CN105956146A - Article information recommending method and device - Google Patents

Abstract

The invention discloses an article information recommending method and device. The method comprises the steps of obtaining attribute information and user behavior data of an access user when an article access request is received; obtaining a corresponding candidate article set; determining articles satisfying preset conditions in the candidate article set based on a similarity matrix, the attribute information and the user behavior data, wherein the similarity matrix is used for indicating the similarities among the candidate articles, and the similarities between the candidate articles and the attribute information; and recommending the information of the articles satisfying the preset conditions to the access user. Through adoption of the similarity matrix, the attribute information of the access user and recent different click and consumption behaviors to the articles, intention prediction is carried out on user access; and therefore, the articles suitable for the user are determined and recommended to the user. Compared with the mode of carrying out recommendation through prediction of the click-through-rate scores of the user to the articles based on a linear model, the method and the device have the advantages of improving individuation of the recommendation results and improving the accuracy of the recommendation results.

Description

A kind of recommendation method and device of Item Information

Technical field

The invention belongs to communication technical field, particularly relate to the recommendation method and device of a kind of Item Information.

Background technology

Along with the continuous propelling of network, user is more and more higher to the requirement of network various functions when accessing network, Such as, as a example by article are recommended, user typically can wish that website or application shop are recommended and article required for it Ware or relative article, thus the article needed can be had access to by the way of comparison query more easily.

At present, traditional article recommend this special recurrence (LR, Logistic Regression) of general employing logic Model prediction user exposure conversion ratio (CTR, the Click-Through-Rate) score to article, such as article At the application shop intraday download time of homepage and the ratio of exposure frequency, according to CTR, article are entered thereafter Row sequence, thus obtain recommendation results.

To in the research of prior art and practice process, it was found by the inventors of the present invention that owing to LR is one Planting linear model, non-linear expression is limited in one's ability, and LR model is a kind of statistical probability based on crowd Model, the nonlinear problem therefore recommended in the face of magnanimity all kinds article, it may appear that recommendation results is personalized Deficiency and the inaccurate problem of recommendation results.

Summary of the invention

It is an object of the invention to provide the recommendation method and device of a kind of Item Information, it is intended to promote article and push away The personalization recommended and the accuracy improving recommendation results.

For solving above-mentioned technical problem, embodiment of the present invention offer techniques below scheme:

A kind of recommendation method of Item Information, including:

When receiving article access request, obtain attribute information and the user behavior data accessing user；

Obtain corresponding candidate item collection；

Based on default similarity matrix, the attribute information of described access user and user behavior data, in institute Stating candidate item to concentrate and determine and meet pre-conditioned article, described default similarity matrix is used for indicating time Select the similarity between the similarity between article and candidate item and attribute information；

The described Item Information met corresponding to pre-conditioned article is recommended to accessing user.

For solving above-mentioned technical problem, the embodiment of the present invention also provides for techniques below scheme:

A kind of recommendation apparatus of Item Information, including:

First acquiring unit, for when receiving article access request, obtains the attribute information accessing user And user behavior data；

Second acquisition unit, is used for obtaining corresponding candidate item collection；

Determine unit, for based on default similarity matrix, the attribute information of described access user and user Behavioral data, determines in described candidate item concentration and meets pre-conditioned article, described default similarity Matrix is for indicating the similarity between the similarity between candidate item and candidate item and attribute information；

Recommendation unit, is used for the described Item Information met corresponding to pre-conditioned article to accessing user Recommend.

Relative to prior art, the embodiment of the present invention, when receiving article access request, obtain to access and use The attribute information at family and user behavior data；Then, based on default similarity matrix and get Access attribute information and the user behavior data of user, determine in candidate item concentration and meet pre-conditioned thing Product, described default similarity matrix is for indicating the similarity between candidate item and candidate item and genus Similarity between property information；Finally, the Item Information corresponding to pre-conditioned article will be met to this visit Ask that user recommends.I.e. the embodiment of the present invention is by similarity matrix, the attribute information of access user and user's row For data, as recent to the click of article difference and consuming behavior etc., realize user is accessed carrying out being intended in advance Survey, wherein similar between similarity and article and user base attribute between similarity moment matrix representation article Degree, may thereby determine that out the article of applicable user and recommends, relative to based on Linear Model for Prediction user couple The exposure conversion ratio score of article carries out the mode recommended, and not only improves the personalization of recommendation results, changes Kind recommendation effect, also improves the accuracy of recommendation results.

Accompanying drawing explanation

Below in conjunction with the accompanying drawings, by the detailed description of the invention of the present invention is described in detail, the skill of the present invention will be made Art scheme and other beneficial effect are apparent.

