrary user, r _{a, d}for the interest score value of the multimedia classification a of user d, for the interest score value mean value of the multimedia classification a of all users, r _{b, d}for the interest score value of the multimedia classification b of user d, for the interest score value mean value of the multimedia classification b of all users;

13 determines subelement, also for according to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value by following formula:

$r_{i,f}=\stackrel{\‾}{r_f}+\frac{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})(r_{g^{\′},i}-\stackrel{\‾}{r_{g^{\′}}})}{\sqrt{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})}}$

In formula, f is appointment second multimedia classification, r _i,ffor specifying the interest score value of the second multimedia classification f, for all second users are to the interest score value mean value of appointment second multimedia classification f, G is target first multimedia class destination aggregation (mda), g' is arbitrary target first multimedia classification, sim (f, g') be second degree of correlation between target first multimedia classification g' and appointment the second multimedia classification f, r _{g', i}for the interest score value of the target first multimedia classification g' of first user i, for all second users are to the interest score value mean value of target first multimedia classification g'.

In another embodiment, the first determination module 401, also for clicking behavior according to first user to the history of multimedia messages, determines first user each the first multimedia class object number of clicks to first user; According to first user each the first multimedia class object number of clicks to first user, determine the interested first multimedia classification of first user;

Second determination module 402, also for clicking behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determines the third phase pass degree between any two multimedia classifications; According to the third phase pass degree between any two multimedia classifications and the interested first multimedia classification of first user, determine the potential interested second multimedia classification of first user.

Fig. 5 is the structural representation of a kind of server according to an exemplary embodiment, and this server may be used for the recommend method performing the multimedia messages that embodiment provides corresponding to above-mentioned Fig. 1 or Fig. 2.With reference to Fig. 5, server 500 comprises processing components 522, and it comprises one or more processor further, and the memory resource representated by storer 532, can such as, by the instruction of the execution of processing components 522, application program for storing.The application program stored in storer 532 can comprise each module corresponding to one group of instruction one or more.In addition, processing components 522 is configured to perform instruction, to perform the recommend method of the multimedia messages that embodiment provides corresponding to above-mentioned Fig. 1 or Fig. 2.

Server 500 can also comprise the power management that a power supply module 526 is configured to perform server 500, and a wired or wireless network interface 550 is configured to server 500 to be connected to network, and input and output (I/O) interface 558.Server 500 can operate the operating system based on being stored in storer 532, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or similar.

Wherein, more than one or one program is stored in storer, and is configured to be performed by more than one or one processor, and described more than one or one routine package is containing the instruction for carrying out following operation:

According to first user, behavior is clicked to the history of multimedia messages, determine the interested first multimedia classification of first user, wherein, classification belonging to the multimedia messages that the first multimedia classification is clicked for user's history;

Behavior is clicked according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, wherein, classification belonging to the multimedia messages do not clicked for user's history of the second multimedia classification;

According to the interested first multimedia classification of first user and the potential interested second multimedia classification of first user, determine destination multimedia classification;

The multimedia messages of destination multimedia classification is recommended to first user.

Suppose that above-mentioned is the first possible embodiment, then, in the embodiment that the second provided based on the embodiment that the first is possible is possible, in the storer of server, also comprise the instruction for performing following operation:

According to first user, behavior is clicked to the history of multimedia messages, determines the interested first multimedia classification of first user, comprising:

According to first user, behavior is clicked to the history of multimedia messages, determine each the first multimedia class object interest score value of first user;

According to each the first multimedia class object interest score value of first user, determine the interested first multimedia classification of first user;

Click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, comprising:

Click behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of first user;

According to each the second multimedia class object interest score value of first user, determine the potential interested second multimedia classification of first user;

According to the interested first multimedia classification of first user and the potential interested second multimedia classification of first user, determine destination multimedia classification, comprising:

According to the interested first multimedia class object interest score value of first user and the potential interested second multimedia class object interest score value of first user, determine the destination multimedia classification recommended to first user.

In the third the possible embodiment provided based on the embodiment that the second is possible, in the storer of server, also comprise the instruction for performing following operation:

According to first user, behavior is clicked to the history of multimedia messages, determines each the first multimedia class object interest score value of first user, comprising:

When using arbitrary first multimedia classification of first user as appointment first multimedia classification time, for appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number;

Click behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number;

Obtain and click appointment first multimedia class object number of users and all numbers of users;

According to first number, second number, number of users and all numbers of users, determine appointment first multimedia class object interest score value.

