CN104915359A - Theme label recommending method and device - Google Patents

Abstract

The embodiment of the invention provides a theme label recommending method and device. The theme label recommending method comprises the steps that a first message published by a user on a social network is obtained, wherein the first message is a message which is not labeled with a theme label; according to a theme model of the social network, the theme labels of the first message are obtained; the obtained theme labels of the first message are displayed on a screen to be selected by a user; according to the selection result of the user, the theme label of the first message is determined. By the adoption of the theme label recommending method and the device, the purpose that theme labels are recommended to the user according to the theme model, the theme label is determined finally according to selection of the user, and therefore the finally determined theme label is closest to the theme which the user wants to express.

Description

Theme label recommend method and device

Technical field

The present invention relates to the information processing technology, particularly relate to a kind of theme label recommend method and device.

Background technology

Universal along with Internet and mobile terminal, uses the user of social networks to get more and more.Everybody can issue or forward some message by social networks, and along with increasing of user, the size of message that social networks is issued is also increasing.

In order to promote the experience of user on social networks, existing social networks proposes theme label (hashtag) function, namely isolate vocabulary to demarcate segregate vocabulary as hashtag by " # " number when user gives out information, this hashtag is the theme that given out information.Such hashtag can help user to obtain the message such as specific topics, dependent event.But this mode must rely on when user gives out information and all incidentally go up assigned tags to demarcate hashtag at every turn, if user does not demarcate hashtag just can not determine out the theme that user gives out information.

For the message of not demarcating hashtag, determine the mode of message subject in prior art, main employing message cluster, is occurred that a fairly large number of vocabulary is as theme label in the message that user is issued.

But the theme label adopting this clustering technique to obtain often accurately can not express the theme that user gives out information.

Summary of the invention

The embodiment of the present invention provides a kind of theme label recommend method and device, and the theme label for solving recommendation accurately can not express the problem of the theme that user gives out information.

Embodiment of the present invention first aspect provides a kind of theme label recommend method, comprising:

Obtain the first message that user issues on social networks, described first message refers to the message not being labeled theme label;

According to the topic model of described social networks, obtain the theme label of described first message;

The theme label of the first message of described acquisition is presented on screen and selects for described user;

According to the result that described user selects, determine the theme label of described first message.

In conjunction with first aspect, in the first possible embodiment of first aspect, the described topic model according to described social networks, before obtaining the theme label of described first message, also comprises:

From described social networks, obtain the second message, described second message refers to the message being marked with theme label;

Pre-service is carried out to described second message, and preserves described pretreated result;

By using machine learning method to train described pretreated result, obtain the topic model of described social networks.

In conjunction with the first possible embodiment of first aspect, in the embodiment that the second of first aspect is possible, described pre-service is carried out to described second message, and preserves described pretreated result, comprising:

Described second message is carried out cutting by vocabulary;

From the result of described cutting, phrase is obtained according to default part of speech;

Described phrase is stored according to preset format.

In conjunction with the first possible embodiment of first aspect or the possible embodiment of the second, in the embodiment that the second of first aspect is possible, described by using machine learning method to train described pretreated result, the topic model obtaining described social networks comprises:

Use machine learning method to sample to described pretreated result, obtain theme-vocabulary probability distribution and theme-theme label probability distribution according to described sampled result;

According to described theme-vocabulary probability distribution and described theme-theme label probability distribution, obtain the topic model of described social networks.

In conjunction with first aspect to first aspect the third possible embodiment any one of, in the 4th kind of possible embodiment of first aspect, the described topic model according to described social networks, the theme label obtaining described first message comprises:

Adopt formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, and the value obtaining described p (hdu) is greater than the theme label of the first predetermined threshold value, wherein: ψ _t,hrepresent the probability occurring theme label h in the theme label of theme t, η _u,rrepresent that described user selects the probability of the history theme of user r, θ _r,trepresent that the message of user r issue belongs to the probability of theme t, π ₀represent that a vocabulary belongs to the probability of not a theme vocabulary, represent in not a theme vocabulary and occur w _{u, d, n}probability, π ₁represent that a vocabulary belongs to the probability of theme vocabulary, represent to belong in the vocabulary of theme t and occur w _{u, d, n}probability, w _{u, d, n}represent the vocabulary n in described first message;

The described theme label by the first message of described acquisition is presented on screen and selects to comprise for described user:

By described p (h|d _u) the value theme label that is greater than the first predetermined threshold value be presented on screen and select for described user.

In conjunction with the 4th kind of possible embodiment of first aspect, in the 5th kind of possible embodiment of first aspect, described employing formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, comprising:

By the ψ in the topic model of described social networks _t,harrange from big to small, and according to ψ _t,horder from big to small substitutes into formula successively $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate, when calculating the p (h|d obtained _u) value stops calculating when being less than the second predetermined threshold value, described second predetermined threshold value is less than or equal to described first predetermined threshold value.

Embodiment of the present invention second aspect provides a kind of theme label recommendation apparatus, comprising:

Acquisition module, for obtaining the first message that user issues on social networks, described first message refers to the message not being labeled theme label; According to the topic model of described social networks, obtain the theme label of described first message;

Recommending module, the theme label for the first message by described acquisition is presented on screen and selects for described user;

Determination module, for the result selected according to described user, determines the theme label of described first message.