Fig. 1 a is the scene schematic diagram of the commending system of the Item Information that the embodiment of the present invention provides；

Fig. 1 b is the schematic flow sheet of the recommendation method of the Item Information that first embodiment of the invention provides；

Fig. 1 c is the principle schematic of the conduction of heat algorithm that first embodiment of the invention provides；

Fig. 1 d is the Computing Principle schematic diagram of the conduction of heat algorithm that first embodiment of the invention provides；

The schematic flow sheet of the recommendation method of the Item Information that Fig. 2 a provides for second embodiment of the invention；

The article that Fig. 2 b provides for second embodiment of the invention recommend interface schematic diagram；

The structural representation of the recommendation apparatus of the Item Information that Fig. 3 a provides for third embodiment of the invention；

Another structural representation of the recommendation apparatus of the Item Information that Fig. 3 b provides for third embodiment of the invention；

The structural representation of the server that Fig. 4 provides for fourth embodiment of the invention.

Detailed description of the invention

Refer to graphic, the most identical element numbers represents identical assembly, and the principle of the present invention is with reality The computing environment that Shi Yi is suitable illustrates.The following description is concrete based on the illustrated present invention Embodiment, it is not construed as limiting other specific embodiment that the present invention is the most detailed herein.

In the following description, the specific embodiment of the present invention will be with reference to by performed by one or multi-section computer Step and symbol illustrate, unless otherwise stating clearly.Therefore, these steps and operation will have mention for several times by Computer performs, and computer as referred to herein performs to include by representing with the data in a structuring pattern The operation of computer processing unit of electronic signal.This operation is changed these data or is maintained at this calculating Position in the memory system of machine, it is reconfigurable or other with the side known to the tester of this area Formula changes the running of this computer.The data structure that these data are maintained is the provider location of this internal memory, its Have by particular characteristics defined in this data form.But, the principle of the invention illustrates with above-mentioned word, It is not represented as a kind of restriction, and this area tester will appreciate that plurality of step and the behaviour of the following stated Also may be implemented in the middle of hardware.

Term as used herein " module " can regard the software object as performing on this arithmetic system as.This Different assemblies, module, engine and service described in literary composition can be regarded as the objective for implementation on this arithmetic system. And device and method as herein described is preferably implemented in the way of software, the most also can be enterprising at hardware Row is implemented, all within scope.

The embodiment of the present invention provides the recommendation method and device of a kind of Item Information.

See Fig. 1 a, the scene signal of the commending system of the Item Information that this figure is provided by the embodiment of the present invention Figure, the commending system of this Item Information can include the recommendation apparatus of Item Information, the recommendation of this Item Information Device specifically can integrated be mainly used in when receiving article access request in the server, obtains and accesses The attribute information of user and user behavior data；Then, corresponding candidate item collection is obtained；Based on default Similarity matrix, the attribute information of described access user and user behavior data, concentrate at described candidate item Determining and meet pre-conditioned article, wherein, this similarity matrix is similar for indicate between candidate item Similarity between degree and candidate item and attribute information, finally meets pre-conditioned article institute by described Corresponding Item Information is recommended to accessing user, etc..

Additionally, the commending system of this Item Information can also include memorizer, it is mainly used in storing the genus of user Property information and user behavior data, such as, the attribute such as the age of user, sex and user are in the recent period to article The data such as different clicks and consuming behavior, it addition, this memorizer can also include Item Information, such as item name Title, goods categories etc..Certainly, the commending system of this Item Information can also include user terminal, mainly uses In receiving the article access request accessing user's input and by the recommendation apparatus of Item Information to accessing user The article recommended carry out display etc..

To be described in detail respectively below.

First embodiment

In the present embodiment, will be described from the angle of the recommendation apparatus of Item Information, this Item Information Recommendation apparatus specifically can be integrated in the network equipment such as server or gateway.

A kind of recommendation method of Item Information, including: when receiving article access request, obtain to access and use The attribute information at family and user behavior data；Obtain corresponding candidate item collection；Based on default similarity moment Battle array, the described attribute information accessing user and user behavior data, determine satisfied in described candidate item concentration Pre-conditioned article, described default similarity matrix for indicate the similarity between candidate item and Similarity between candidate item and attribute information；By the described article met corresponding to pre-conditioned article Information is recommended to accessing user.

Refer to the stream that Fig. 1 b, Fig. 1 b is the recommendation method of the Item Information that first embodiment of the invention provides Journey schematic diagram.Described method includes:

In step S101, when receiving article access request, obtain access user attribute information and User behavior data.