In the 4th kind of possible embodiment provided based on the embodiment that the third is possible, in the storer of server, also comprise the instruction for performing following operation: according to first number, second number, number of users and all numbers of users, determine appointment first multimedia class object interest score value, comprising:

According to first number, second number, number of users and all numbers of users, determine appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_1=\frac{{\mathrm{CK}}_{i,j}}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{CK}}_{i,j}}$

${\mathrm{IDF}}_1=\log \frac{N}{N_j}$

S ₁＝TF ₁×IDF ₁

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, S ₁for specifying the interest score value of the first multimedia classification j.

In the 5th kind of possible embodiment provided based on the embodiment that the second is possible, in the storer of server, also comprise the instruction for performing following operation:

According to first user, behavior is clicked to the history of multimedia messages, determines each the first multimedia class object interest score value of first user, comprising:

When using arbitrary for first user the first multimedia classification as appointment first multimedia classification time, for appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that first user clicks appointment first multimedia class object number of times is first number;

Click behavior according to first user to the history of multimedia messages, determining that first user clicks all multimedia class object number of times sums is second number;

Obtain and click appointment first multimedia class object number of users and all numbers of users;

Click the appointment first multimedia class object moment according to first user, determine appointment first multimedia class object weight;

According to first number, second number, number of users, all numbers of users and appointment the first multimedia class object weight, determine appointment first multimedia class object interest score value.

In the 6th kind of possible embodiment provided based on the 5th kind of possible embodiment, in the storer of server, also comprise the instruction for performing following operation: click the appointment first multimedia class object moment according to first user, determine appointment first multimedia class object weight, comprising:

Click the appointment first multimedia class object moment according to first user, determine appointment first multimedia class object weight by following formula:

$w_j=\frac{1}{1+e^{-t}}$

In formula, t represents that first user clicks the difference between the moment of appointment first multimedia classification j and current time;

According to first number, second number, number of users, all numbers of users and appointment the first multimedia class object weight, determine appointment first multimedia class object interest score value, comprising:

According to first number, second number, number of users, all numbers of users and appointment the first multimedia class object weight, determine appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_2=\frac{{\mathrm{CK}}_{i,j}\×w_j}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{CK}}_{i,j}\×w_j)}$

${\mathrm{IDF}}_2=\log \frac{N}{N_j}$

S ₂＝TF ₂×IDF ₂

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, w _jfor specifying the weight of the first multimedia classification j, S ₂for specifying the interest score value of the first multimedia classification j.

In the 7th kind of possible embodiment provided based on the embodiment that the second is possible, in the storer of server, also comprise the instruction for performing following operation:

Click behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of first user, comprising:

For arbitrary second user, the history according to the second user on multimedia information clicks behavior, determines each the first multimedia class object interest score value of the second user;

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, determine first user and the second user first degree of correlation when clicking multimedia messages;

From all second users, select target second user of the first numerical value, wherein, arbitrary target second user and first user first degree of correlation when clicking multimedia messages meets the first specified requirements;

When using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, according to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value.

In the 8th kind of possible embodiment provided based on the 7th kind of possible embodiment, in the storer of server, also comprise the instruction for performing following operation:

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, determining first user and the second user first degree of correlation when clicking multimedia messages, comprising:

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of the second user, first degree of correlation when clicking multimedia messages by following formula determination first user and the second user:

$\mathrm{sim}(i,y)=\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}(r_{i,k}-\stackrel{\‾}{r_i})(r_{y,k}-\stackrel{\‾}{r_y})}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{i,k}-\stackrel{\‾}{r_i})}^2\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{y,k}-\stackrel{\‾}{r_y})}^2}}$

In formula, i is first user, y is arbitrary second user, m has multimedia class object quantity in the first multimedia classification of first user i with the first multimedia classification of the second user y, k is arbitrary multimedia classification total in the first multimedia classification of first user and the first multimedia classification of the second user, r _{i, k}for the interest score value of the first multimedia classification k of first user i, r _{y, k}be the interest score value of the first multimedia classification k of the second user y, for all first multimedia class object interest score value mean values of first user i, be all first multimedia class object interest score value mean values of the second user y;

According to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value, comprising:

According to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between first user and each target second user of first user, determine appointment second multimedia class object interest score value by following formula:

$r_{i,h}=\stackrel{\‾}{r_i}+\frac{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})(r_{u^{\′},h}-\stackrel{\‾}{r_{u^{\′}}})}{\sqrt{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})}}$

In formula, U is the set of target second user, and u' is arbitrary target second user, and h is appointment second multimedia classification, r _i,hfor specifying the interest score value of the second multimedia classification h, sim (i, u') is first degree of correlation between first user i and target second user u', r _{u', h}for the interest score value of the first multimedia classification h of target second user u', for all first multimedia class object interest score value mean values of target second user u'.