In conjunction with second aspect, in the first possible embodiment of second aspect, described acquisition module, also for obtaining the second message from described social networks, described second message refers to the message being marked with theme label;

Described device also comprises:

Pretreatment module, for carrying out pre-service to described second message, and preserves described pretreated result;

Study module, for by using machine learning method to train described pretreated result, obtains the topic model of described social networks;

Described acquisition module, specifically for the topic model stating social networks obtained according to described study module, obtains the theme label of described first message.

In conjunction with the first possible embodiment of second aspect, in the embodiment that the second of second aspect is possible, described pretreatment module, specifically for carrying out cutting by described second message by vocabulary; From the result of described cutting, phrase is obtained according to default part of speech; Described phrase is stored according to preset format.

In conjunction with the first possible embodiment of second aspect or the possible embodiment of the second, in the embodiment that the second of second aspect is possible, described study module, specifically for using machine learning method to sample to described pretreated result, obtain theme-vocabulary probability distribution and theme-theme label probability distribution according to described sampled result; According to described theme-vocabulary probability distribution and described theme-theme label probability distribution, obtain the topic model of described social networks.

In conjunction with second aspect to second aspect the third possible embodiment any one of, in the 4th kind of possible embodiment of second aspect, described acquisition module, specifically for adopt formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, and obtain described p (h|d _u) be greater than the theme label of the first predetermined threshold value, wherein: ψ _t,hrepresent the probability occurring theme label h in the theme label of theme t, η _u,rrepresent that user u selects the probability of the history theme of user r, θ _r,trepresent that the message of user r issue belongs to the probability of theme t, π ₀represent that a vocabulary belongs to the probability of not a theme vocabulary, represent in not a theme vocabulary and occur w _{u, d, n}probability, π ₁represent that a vocabulary belongs to the probability of theme vocabulary, represent to belong in the vocabulary of theme t and occur w _{u, d, n}probability;

Described recommending module, specifically for by described p (h|d _u) the value theme label that is greater than the first predetermined threshold value be presented on screen and select for described user.

In conjunction with the 4th kind of possible embodiment of second aspect, in the 5th kind of possible embodiment of second aspect, described acquisition module, specifically for by the ψ in the topic model of described social networks _t,harrange from big to small, and according to ψ _t,horder from big to small substitutes into formula successively $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate, when calculating the p (h|d obtained _u) value stops calculating when being less than the second predetermined threshold value, described second predetermined threshold value is less than or equal to described first predetermined threshold value.

In the embodiment of the present invention, obtain the first message not marking theme label that user issues on social networks, according to the topic model of social networks, obtain the theme label of this first message, and these theme label are presented on screen for user's selection, according to the selection result of user, determine the theme label of this first message.Achieve and come to user's proposed topic label by topic model, and finally determine theme label by the selection of user, the final like this theme label determined could think the theme of expression originally closest to user.

Accompanying drawing explanation

In order to be illustrated more clearly in the technical scheme of the embodiment of the present invention, be briefly described to the accompanying drawing used required in embodiment below, apparently, accompanying drawing in the following describes is 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 schematic flow sheet of theme label recommendation process embodiment of the method one provided by the invention;

Fig. 2 is the schematic flow sheet of theme label recommendation process embodiment of the method two provided by the invention;

Fig. 3 is the structural representation of theme label recommendation process device embodiment one provided by the invention;

Fig. 4 is the structural representation of theme label recommendation process device embodiment two provided by the invention;

Fig. 5 is the structural representation of theme label recommendation process device embodiment three provided by the invention.

Embodiment

For making the object of the embodiment of the present invention, technical scheme and advantage clearly, below in conjunction with the accompanying drawing in the embodiment of the present invention, technical scheme in the embodiment of the present invention is clearly and completely described, obviously, described embodiment is the present invention's part embodiment, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.

Fig. 1 is the schematic flow sheet of theme label recommendation process embodiment of the method one provided by the invention, and as shown in Figure 1, the method comprises:

The first message that S101, acquisition user issue on social networks, this first message refers to the message not being labeled theme label.

Above-mentioned social networks can be, the microblogging that active user commonly uses, Renren Network, micro-letter circle of friends, push away spy (twitter), the types of facial makeup in Beijing operas (facebook) etc., but not as limit, user can give out information on these social networks, or forward others' message, user likely can add theme label (hashtag) when giving out information, and also may not add hashtag because of trouble and directly issues.The first message herein just refers to the message not adding hashtag.

S102, topic model according to above-mentioned social networks, obtain the theme label of the first message.

This topic model is used for proposed topic label, has broad applicability.

S103, the theme label of the first message of above-mentioned acquisition is presented on screen and selects for above-mentioned user.

Can get multiple theme label for above-mentioned first message according to topic model, these theme label are all shown to user by the display screen of subscriber equipment by these, and user can select wherein, determine oneself to think most suitable theme label.

S104, the result selected according to user, determine the theme label of above-mentioned first message.

User selects one oneself to think most suitable theme label in the theme label of above-mentioned acquisition, and after determining, this theme label is exactly the theme label of the first message.

In the present embodiment, obtain the first message not marking theme label that user issues on social networks, according to the topic model of social networks, obtain the theme label of this first message, and these theme label are presented on screen for user's selection, according to the selection result of user, determine the theme label of this first message.Achieve and come to user's proposed topic label by topic model, and finally determine theme label by the selection of user, the final like this theme label determined could think the theme of expression originally closest to user.Meanwhile, by the first message determination theme label, also facilitate the interested theme of user search or event, help network to carry out the function such as time detecting and monitoring.