Such as, access user and can pass through cell-phone customer terminal or PC (personal computer, Ge Ren electricity Brain) end etc. sends article access request to the recommendation apparatus of Item Information, the recommendation apparatus of Item Information according to This article access request, obtains attribute information and the user behavior data of this access user.

In the embodiment of the present invention, described access user attribute information be primarily referred to as user's sex, the age, The demographic attributes such as territory；It is clear at PC end or cell-phone customer terminal that described user behavior data refers mainly to user Look at, click on, buy, download, the user behaviors log such as installation.

Described article can refer specifically to the thing provided on the application program in mobile phone terminal application store, shopping platform Product etc., described Item Information can refer specifically to Item Title, goods categories, article size, article tag Etc. information, it is not especially limited herein.

In step s 102, corresponding candidate item collection is obtained.

Such as, when receiving article access request by terminal applies (as mobile phone terminal applies store), can To get this terminal applies corresponding item information data storehouse, and by the article in this item information data storehouse Information setting is candidate item collection.

In step s 103, based on default similarity matrix, the attribute information of described access user and use Family behavioral data, concentrates at described candidate item and determines and meet pre-conditioned article, described default similar Degree matrix is similar for indicate between the similarity between candidate item and candidate item and attribute information Degree.

In the embodiment of the present invention, this similarity matrix can be set in advance in the recommendation apparatus of Item Information, Then when receiving article access request, the attribute information of acquisition access user and user behavior data are (i.e. Step S101) before, need generate similarity matrix and store, may include that

Step 1, collection Item Information, the attribute information of user and user behavior data.

Step 2, based on conduction of heat algorithm, and according to the article indicated by described Item Information, the genus of user Property information and user behavior data, generate the similarity matrix preset.

It is to say, the recommendation apparatus of Item Information need first to collect Item Information, the attribute information of user and The data sources such as user behavior data, and learn heat conduction model based on these data sources, thus generate similarity Matrix.

It is understood that the thought of conduction of heat algorithm is that the relation of user and article analogizes to one two Figure, i.e. user is a category node, and article are another kind of nodes, there is even limit between user and article, but Even limit is there is not between user and user and between article and article, can be in the lump with reference to Fig. 1 c.Wherein, exist On the basis of bigraph (bipartite graph), conduction of heat algorithm, according to two kinds of diffusion principles of physics, can obtain article (item) Between strength of association, two kinds of diffusion principles are material diffusion respectively and heat conducts.Material spreads, and meets In law of conservation of energy, i.e. matter and energy diffusion process, gross energy keeps constant, it is intended to recommend popular item.Heat conducts, one or more constant temperature source drive, be unsatisfactory for preservation of energy, it is intended to recommend Unexpected winner item.Can be in the lump with reference to Fig. 1 d, for the process signal of the diffusion of concrete material and energy conduction.

The behavior of certain article of customer consumption is it can be understood as energy (Energy) or heat (Heat). As shown in (a) in Fig. 1 d, it is assumed that the recommendation ability of user's goods for consumption information is 1, non-goods for consumption information Recommendation ability be 0, as a example by the user (user) being labelled with *, for material spread, article first Its energy is divided equally to its user of post-consumer, and as Suo Shi (b), then oneself energy is divided equally to article by user, As Suo Shi (c), it is possible to obtain in the article that * user does not consumes, the highest scoring of the 3rd article, connect down Recommend the 3rd article to * user；Conducting for heat, first one user's temperature is bought equal to him Crossing the meansigma methods of article temperature, as Suo Shi (d), next each article temperature bought its equal to all User's temperature averages, as Suo Shi (e), it can be seen that in the article that * user does not consumes, it should push away to * user Recommend the 4th article.Then, according to material diffusion and two processes of heat conduction, meter can be respectively obtained Calculate similarity formula between article as follows:

Material spreads:

Heat conducts:

According to (1) and (2), aggregative formula can be obtained:

Wherein α, β identify two different article, k_α、k_βMarking articles α and article β is consumed number of times, k_j The article number of mark user's j post-consumer, a_αi、a_βiWhether mark user i goods for consumption α, β, be to be 1, it is otherwise 0.

In the embodiment of the present invention, user is regarded as the user in conduction of heat algorithm, by article and the genus of user Property information regard as item, according to aggregative formula (3), i.e. can get the similarity between article and article and Similarity between attribute information, thus obtain similarity matrix.

Further, based on default similarity matrix, described attribute information and user behavior data, Described candidate item is concentrated before determining and meeting pre-conditioned article (step S103), it is also possible to including:

According to the described attribute information accessing user and user behavior data, structuring user's characteristic vector.