In the 9th kind of possible embodiment provided based on the embodiment that the second is possible, in the storer of server, also comprise the instruction for performing following operation:

Click behavior according to each the first multimedia class object interest score value of first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of first user, comprising:

History according to each second user on multimedia information clicks behavior, determines each the first multimedia class object interest score value of each second user;

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications;

When using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for appointment second multimedia classification, the target first multimedia classification of second value is selected from all first multimedia classifications of first user, wherein, second degree of correlation between arbitrary target first multimedia classification and appointment the second multimedia classification meets the second specified requirements;

According to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value.

In the tenth kind of possible embodiment provided based on the 9th kind of possible embodiment, in the storer of server, also comprise the instruction for performing following operation:

According to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications, comprising:

For any two multimedia classifications, according to each the first multimedia class object interest score value of first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation in all multimedia classifications between any two multimedia classifications by following formula:

$\mathrm{sim}(a,b)=\frac{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}(r_{a,d}-\stackrel{\‾}{r_a})(r_{b,d}-\stackrel{\‾}{r_b})}{\sqrt{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{a,d}-\stackrel{\‾}{r_a})}^2\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{b,d}-\stackrel{\‾}{r_b})}^2}}$

In formula, a is arbitrary multimedia classification, and b is other arbitrary multimedia classification different from a, and N is all numbers of users, and d is arbitrary user, r _{a, d}for the interest score value of the multimedia classification a of user d, for the interest score value mean value of the multimedia classification a of all users, r _{b, d}for the interest score value of the multimedia classification b of user d, for the interest score value mean value of the multimedia classification b of all users;

According to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value, comprising:

According to each target first multimedia class object interest score value, each target first multimedia classification and specify second degree of correlation between the second multimedia classification and all second users to appointment second multimedia class object interest score value, determine appointment second multimedia class object interest score value by following formula:

$r_{i,f}=\stackrel{\‾}{r_f}+\frac{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})(r_{g^{\′},i}-\stackrel{\‾}{r_{g^{\′}}})}{\sqrt{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})}}$

In formula, f is appointment second multimedia classification, r _i,ffor specifying the interest score value of the second multimedia classification f, for all second users are to the interest score value mean value of appointment second multimedia classification f, G is target first multimedia class destination aggregation (mda), g' is arbitrary target first multimedia classification, sim (f, g') be second degree of correlation between target first multimedia classification g' and appointment the second multimedia classification f, r _{g', i}for the interest score value of the target first multimedia classification g' of first user i, for all second users are to the interest score value mean value of target first multimedia classification g'.

In the 11 kind of possible embodiment provided based on the embodiment that the first is possible, in the storer of server, also comprise the instruction for performing following operation:

According to first user, behavior is clicked to the history of multimedia messages, determines the interested first multimedia classification of first user, comprising:

According to first user, behavior is clicked to the history of multimedia messages, determine first user each the first multimedia class object number of clicks to first user;

According to first user each the first multimedia class object number of clicks to first user, determine the interested first multimedia classification of first user;

Click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of first user, comprising:

Click behavior according to each the first multimedia classification of first user and the history of multiple second user on multimedia information, determine the third phase pass degree between any two multimedia classifications;

According to the third phase pass degree between any two multimedia classifications and the interested first multimedia classification of first user, determine the potential interested second multimedia classification of first user.

The server that the embodiment of the present invention provides, by determining the interested first multimedia classification of first user, and determine the potential interested second multimedia classification that first user history is not clicked, thus according to the interested first multimedia classification of first user and the potential interested second classification determination destination multimedia classification of first user, make when recommending multimedia messages, except the multimedia classification belonging to the multimedia messages clicked according to first user history realizes, also realize according to the potential interested multimedia classification of first user, the destination multimedia classification determined is more accurate, thus make the multimedia messages of recommended destination multimedia classification fully to cause user to note, the clicking rate of the multimedia messages of recommendation can not only be improved, the multimedia messages recommended is made to have more specific aim, and can saving resource and financial resources.In addition, when destination multimedia classification comprises the potential interested arbitrary second multimedia classification of first user, because first user history does not click this second multimedia classification, therefore, the multimedia messages of recommendation has certain freshness to first user.