Fig. 2 is the schematic flow sheet of theme label recommendation process embodiment of the method two provided by the invention, as shown in Figure 2, before above-mentioned S102, namely according to the topic model of above-mentioned social networks, before obtaining the theme label of the first message, also comprises:

S201, from above-mentioned social networks, obtain the second message, this second message refers to the message being marked with theme label.

While obtaining the second message, the social networks message obtained in social networks can also be comprised.This social networks message can comprise the circle of friends message of user message on social networks and these users.

S202, pre-service is carried out to above-mentioned second message, and preserve this pretreated result.

Particularly, pre-service is carried out to above-mentioned second message, and preserve this pretreated result, Ke Yishi: above-mentioned second message is carried out cutting by vocabulary; From the result of above-mentioned cutting, phrase is obtained according to default part of speech; Above-mentioned phrase is stored according to preset format.

Illustrate, suppose that second message obtained is for " I has spent Christmas Day with several friend together with ", the word combination obtained after carrying out cutting by vocabulary be I with, several, friend, together with, spend, Christmas Day.And wherein " Christmas Day " theme label that is this message.

From the result of above-mentioned cutting, obtain phrase according to default part of speech, can be further above-mentioned word combination is carried out deleting optimization, such as, delete stop words, waste advertisements word etc." several " in above-mentioned word combination, stop words that " " is such are deleted, the word combination after renewal for I with, friend, together with, spend, Christmas Day.

When carrying out storing, the word combination of acquisition and its theme label are stored, and indicate the corresponding relation of vocabulary and theme label.

S203, by using machine learning method to train above-mentioned pretreated result, obtain the topic model of above-mentioned social networks.

When training, the theme label obtained can be trained as existing theme from the second message.

Further, by using machine learning method, above-mentioned pretreated result is trained, obtain the topic model of above-mentioned social networks, be specifically as follows: use machine learning method to sample to above-mentioned pretreated result, obtain theme-vocabulary probability distribution and theme-theme label probability distribution according to above-mentioned sampled result; According to theme-vocabulary probability distribution and theme-theme label probability distribution, obtain the topic model of above-mentioned social networks.

Wherein, theme-vocabulary probability distribution specifically refers to belong to the probability distribution of each vocabulary under same subject; Theme-theme label probability distribution specifically refers to belong to the probability distribution of each theme label under same subject.

In concrete sampling process, above-mentioned pretreated result as sample, can be sampled at random, also can sample according to certain rule.

Using machine learning method to sample to above-mentioned pretreated result, can be carry out gibbs sampler (Gibbs Sampling) to pretreated result.In fact namely the second message is sampled.

But not as limit in the embodiment of the present invention, the variational method (Variational Method) can also be adopted to carry out the study of topic model.

More specifically, be described for concrete certain the second message, pretreated result sampled, mainly comprise three aspects:

(1) the history theme of history theme or the friend of this user that the theme of above-mentioned second message of every bar comes from above-mentioned user is sampled.Namely the theme of a piece of news that user u issues may come from user u once used theme, also may come from certain friend once used theme in user u social networks.Certain sample of such as user u is the theme label of { I, like, recreation ground } is " recreation ground ", and the history theme that sampling analysis " recreation ground " this theme comes from this user u oneself still comes from the history theme of user friend u.

(2) theme of the second message is sampled.What the theme of the second message of namely sampling is.

The theme of the second message may be exactly the theme label of this message, also may be the theme belonging to this theme label." Christmas Day ", " Christmas ", " Christmas ", " silent night " these theme label can belong to " Christmas Day " this theme.

(3) vocabulary in the second message is belonged to theme vocabulary or belongs to not a theme vocabulary sample.I.e. sampling analysis, the vocabulary in certain second message is specifically relevant to theme or irrelevant with theme.Such as " I ", " together ", " spending " this vocabulary that can not embody theme are not a theme vocabulary, and picture " friend ", " Christmas Day " this vocabulary that may become to be the theme are the theme vocabulary.Theme vocabulary can be identical with theme vocabulary, can be also the vocabulary relevant to theme, the vocabulary such as belonging to " Christmas Day " this theme can comprise: " Christmas Day ", " Christmas ", " Christmas ", " silent night " etc. vocabulary.

It should be noted that, when specifically performing, above-mentioned (1), (2), (3) are order in no particular order, can perform simultaneously, also can perform in any order, be not construed as limiting in the embodiment of the present invention.