It should be noted that: the recommendation apparatus of the multimedia messages that above-described embodiment provides is when recommending multimedia messages, only be illustrated with the division of above-mentioned each functional module, in practical application, can distribute as required and by above-mentioned functions and be completed by different functional modules, inner structure by equipment is divided into different functional modules, to complete all or part of function described above.In addition, the recommend method embodiment of the recommendation apparatus of the multimedia messages that above-described embodiment provides and server and multimedia messages belongs to same design, and its specific implementation process refers to embodiment of the method, repeats no more here.

One of ordinary skill in the art will appreciate that all or part of step realizing above-described embodiment can have been come by hardware, the hardware that also can carry out instruction relevant by program completes, described program can be stored in a kind of computer-readable recording medium, the above-mentioned storage medium mentioned can be ROM (read-only memory), disk or CD etc.

The foregoing is only preferred embodiment of the present invention, not in order to limit the present invention, within the spirit and principles in the present invention all, any amendment done, equivalent replacement, improvement etc., all should be included within protection scope of the present invention.

Claims (22)

1. a recommend method for multimedia messages, is characterized in that, described method comprises:

According to first user, behavior is clicked to the history of multimedia messages, determine the interested first multimedia classification of described first user, classification belonging to the multimedia messages that described first multimedia classification is clicked for user's history;

Behavior is clicked according to each the first multimedia classification of described first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of described first user, classification belonging to the multimedia messages that described second multimedia classification is not clicked for user's history;

According to the interested first multimedia classification of described first user and the potential interested second multimedia classification of described first user, determine destination multimedia classification;

The multimedia messages of described destination multimedia classification is recommended to described first user.

2. method according to claim 1, is characterized in that, described according to the history click behavior of first user to multimedia messages, determines the interested first multimedia classification of described first user, comprising:

According to first user, behavior is clicked to the history of multimedia messages, determine each the first multimedia class object interest score value of described first user;

According to each the first multimedia class object interest score value of described first user, determine the interested first multimedia classification of described first user;

The history of described each the first multimedia classification according to described first user and multiple second user on multimedia information clicks behavior, determines the potential interested second multimedia classification of described first user, comprising:

Click behavior according to each the first multimedia class object interest score value of described first user and the history of multiple second user on multimedia information, determine each the second multimedia class object interest score value of described first user;

According to each the second multimedia class object interest score value of described first user, determine the potential interested second multimedia classification of described first user;

Described according to the interested first multimedia classification of described first user and the potential interested second multimedia classification of described first user, determine destination multimedia classification, comprising:

According to the interested first multimedia class object interest score value of described first user and the potential interested second multimedia class object interest score value of described first user, determine the destination multimedia classification recommended to described first user.

3. method according to claim 2, is characterized in that, described according to the history click behavior of first user to multimedia messages, determines each the first multimedia class object interest score value of described first user, comprising:

When using arbitrary first multimedia classification of first user as appointment first multimedia classification time, for described appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that described first user clicks described appointment first multimedia class object number of times is first number;

Click behavior according to described first user to the history of multimedia messages, determining that described first user clicks all multimedia class object number of times sums is second number;

Obtain and click described appointment first multimedia class object number of users and all numbers of users;

According to described first number, described second number, described number of users and described all numbers of users, determine described appointment first multimedia class object interest score value.

4. method according to claim 3, is characterized in that, described according to described first number, described second number, described number of users and described all numbers of users, determines described appointment first multimedia class object interest score value, comprising:

According to described first number, described second number, described number of users and described all numbers of users, determine described appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_1=\frac{{\mathrm{CK}}_{i,j}}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{CK}}_{i,j}}$

${\mathrm{IDF}}_1=\log \frac{N}{N_j}$

S ₁＝TF ₁×IDF ₁

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, S ₁for specifying the interest score value of the first multimedia classification j.

5. method according to claim 2, is characterized in that, described according to the history click behavior of first user to multimedia messages, determines each the first multimedia class object interest score value of described first user, comprising:

When using arbitrary for first user the first multimedia classification as appointment first multimedia classification time, for described appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that described first user clicks described appointment first multimedia class object number of times is first number;

Click behavior according to described first user to the history of multimedia messages, determining that described first user clicks all multimedia class object number of times sums is second number;

Obtain and click described appointment first multimedia class object number of users and all numbers of users;

Click the described appointment first multimedia class object moment according to described first user, determine described appointment first multimedia class object weight;

According to described first number, described second number, described number of users, described all numbers of users and described appointment first multimedia class object weight, determine described appointment first multimedia class object interest score value.

6. method according to claim 5, is characterized in that, describedly clicks the described appointment first multimedia class object moment according to described first user, determines described appointment first multimedia class object weight, comprising:

Click the described appointment first multimedia class object moment according to described first user, determine described appointment first multimedia class object weight by following formula:

$w_j=\frac{1}{1+e^{-t}}$

In formula, t represents that described first user clicks the difference between the moment of described appointment first multimedia classification j and current time;

Described according to described first number, described second number, described number of users, described all numbers of users and described appointment first multimedia class object weight, determine described appointment first multimedia class object interest score value, comprising:

According to described first number, described second number, described number of users, described all numbers of users and described appointment first multimedia class object weight, determine described appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_2=\frac{{\mathrm{CK}}_{i,j}\×w_j}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{CK}}_{i,j}\×w_j)}$

${\mathrm{IDF}}_2=\log \frac{N}{N_j}$

S ₂＝TF ₂×IDF ₂

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, w _jfor specifying the weight of the first multimedia classification j, S ₂for specifying the interest score value of the first multimedia classification j.

7. method according to claim 2, it is characterized in that, the history of described each the first multimedia class object interest score value according to described first user and multiple second user on multimedia information clicks behavior, determine each the second multimedia class object interest score value of described first user, comprising:

For arbitrary second user, the history according to described second user on multimedia information clicks behavior, determines each the first multimedia class object interest score value of described second user;

According to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of described second user, determine described first user and first degree of correlation of described second user when clicking multimedia messages;

From all second users, select target second user of the first numerical value, arbitrary target second user and described first user first degree of correlation when clicking multimedia messages meets the first specified requirements;

When using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for described appointment second multimedia classification, according to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between described first user and each target second user of described first user, determine described appointment second multimedia class object interest score value.

8. method according to claim 7, it is characterized in that, each the first multimedia class object interest score value of described each the first multimedia class object interest score value according to described first user and described second user, determining described first user and first degree of correlation of described second user when clicking multimedia messages, comprising:

According to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of described second user, determine described first user and first degree of correlation of described second user when clicking multimedia messages by following formula:

$\mathrm{sim}(i,y)=\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}(r_{i,k}-\stackrel{\‾}{r_i})(r_{y,k}-\stackrel{\‾}{r_y})}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{i,j}-\stackrel{\‾}{r_i})}^2\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{y,k}-\stackrel{\‾}{r_y})}^2}}$

In formula, i is first user, y is arbitrary second user, m has multimedia class object quantity in the first multimedia classification of first user i with the first multimedia classification of the second user y, k is arbitrary multimedia classification total in the first multimedia classification of first user and the first multimedia classification of the second user, r _{i, k}for the interest score value of the first multimedia classification k of first user i, r _{y, k}be the interest score value of the first multimedia classification k of the second user y, for all first multimedia class object interest score value mean values of first user i, be all first multimedia class object interest score value mean values of the second user y;

Described each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between described first user and each target second user according to described first user, determine described appointment second multimedia class object interest score value, comprising:

According to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between described first user and each target second user of described first user, determine described appointment second multimedia class object interest score value by following formula:

$r_{i,h}=\stackrel{\‾}{r_i}+\frac{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})(r_{u^{\′},h}-\stackrel{\‾}{r_{u^{\′}}})}{\sqrt{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})}}$

In formula, U is the set of target second user, and u' is arbitrary target second user, and h is appointment second multimedia classification, r _i,hfor specifying the interest score value of the second multimedia classification h, sim (i, u') is first degree of correlation between first user i and target second user u', r _{u', h}for the interest score value of the first multimedia classification h of target second user u', for all first multimedia class object interest score value mean values of target second user u'.

9. method according to claim 2, it is characterized in that, the history of described each the first multimedia class object interest score value according to described first user and multiple second user on multimedia information clicks behavior, determine each the second multimedia class object interest score value of described first user, comprising:

History according to each second user on multimedia information clicks behavior, determines each the first multimedia class object interest score value of each second user;

According to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications;

When using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for described appointment second multimedia classification, from all first multimedia classifications of first user, select the target first multimedia classification of second value, second degree of correlation between arbitrary target first multimedia classification and described appointment second multimedia classification meets the second specified requirements;

According to each target first multimedia class object interest score value, second degree of correlation between each target first multimedia classification and described appointment second multimedia classification and all second users to described appointment second multimedia class object interest score value, determine described appointment second multimedia class object interest score value.