More specifically, for above-mentioned (1) aspect, sample to the history theme of history theme or the friend of this user that the theme of the second message comes from above-mentioned user, Ke Yishi, adopts formula $p(f_{u,d}=r|f_{\⫬(u,d)},z,y,w,h)\∝\frac{{\mathrm{cf}}_{u,r,\⫬(u,d)}+\mathrm{\δ}}{{\mathrm{cf}}_{u,(\·),\⫬(u,d)}+\left(\right|F_u|+1)\mathrm{\δ}}\×\frac{{\mathrm{ct}}_{r,z_{u,d},\⫬(u,d)}+\mathrm{\α}}{{\mathrm{ct}}_{r,(\·),\⫬(u,d)}+\left(\right|T\left|\right)\mathrm{\α}}$ (being designated as formula 1) history theme to history theme or the friend of this user that the theme of the second message comes from above-mentioned user is sampled.History theme refers to the used theme of user.Wherein: represent that the theme of above-mentioned second message comes from the probability of the history theme of user r, user r is the friend of above-mentioned user (being designated as: user u) oneself or user u; | F _u| represent friend's sum of user u; represent the number of times employing the history theme of user r in the message that user u issues except above-mentioned second message; represent and own except above-mentioned second message the summation of value, namely except above-mentioned second message, for all different user r's the summation of value, namely substitutes into that all different r values obtain the summation of value; z _u,drepresent the theme of this second message; what expression described user r except above-mentioned second message issued belongs to described theme z _u,dthe quantity of message; represent the sum of the message that user r issues except above-mentioned second message; | T| represents the current quantity that there is theme.These parameters substituting into formula can be obtained by statistics.

For above-mentioned (2) aspect, sample to the theme of above-mentioned second message, Ke Yishi, adopts formula:

$p(z_{u,d}=i|z_{\⫬(u,d)},f,y,w,h)\∝\frac{{\mathrm{ct}}_{f_{u,d},i,\⫬(u,d)}+\mathrm{\α}}{{\mathrm{ct}}_{f_{u,d},(\·),\⫬(u,d)}+\left(\right|T\left|\right)\mathrm{\α}}$

$\×\left(\underset{k=0}{\overset{{\mathrm{nw}}_{i,(\·)}-1}{\mathrm{\Π}}}\frac{1}{{\mathrm{cw}}_{i,(\·),\⫬(u,d)}+\left|V\right|\mathrm{\β}+k}\right)$ (being designated as formula 2)

$\×\left(\underset{v\∈w_{u,d}}{\mathrm{\Π}}\underset{k=0}{\overset{{\mathrm{nw}}_{i,w}-1}{\mathrm{\Π}}}({\mathrm{cw}}_{i,v,\⫬(u,d)}+\mathrm{\β}+k)\right)$

$\×\left(\underset{k=0}{\overset{M_{u,d}-1}{\mathrm{\Π}}}\frac{1}{{\mathrm{ch}}_{i,(\·),\⫬(u,d)}+\left|H\right|\mathrm{\μ}+k}\right)\left(\underset{g\∈h_{u,d}}{\mathrm{\Π}}({\mathrm{ch}}_{i,g,\⫬(u,d)}+\mathrm{\μ})\right)$

The theme of above-mentioned second message is sampled, wherein: represent that above-mentioned second message subject is the theme the probability of i; represent the quantity belonging to the message of theme i in the message that above-mentioned user issues except above-mentioned second message; represent the sum of message except above-mentioned second message that above-mentioned user issues, such as user has issued 100 message, then be 99; represent that vocabulary v(vocabulary v represents the vocabulary that in above-mentioned second message one is concrete except above-mentioned second message) belong to the number of times of the vocabulary of theme i; represent the sum belonging to the vocabulary of theme i except above-mentioned second message; w _u,drepresent all vocabulary in above-mentioned second message, vocabulary v just represents w _u,done of them; Nw _i,vrepresent that vocabulary v belongs to the number of times of the vocabulary of theme i; | V| represents the current quantity that there is vocabulary; | H| represents the current quantity that there is theme label; represent and except above-mentioned second message, include theme label g and the quantity belonging to the message of theme i; represent all the summation of value, namely includes the sum of the message of theme label g except above-mentioned second message.These parameters substituting into formula can be obtained by statistics.

For above-mentioned (3) aspect, the vocabulary in above-mentioned second message is belonged to theme vocabulary or belongs to not a theme vocabulary and samples, Ke Yishi, adopt formula $p(y_{u,d,n}=1|y_{\⫬(u,d,n)},f,z,w,h)\∝\frac{{\mathrm{cy}}_{1,\⫬(u,d,n)}+\mathrm{\γ}}{{\mathrm{cy}}_{(\·),\⫬(u,d,n)}+2\mathrm{\γ}}\×\frac{{\mathrm{cw}}_{i,w_{u,d,n},\⫬(u,d,n)}+\mathrm{\β}}{{\mathrm{cw}}_{i,(\·),\⫬(u,d,n)}+\left|V\right|\mathrm{\β}}$ (being designated as formula 3) belongs to theme vocabulary (non-background vocabulary) to the vocabulary in above-mentioned second message and samples.Adopt formula $p(y_{u,d,n}=0|y_{\⫬(u,d,n)},f,z,w,h)\∝\frac{{\mathrm{cy}}_{0,\⫬(u,d,n)}+\mathrm{\γ}}{{\mathrm{cy}}_{(\·),\⫬(u,d,n)}+2\mathrm{\γ}}\×\frac{{\mathrm{cw}}_{B,w_{u,d,n},\⫬(u,d,n)}+\mathrm{\β}}{{\mathrm{cw}}_{B,(\·),\⫬(u,d,n)}+\left|V\right|\mathrm{\β}}$ (being designated as formula 4) belongs to theme vocabulary (background vocabulary) to the vocabulary in above-mentioned second message and samples.

Wherein: the vocabulary n(vocabulary n representing in above-mentioned second message represents the vocabulary that in above-mentioned second message is concrete) belong to the probability of theme vocabulary, represent that the vocabulary n in above-mentioned second message belongs to the probability of not a theme vocabulary; represent the vocabulary quantity belonging to theme vocabulary except the vocabulary n in described second message; represent the sum of vocabulary except the vocabulary n in the second message; represent the vocabulary quantity belonging to not a theme vocabulary except the vocabulary n in described second message; represent w _{u, d, n}the number of times of theme i is being appeared at, wherein w except the vocabulary n in described second message _u,drepresent all vocabulary in above-mentioned second message, vocabulary n just represents w _u,done of them; represent the sum of different vocabulary there is theme i except the vocabulary n in described second message under, | V| represents the current quantity that there is vocabulary.These parameters substituting into formula can be obtained by statistics.

According to the result of above-mentioned Gibbs Sampling, the topic model got comprises following parameter: 1) vocabulary belongs to the probability of not a theme vocabulary or belongs to the probability of theme vocabulary, is designated as π _k, k=0,1, represent that as k=0 this vocabulary is not a theme vocabulary, during k=1, represent that this vocabulary is the theme vocabulary.2) message that above-mentioned user (being designated as user u) issues belongs to the probability of theme t, is designated as θ _u,t.3) in not a theme vocabulary, there is the probability of vocabulary v, be designated as φ _b,v.4) in the vocabulary belonging to theme t, there is the probability of vocabulary v, be designated as φ _t,v.5) in the theme label of theme t, there is the probability of theme label h, be designated as ψ _t,h.6) above-mentioned user (user u) selects the probability of the history theme of user r, is designated as η _u,r, user r is the friend of user u or user u itself, and in the embodiment of the present invention, friend all refers to the friend of user on social networks.

α, β, γ, μ, δ in above-mentioned sampling are respectively the Di Li Cray priori (Dirichlet prior) of parameter θ, φ, π, ψ, η in above-mentioned topic model.

More specifically, can, according to the result of Gibbs Sampling, following formula be adopted to calculate above-mentioned parameter respectively: 1) k=0,1(are designated as formula 5); 2) t ∈ T(is designated as formula 6); 3) (being designated as formula 7); 4) ${\mathrm{\φ}}_{t,v}=\frac{{\mathrm{cw}}_{t,v}+\mathrm{\β}}{{\mathrm{cw}}_{t,(\·)}+\mathrm{\β}},\∀t\∈T,$ V ∈ V(is designated as formula 8); 5) ${\mathrm{\ψ}}_{t,h}=\frac{{\mathrm{ch}}_{t,h}+\mathrm{\μ}}{{\mathrm{ch}}_{t,(\·)}+\left|H\right|\mathrm{\μ}},\∀t\∈T,$ H ∈ H(is designated as formula 8); 6) ${\mathrm{\η}}_{u,r}=\frac{{\mathrm{cF}}_{u,r}+\mathrm{\δ}}{{\mathrm{cF}}_{u,(\·)}+\left(\right|F_u|+1)\mathrm{\δ}},\∀u\∈U,r\∈\left|F_u\right|\∪\left\{u\right\}$ (being designated as formula 8).

Wherein, U represents the set of all users crawled; Cy _krepresent and belong to the quantity of not a theme vocabulary or the quantity of theme vocabulary, wherein represent the quantity of not a theme vocabulary during k=0, during k=1, represent the quantity of theme vocabulary; Cy ₍₎represent the summation of the quantity of theme vocabulary and the quantity of not a theme vocabulary; Ct _u,trepresent the quantity belonging to the message of theme t that user u issues; Ct _{u, ()}represent the sum of the message that user u issues; Cw _b,vrepresent that vocabulary v is the number of times of not a theme vocabulary; Cw _{b, ()}represent the summation of not a theme vocabulary occurrence number; Cw _t,vrepresent that vocabulary v belongs to the number of times of theme t; Cw _{t, ()}represent the summation of the number of times that the vocabulary belonging to theme t occurs; Ch _t,hrepresent the number of times occurring theme label h in the theme label of theme t; Ch _{t, ()}represent the sum of all theme label of theme t; Cf _u,rrepresent that user u uses the number of times of the history theme of user r; Cf _{u, ()}represent the sum of the used theme of user u.These parameters come from aforementioned sample result and statistics.

It should be noted that, in order to improve operation efficiency, the process of above-mentioned acquisition topic model can be processed by multiple parallel processor simultaneously, is finally arranged by keeper (manager) again, exports the topic model got.

Certainly, in specific implementation process, constantly can swash from social networks according to the time interval of presetting and fetch data, also above-mentioned flow process can constantly be repeated, for the new data obtained, if just get the message that user newly issues, the theme label wherein adopted currently there is label, so this message performs above-mentioned sampling as new sample, and upgrades topic model; If the new information got, further comprises new theme label, so first on the vector dimension of current already present theme label, add 1, then above-mentioned sampling is carried out to this sample, and upgrade topic model.

Further, in order to improve the efficiency that theme label is recommended, can also be optimized topic model.In specific implementation process, after obtaining topic model, can will be less than the ψ of the 3rd predetermined threshold value in above-mentioned topic model on the one hand _t,hcorresponding theme label from current existed theme label remove, namely preset the 3rd predetermined threshold value and existing ψ _t,hcompare, will the ψ of the 3rd predetermined threshold value be less than _t,hcorresponding theme label from stored current existed theme label remove, just need not consider the theme label removed during such subsequent recommendation theme label, reduce operand, raise the efficiency.Similarly, on the other hand, the θ of the 4th predetermined threshold value will can be less than in above-mentioned topic model _u,tcorresponding theme from current existed theme remove, be exactly without the theme that these remove in subsequent recommendation process.Particularly, in specific implementation process, the optimization of above-mentioned two aspects can be performed simultaneously, also only can perform one of them as the case may be.