10. method according to claim 9, it is characterized in that, each the first multimedia class object interest score value of described each the first multimedia class object interest score value according to described first user and all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications, comprising:

For any two multimedia classifications, according to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation in all multimedia classifications between any two multimedia classifications by following formula:

$\mathrm{sim}(a,b)=\frac{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}(r_{a,b}-\stackrel{\‾}{r_a})(r_{b,d}-\stackrel{\‾}{r_b})}{\sqrt{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{a,d}-\stackrel{\‾}{r_a})}^2\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{b,d}-\stackrel{\‾}{r_b})}^2}}$

In formula, a is arbitrary multimedia classification, and b is other arbitrary multimedia classification different from a, and N is all numbers of users, and d is arbitrary user, r _{a, d}for the interest score value of the multimedia classification a of user d, for the interest score value mean value of the multimedia classification a of all users, r _{b, d}for the interest score value of the multimedia classification b of user d, for the interest score value mean value of the multimedia classification b of all users;

Described according to each target first multimedia class object interest score value, second degree of correlation between each target first multimedia classification and described appointment second multimedia classification and all second users to described appointment second multimedia class object interest score value, determine described appointment second multimedia class object interest score value, comprising:

According to each target first multimedia class object interest score value, second degree of correlation between each target first multimedia classification and described appointment second multimedia classification and all second users to described appointment second multimedia class object interest score value, determine described appointment second multimedia class object interest score value by following formula:

$r_{i,f}=\stackrel{\‾}{r_f}+\frac{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})(r_{g^{\′},i}-\stackrel{\‾}{r_{g^{\′}}})}{\sqrt{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})}}$

In formula, f is appointment second multimedia classification, r _i,ffor specifying the interest score value of the second multimedia classification f, for all second users are to the interest score value mean value of appointment second multimedia classification f, G is target first multimedia class destination aggregation (mda), g' is arbitrary target first multimedia classification, sim (f, g') be second degree of correlation between target first multimedia classification g' and appointment the second multimedia classification f, r _{g', i}for the interest score value of the target first multimedia classification g' of first user i, for all second users are to the interest score value mean value of target first multimedia classification g'.

11. methods according to claim 1, is characterized in that, described according to the history click behavior of first user to multimedia messages, determine the interested first multimedia classification of described first user, comprising:

According to first user, behavior is clicked to the history of multimedia messages, determine described first user each the first multimedia class object number of clicks to described first user;

According to described first user each the first multimedia class object number of clicks to described first user, determine the interested first multimedia classification of described first user;

The history of described each the first multimedia classification according to described first user and multiple second user on multimedia information clicks behavior, determines the potential interested second multimedia classification of described first user, comprising:

The history of described each the first multimedia classification according to described first user and multiple second user on multimedia information clicks behavior, determines the third phase pass degree between any two multimedia classifications;

According to the third phase pass degree between any two multimedia classifications and the interested first multimedia classification of described first user, determine the potential interested second multimedia classification of described first user.

The recommendation apparatus of 12. 1 kinds of multimedia messagess, is characterized in that, described device comprises:

First determination module, for clicking behavior according to first user to the history of multimedia messages, determines the interested first multimedia classification of described first user, classification belonging to the multimedia messages that described first multimedia classification is clicked for user's history;

Second determination module, for clicking behavior according to each the first multimedia classification of described first user and the history of multiple second user on multimedia information, determine the potential interested second multimedia classification of described first user, classification belonging to the multimedia messages that described second multimedia classification is not clicked for user's history;

3rd determination module, for according to the interested first multimedia classification of described first user and the potential interested second multimedia classification of described first user, determines destination multimedia classification;

Recommending module, for recommending the multimedia messages of described destination multimedia classification to described first user.

13. devices according to claim 12, is characterized in that, described first determination module comprises:

First determining unit, for clicking behavior according to first user to the history of multimedia messages, determines each the first multimedia class object interest score value of described first user;

Second determining unit, for each the first multimedia class object interest score value according to described first user, determines the interested first multimedia classification of described first user;

Described second determination module comprises:

3rd determining unit, for clicking behavior according to each the first multimedia class object interest score value of described first user and the history of multiple second user on multimedia information, determines each the second multimedia class object interest score value of described first user;

4th determining unit, for each the second multimedia class object interest score value according to described first user, determines the potential interested second multimedia classification of described first user;

Described 3rd determination module, for according to the interested first multimedia class object interest score value of described first user and the potential interested second multimedia class object interest score value of described first user, determine the destination multimedia classification recommended to described first user.

14. devices according to claim 13, is characterized in that, described first determining unit comprises:

First determines subelement, for when using arbitrary first multimedia classification of first user as appointment first multimedia classification time, for described appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that described first user clicks described appointment first multimedia class object number of times is first number;

Second determines subelement, and for clicking behavior according to described first user to the history of multimedia messages, determining that described first user clicks all multimedia class object number of times sums is second number;

First obtains subelement, clicks described appointment first multimedia class object number of users and all numbers of users for obtaining;

3rd determines subelement, for according to described first number, described second number, described number of users and described all numbers of users, determines described appointment first multimedia class object interest score value.