Further, the above-mentioned topic model according to social networks, obtains the theme label of above-mentioned first message, is specially employing formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, and obtain described p (h|d _u) value be greater than the theme label of the first predetermined threshold value.Wherein: ψ _t,hrepresent the probability occurring theme label h in the theme label of theme t, η _u,rrepresent that above-mentioned user (user u) selects the probability of the history theme of user r, θ _r,trepresent that the message of user r issue belongs to the probability of theme t, π ₀represent that a vocabulary belongs to the probability of not a theme vocabulary, represent in not a theme vocabulary and occur w _{u, d, n}probability, π ₁represent that a vocabulary belongs to the probability of theme vocabulary, represent to belong in the vocabulary of theme t and occur w _{u, d, n}probability, w _{u, d, n}represent the vocabulary n in described first message.W _{u, d, n}represent the vocabulary n in described first message.

Correspondingly, the above-mentioned theme label by the first message obtained is presented on screen and selects for user, is specially: by above-mentioned p (h|d _u) the value theme label that is greater than the first predetermined threshold value be presented on screen and select for described user.

In another kind of mode, also can be, by the p (h|d calculated _u) value arranges, from big to small in order by p (h|d _u) maximum N number of theme label be presented on screen for user select, wherein N is the number of default proposed topic label.

Theme label h be current any one that existed in theme label, substitute into these values successively for different theme label and carry out calculating p (h|d _u) value.Owing to being independently for the computing of each theme label h, during specific implementation, in order to improve operand, different theme label can be processed by multiple processor simultaneously, obtaining p (h|d _u) value, finally carry out arranging by keeper (manager) theme label obtaining being shown to user again.

Further, in order to improve the efficiency of proposed topic label, above-mentioned employing formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate above-mentioned first message d _utheme label be the probability of h, Ke Yishi, by the ψ in the topic model of above-mentioned social networks _t,harrange from big to small, and according to ψ _t,horder from big to small substitutes into formula successively $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate, when calculating the p (h|d obtained _u) value stops calculating when being less than the second predetermined threshold value, described second predetermined threshold value is less than or equal to described first predetermined threshold value.Preset second predetermined threshold value like this, just unnecessary all parameters substitution calculates, and reduces operand, improves work efficiency.

Fig. 3 is the structural representation of theme label recommendation process device embodiment one provided by the invention, and as shown in Figure 3, this device comprises: acquisition module 301, recommending module 302 and determination module 304, wherein:

Acquisition module 301, for obtaining the first message that user issues on social networks, described first message refers to the message not being labeled theme label; According to the topic model of described social networks, obtain the theme label of described first message.

Recommending module 302, the theme label for the first message by described acquisition is presented on screen and selects for described user.

Determination module 303, for the result selected according to described user, determines the theme label of described first message.

This device is for performing preceding method embodiment, and it realizes principle and technique effect is similar, does not repeat them here.

Fig. 4 is the structural representation of theme label recommendation process device embodiment two provided by the invention, and as shown in Figure 4, on the basis of Fig. 3, this device can also comprise: pretreatment module 401 and study module 402.

In specific implementation process:

Above-mentioned acquisition module 301, also for obtaining the second message from described social networks, described second message refers to the message being marked with theme label.

Pretreatment module 401, for carrying out pre-service to described second message, and preserves described pretreated result.

Study module 402, for by using machine learning method to train described pretreated result, obtains the topic model of described social networks.

Acquisition module 301, specifically for the topic model stating social networks obtained according to study module 402, obtains the theme label of described first message.

Further, pretreatment module 401, specifically for carrying out cutting by described second message by vocabulary; From the result of described cutting, phrase is obtained according to default part of speech; Described phrase is stored according to preset format.

Study module 402, specifically for using machine learning method to sample to described pretreated result, obtains theme-vocabulary probability distribution and theme-theme label probability distribution according to described sampled result; According to described theme-vocabulary probability distribution and described theme-theme label probability distribution, obtain the topic model of described social networks.

Further, above-mentioned acquisition module 301, specifically for adopting formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, and obtain described p (h|d _u) value be greater than the theme label of the first predetermined threshold value, wherein: ψ _t,hrepresent the probability occurring theme label h in the theme label of theme t, η _u,rrepresent that described user selects the probability of the history theme of user r, θ _r,trepresent that the message of user r issue belongs to the probability of theme t, π ₀represent that a vocabulary belongs to the probability of not a theme vocabulary, represent in not a theme vocabulary and occur w _{u, d, n}probability, π ₁represent that a vocabulary belongs to the probability of theme vocabulary, represent to belong in the vocabulary of theme t and occur w _{u, d, n}probability, w _{u, d, n}represent the vocabulary n in described first message.

Recommending module 302, specifically for by described p (h|d _u) the value theme label that is greater than the first predetermined threshold value be presented on screen and select for described user.

Alternatively, acquisition module 301, specifically for by the ψ in the topic model of described social networks _t,harrange from big to small, and according to ψ _t,horder from big to small substitutes into formula successively $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate, when calculating the p (h|d obtained _u) value stops calculating when being less than the second predetermined threshold value, described second predetermined threshold value is less than or equal to described first predetermined threshold value.