15. devices according to claim 14, it is characterized in that, described 3rd determines subelement, also for according to described first number, described second number, described number of users and described all numbers of users, determines described appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_1=\frac{{\mathrm{CK}}_{i,j}}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{CK}}_{i,j}}$

${\mathrm{IDF}}_1=\log \frac{N}{N_j}$

S ₁＝TF ₁×IDF ₁

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, S ₁for specifying the interest score value of the first multimedia classification j.

16. devices according to claim 13, is characterized in that, described first determining unit comprises:

4th determines subelement, for when using arbitrary for first user the first multimedia classification as appointment first multimedia classification time, for described appointment first multimedia classification, click behavior according to first user to the history of multimedia messages, determining that described first user clicks described appointment first multimedia class object number of times is first number;

5th determines subelement, and for clicking behavior according to described first user to the history of multimedia messages, determining that described first user clicks all multimedia class object number of times sums is second number;

Second obtains subelement, clicks described appointment first multimedia class object number of users and all numbers of users for obtaining;

6th determines subelement, for clicking the described appointment first multimedia class object moment according to described first user, determines described appointment first multimedia class object weight;

7th determines subelement, for according to described first number, described second number, described number of users, described all numbers of users and described appointment first multimedia class object weight, determines described appointment first multimedia class object interest score value.

17. devices according to claim 16, is characterized in that, the described 6th determines subelement, also for clicking the described appointment first multimedia class object moment according to described first user, determine described appointment first multimedia class object weight by following formula:

$w_j=\frac{1}{1+e^{-t}}$

In formula, t represents that described first user clicks the difference between the moment of described appointment first multimedia classification j and current time;

Described 7th determines subelement, also for according to described first number, described second number, described number of users, described all numbers of users and described appointment first multimedia class object weight, determine described appointment first multimedia class object interest score value by following formula:

${\mathrm{TF}}_2=\frac{{\mathrm{CK}}_{i,j}\×w_j}{\underset{j=1}{\overset{n}{\mathrm{\Σ}}}({\mathrm{CK}}_{i,j}\×w_j)}$

${\mathrm{IDF}}_2=\log \frac{N}{N_j}$

S ₂＝TF ₂×IDF ₂

In formula, j represents appointment first multimedia classification, and i represents first user, CK _{i, j}for first user i clicks first number of appointment first multimedia classification j, n is all multimedia class object quantity, represent that first user i clicks all multimedia class objects second number, N _jfor clicking the number of users of appointment first multimedia classification j, N is all numbers of users, w _jfor specifying the weight of the first multimedia classification j, S ₂for specifying the interest score value of the first multimedia classification j.

18. devices according to claim 13, is characterized in that, described 3rd determining unit comprises:

8th determines subelement, for for arbitrary second user, according to the history click behavior of described second user on multimedia information, determines each the first multimedia class object interest score value of described second user;

9th determines subelement, for according to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of described second user, determine described first user and first degree of correlation of described second user when clicking multimedia messages;

First chooser unit, for selecting target second user of the first numerical value from all second users, arbitrary target second user and described first user first degree of correlation when clicking multimedia messages meets the first specified requirements;

Tenth determines subelement, for when using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for described appointment second multimedia classification, according to each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between described first user and each target second user of described first user, determine described appointment second multimedia class object interest score value.

19. devices according to claim 18, it is characterized in that, described 9th determines subelement, also for according to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of described second user, determine described first user and first degree of correlation of described second user when clicking multimedia messages by following formula:

$\mathrm{sim}(i,y)=\frac{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}(r_{i,k}-\stackrel{\‾}{r_i})(r_{y,k}-\stackrel{\‾}{r_y})}{\sqrt{\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{i,j}-\stackrel{\‾}{r_i})}^2\underset{k=1}{\overset{m}{\mathrm{\Σ}}}{(r_{y,k}-\stackrel{\‾}{r_y})}^2}}$

In formula, i is first user, y is arbitrary second user, m has multimedia class object quantity in the first multimedia classification of first user i with the first multimedia classification of the second user y, k is arbitrary multimedia classification total in the first multimedia classification of first user and the first multimedia classification of the second user, r _{i, k}for the interest score value of the first multimedia classification k of first user i, r _{y, k}be the interest score value of the first multimedia classification k of the second user y, for all first multimedia class object interest score value mean values of first user i, be all first multimedia class object interest score value mean values of the second user y;

Described tenth determines subelement, also for each the first multimedia class object interest score value, each the first multimedia class object interest score value of each target second user and first degree of correlation between described first user and each target second user according to described first user, determine described appointment second multimedia class object interest score value by following formula:

$r_{i,h}=\stackrel{\‾}{r_i}+\frac{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})(r_{u^{\′},h}-\stackrel{\‾}{r_{u^{\′}}})}{\sqrt{\underset{u^{\′}\∈U}{\mathrm{\Σ}}\mathrm{sim}(i,u^{\′})}}$

In formula, U is the set of target second user, and u' is arbitrary target second user, and h is appointment second multimedia classification, r _i,hfor specifying the interest score value of the second multimedia classification h, sim (i, u') is first degree of correlation between first user i and target second user u', r _{u', h}for the interest score value of the first multimedia classification h of target second user u', for all first multimedia class object interest score value mean values of target second user u'.

20. devices according to claim 13, is characterized in that, described 3rd determining unit comprises:

11 determines subelement, clicks behavior, determine each the first multimedia class object interest score value of each second user for the history according to each second user on multimedia information;

12 determines subelement, for according to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation between any two multimedia classifications in all multimedia classifications;

Second chooser unit, for when using arbitrary second multimedia classification of first user as appointment second multimedia classification time, for described appointment second multimedia classification, from all first multimedia classifications of first user, select the target first multimedia classification of second value, second degree of correlation between arbitrary target first multimedia classification and described appointment second multimedia classification meets the second specified requirements;

13 determines subelement, for according to each target first multimedia class object interest score value, second degree of correlation between each target first multimedia classification and described appointment second multimedia classification and all second users to described appointment second multimedia class object interest score value, determine described appointment second multimedia class object interest score value.

21. devices according to claim 20, it is characterized in that, described 12 determines subelement, also for for any two multimedia classifications, according to each the first multimedia class object interest score value of described first user and each the first multimedia class object interest score value of all second users, determine second degree of correlation in all multimedia classifications between any two multimedia classifications by following formula:

$\mathrm{sim}(a,b)=\frac{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}(r_{a,b}-\stackrel{\‾}{r_a})(r_{b,d}-\stackrel{\‾}{r_b})}{\sqrt{\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{a,d}-\stackrel{\‾}{r_a})}^2\underset{d=1}{\overset{N}{\mathrm{\Σ}}}{(r_{b,d}-\stackrel{\‾}{r_b})}^2}}$

In formula, a is arbitrary multimedia classification, and b is other arbitrary multimedia classification different from a, and N is all numbers of users, and d is arbitrary user, r _{a, d}for the interest score value of the multimedia classification a of user d, for the interest score value mean value of the multimedia classification a of all users, r _{b, d}for the interest score value of the multimedia classification b of user d, for the interest score value mean value of the multimedia classification b of all users;

Described 13 determines subelement, also for according to each target first multimedia class object interest score value, second degree of correlation between each target first multimedia classification and described appointment second multimedia classification and all second users to described appointment second multimedia class object interest score value, determine described appointment second multimedia class object interest score value by following formula:

$r_{i,f}=\stackrel{\‾}{r_f}+\frac{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})(r_{g^{\′},i}-\stackrel{\‾}{r_{g^{\′}}})}{\sqrt{\underset{g^{\′}\∈G}{\mathrm{\Σ}}\mathrm{sim}(f,g^{\′})}}$

In formula, f is appointment second multimedia classification, r _i,ffor specifying the interest score value of the second multimedia classification f, for all second users are to the interest score value mean value of appointment second multimedia classification f, G is target first multimedia class destination aggregation (mda), g' is arbitrary target first multimedia classification, sim (f, g') be second degree of correlation between target first multimedia classification g' and appointment the second multimedia classification f, r _{g', i}for the interest score value of the target first multimedia classification g' of first user i, for all second users are to the interest score value mean value of target first multimedia classification g'.

22. devices according to claim 12, it is characterized in that, described first determination module, also for clicking behavior according to first user to the history of multimedia messages, determines described first user each the first multimedia class object number of clicks to described first user; According to described first user each the first multimedia class object number of clicks to described first user, determine the interested first multimedia classification of described first user;

Described second determination module, the history also for described each the first multimedia classification according to described first user and multiple second user on multimedia information clicks behavior, determines the third phase pass degree between any two multimedia classifications; According to the third phase pass degree between any two multimedia classifications and the interested first multimedia classification of described first user, determine the potential interested second multimedia classification of described first user.