This device is for performing preceding method embodiment, and it realizes principle and technique effect is similar, does not repeat them here.

Fig. 5 is the structural representation of theme label recommendation process device embodiment three provided by the invention, and as shown in Figure 5, this device comprises: storer 501 and processor 502.Wherein:

Storer 501 is for store sets of instructions.This processor 502 is configured to call the instruction set in storer 501, to perform following flow process: obtain the first message that user issues on social networks, described first message refers to the message not being labeled theme label; According to the topic model of described social networks, obtain the theme label of described first message; The theme label of the first message of described acquisition is presented on screen and selects for described user; According to the result that described user selects, determine the theme label of described first message.

This device is for performing preceding method embodiment, and it realizes principle and technique effect is similar, does not repeat them here.

Processor 502, also for obtaining the second message from described social networks, described second message refers to the message being marked with theme label; Pre-service is carried out to described second message, and preserves described pretreated result; By using machine learning method to train described pretreated result, obtain the topic model of described social networks.

Further, processor 502, specifically for carrying out cutting by described second message by vocabulary; From the result of described cutting, phrase is obtained according to default part of speech; Described phrase is stored according to preset format.

Processor 502, specifically for using machine learning method to sample to described pretreated result, obtains theme-vocabulary probability distribution and theme-theme label probability distribution according to described sampled result; According to described theme-vocabulary probability distribution and described theme-theme label probability distribution, obtain the topic model of described social networks.

Processor 502, specifically for adopting formula

$p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, and obtain described p (h|d _u) value be greater than the theme label of the first predetermined threshold value, wherein: ψ _t,hrepresent the probability occurring theme label h in the theme label of theme t, η _u,rrepresent that user u selects the probability of the history theme of user r, θ _r,trepresent that the message of user r issue belongs to the probability of theme t, π ₀represent that a vocabulary belongs to the probability of not a theme vocabulary, represent in not a theme vocabulary and occur w _{u, d, n}probability, π ₁represent that a vocabulary belongs to the probability of theme vocabulary, represent with the vocabulary belonging to theme t in there is w _{u, d, n}probability.

Processor 502, specifically for by described p (h|d _u) the value theme label that is greater than the first predetermined threshold value be presented on screen and select for described user.

Alternatively, processor 502, specifically for by the ψ in the topic model of described social networks _t,harrange from big to small, and according to ψ _t,horder from big to small substitutes into formula successively $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate, when calculating the p (h|d obtained _u) value stops calculating when being less than the second predetermined threshold value, described second predetermined threshold value is less than or equal to described first predetermined threshold value.

In several embodiment provided by the present invention, should be understood that, disclosed apparatus and method, can realize by another way.Such as, device embodiment described above is only schematic, such as, the division of described unit, be only a kind of logic function to divide, actual can have other dividing mode when realizing, such as multiple unit or assembly can in conjunction with or another system can be integrated into, or some features can be ignored, or do not perform.Another point, shown or discussed coupling each other or direct-coupling or communication connection can be by some interfaces, and the indirect coupling of device or unit or communication connection can be electrical, machinery or other form.

The described unit illustrated as separating component or can may not be and physically separates, and the parts as unit display can be or may not be physical location, namely can be positioned at a place, or also can be distributed in multiple network element.Some or all of unit wherein can be selected according to the actual needs to realize the object of the present embodiment scheme.

In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, also can be that the independent physics of unit exists, also can two or more unit in a unit integrated.Above-mentioned integrated unit both can adopt the form of hardware to realize, and the form that hardware also can be adopted to add SFU software functional unit realizes.

The above-mentioned integrated unit realized with the form of SFU software functional unit, can be stored in a computer read/write memory medium.Above-mentioned SFU software functional unit is stored in a storage medium, comprising some instructions in order to make a computer equipment (can be personal computer, server, or the network equipment etc.) or processor (processor) perform the part steps of method described in each embodiment of the present invention.And aforesaid storage medium comprises: USB flash disk, portable hard drive, ROM (read-only memory) (Read-Only Memory, ROM), random access memory (Random Access Memory, RAM), magnetic disc or CD etc. various can be program code stored medium.

Last it is noted that above each embodiment is only in order to illustrate technical scheme of the present invention, be not intended to limit; Although with reference to foregoing embodiments to invention has been detailed description, those of ordinary skill in the art is to be understood that: it still can be modified to the technical scheme described in foregoing embodiments, or carries out equivalent replacement to wherein some or all of technical characteristic; And these amendments or replacement, do not make the essence of appropriate technical solution depart from the scope of various embodiments of the present invention technical scheme.

Claims (12)

1. a theme label recommend method, is characterized in that, comprising:

Obtain the first message that user issues on social networks, described first message refers to the message not being labeled theme label;

According to the topic model of described social networks, obtain the theme label of described first message;

The theme label of the first message of described acquisition is presented on screen and selects for described user;

According to the result that described user selects, determine the theme label of described first message.

2. method according to claim 1, is characterized in that, the described topic model according to described social networks, before obtaining the theme label of described first message, also comprises:

From described social networks, obtain the second message, described second message refers to the message being marked with theme label;

Pre-service is carried out to described second message, and preserves described pretreated result;

By using machine learning method to train described pretreated result, obtain the topic model of described social networks.

3. method according to claim 2, is characterized in that, describedly carries out pre-service to described second message, and preserves described pretreated result, comprising:

Described second message is carried out cutting by vocabulary;

From the result of described cutting, phrase is obtained according to default part of speech;

Described phrase is stored according to preset format.

4. according to the method in claim 2 or 3, it is characterized in that, described by using machine learning method to train described pretreated result, the topic model obtaining described social networks comprises:

Use machine learning method to sample to described pretreated result, obtain theme-vocabulary probability distribution and theme-theme label probability distribution according to described sampled result;

According to described theme-vocabulary probability distribution and described theme-theme label probability distribution, obtain the topic model of described social networks.

5. the method according to any one of Claims 1-4, is characterized in that, the described topic model according to described social networks, and the theme label obtaining described first message comprises:

Adopt formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, and obtain described p (h|d _u) value be greater than the theme label of the first predetermined threshold value, wherein: ψ _t,hrepresent the probability occurring theme label h in the theme label of theme t, η _u,rrepresent that described user selects the probability of the history theme of user r, θ _r,trepresent that the message of user r issue belongs to the probability of theme t, π ₀represent that a vocabulary belongs to the probability of not a theme vocabulary, represent in not a theme vocabulary and occur w _{u, d, n}probability, π ₁represent that a vocabulary belongs to the probability of theme vocabulary, represent to belong in the vocabulary of theme t and occur w _{u, d, n}probability, w _{u, d, n}represent the vocabulary n in described first message;

The described theme label by the first message of described acquisition is presented on screen and selects to comprise for described user:

By described p (h|d _u) the value theme label that is greater than the first predetermined threshold value be presented on screen and select for described user.

6. method according to claim 5, is characterized in that, described employing formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, comprising:

By the ψ in the topic model of described social networks _t,harrange from big to small, and according to ψ _t,horder from big to small substitutes into formula successively $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate, when calculating the p (h|d obtained _u) value stops calculating when being less than the second predetermined threshold value, described second predetermined threshold value is less than or equal to described first predetermined threshold value.

7. a theme label recommendation apparatus, is characterized in that, comprising:

Acquisition module, for obtaining the first message that user issues on social networks, described first message refers to the message not being labeled theme label; According to the topic model of described social networks, obtain the theme label of described first message;

Recommending module, the theme label for the first message by described acquisition is presented on screen and selects for described user;

Determination module, for the result selected according to described user, determines the theme label of described first message.

8. device according to claim 7, is characterized in that, described acquisition module, and also for obtaining the second message from described social networks, described second message refers to the message being marked with theme label;

Described device also comprises:

Pretreatment module, for carrying out pre-service to described second message, and preserves described pretreated result;

Study module, for by using machine learning method to train described pretreated result, obtains the topic model of described social networks;

Described acquisition module, specifically for the topic model stating social networks obtained according to described study module, obtains the theme label of described first message.

9. device according to claim 8, is characterized in that, described pretreatment module, specifically for described second message is carried out cutting by vocabulary; From the result of described cutting, phrase is obtained according to default part of speech; Described phrase is stored according to preset format.

10. device according to claim 8 or claim 9, it is characterized in that, described study module, specifically for using machine learning method to sample to described pretreated result, obtains theme-vocabulary probability distribution and theme-theme label probability distribution according to described sampled result; According to described theme-vocabulary probability distribution and described theme-theme label probability distribution, obtain the topic model of described social networks.

11. devices according to any one of claim 7 to 10, is characterized in that, described acquisition module, specifically for adopting formula $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate described first message d _utheme label be the probability of h, and obtain described p (h|d _u) be greater than the theme label of the first predetermined threshold value, wherein: ψ _t,hrepresent the probability occurring theme label h in the theme label of theme t, η _u,rrepresent that user u selects the probability of the history theme of user r, θ _r,trepresent that the message of user r issue belongs to the probability of theme t, π ₀represent that a vocabulary belongs to the probability of not a theme vocabulary, represent in not a theme vocabulary and occur w _{u, d, n}probability, π ₁represent that a vocabulary belongs to the probability of theme vocabulary, represent to belong in the vocabulary of theme t and occur w _{u, d, n}probability;

Described recommending module, specifically for by described p (h|d _u) the value theme label that is greater than the first predetermined threshold value be presented on screen and select for described user.

12. devices according to claim 7, is characterized in that, described acquisition module, specifically for by the ψ in the topic model of described social networks _t,harrange from big to small, and according to ψ _t,horder from big to small substitutes into formula successively $p\left(h\right|d_u)=\underset{t}{\mathrm{\Σ}}{\mathrm{\ψ}}_{t,h}\×\left(\left(\underset{r\∈F_u\∪\left\{U\right\}}{\mathrm{\Σ}}{\mathrm{\η}}_{u,r}{\mathrm{\θ}}_{r,t}\right)\underset{w_{u,d,n}}{\mathrm{\Π}}({\mathrm{\π}}_0{\mathrm{\φ}}_{B,w_{u,d,n}}+{\mathrm{\π}}_1{\mathrm{\φ}}_{t,w_{u,d,n}})\right)$ Calculate, when calculating the p (h|d obtained _u) value stops calculating when being less than the second predetermined threshold value, described second predetermined threshold value is less than or equal to described first predetermined threshold value.