Can be concrete, when the recommendation apparatus of Item Information recommends article for access user on line, first obtain Candidate item collection, then according to behavior and user's sex, age, the region etc. of the recent goods for consumption of user Attribute information, constitutes a user characteristics vector A.Such as, 27 years old Shenzhen Female downloaded " XX yesterday Street " App (Application, application program), then user characteristics vector A i.e. comprises [women, 27 Year, Shenzhen, XX street] such dimension.

Based on this, described based on default similarity matrix, described attribute information and user behavior data, Described candidate item is concentrated and is determined that meeting pre-conditioned article (i.e. step S103) can specifically include: base In default similarity matrix and described user characteristics vector, concentrate at described candidate item and determine satisfied presetting The article of condition.

In some embodiments, based on default similarity matrix and described user characteristics vector, described Candidate item is concentrated and is determined that meeting pre-conditioned article can specifically include:

Step A, according to described similarity matrix and described user characteristics vector, to described candidate item collection In article give a mark, obtain give a mark result.

Step B, according to described marking result, determine that target item, described target item are described candidate Article are concentrated, and corresponding marking result exceedes the article of preset fraction threshold value.

Step C, it is defined as meeting pre-conditioned article by described target item.

Further, step A can be specially described similarity matrix and described user characteristics vector It is multiplied, generates score vector, using described score vector as marking result.Thereafter according to marking result Determine the article exceeding preset fraction threshold value, namely article are carried out pre-sequence and screening, thus obtain Target item.

In some embodiments, based on default similarity matrix and described user characteristics vector, described Candidate item is concentrated and is determined that meeting pre-conditioned article can also specifically include:

Step a, according to described similarity matrix and described user characteristics vector, to described candidate item collection In article give a mark, obtain give a mark result.

Step b, according to described marking result, determine that target item, described target item are described candidate Article are concentrated, and corresponding marking result exceedes the article of preset fraction threshold value.

Step c, by this special regression model of logic preset, described target item is ranked up, obtains Target item after sequence.

Step d, it is defined as meeting pre-conditioned article by the target item after described sequence.

It is to say, after determining target item, by this special regression model of default logic to target item Carry out essence row, so that Item Information corresponding for the article being best suitable for user is recommended user.Wherein, step a Can be specially and described similarity matrix and described user characteristics vector are multiplied, generate score vector, Using described score vector as marking result.Thereafter, determine exceed preset fraction threshold value according to marking result Article are namely carried out pre-sequence and screening, thus obtain target item by article.

Preferably, at this special regression model of the logic by presetting, described target item is ranked up, Before target item after sequence, it is also possible to including: according to described Item Information, the attribute information of user And user behavior data, structural feature, described feature includes the intersection between article characteristics and article and user Feature.

It is understood that in the embodiment of the present invention, this special regression model of this logic can be set in advance in thing In the recommendation apparatus of product information, therefore this special regression model of the described logic by presetting, to described object Product are ranked up, and (step c) may include that the target item after being sorted

Step c1, by this special regression model of logic preset, described feature is learnt, generates feature Weight.

Step c2, based on described feature weight, described target item is ranked up, the mesh after being sorted Mark article.

The i.e. recommendation apparatus of Item Information can be in conjunction with Item Information, the attribute information of user and user behavior number According to, by Feature Engineering technology such as feature discretization and characteristic crossover, use this special regression model (LR of logic Model) learn, such that it is able to learn different characteristic weight, wherein these feature weights are to quantify weighing apparatus Measure the influence degree of each factor influential on customer consumption article.Recommendation apparatus at Item Information is determined After target item, according to the feature weight that study obtains, target item is ranked up, the mesh after being sorted Mark article.

In order to be best understood from technical solution of the present invention, below the basic principle of LR model is carried out simple analysis. Due in this recommendation method, only use as two disaggregated models, meet you and time be categorized as with two here Typical case introduces its principle, and many classification problems can extend on this basis.

First, given sample data set X and corresponding label vector Y, (x, y) is one of them sample, it is assumed that The sample following probabilistic model of obedience:

$p(y=\±1|x,w)=\frac{1}{1+\exp (-y(w^{T}x+b))}---\left(4\right)$

If training sample example is x_i, i=1,2,3 ..., l, y_i{ 1 ,-1}, then weight w and constant term b are for waiting to estimate for ∈ Meter parameter, by maximum Likelihood, can obtain following optimization aim:

$\underset{w,b}{min}\underset{i=1}{\overset{l}{\Σ}}log(1+e^{-y_i(w^T{x}_i+b)})$

For carrying out easier calculating, typically can do and simplify as follows:

$\begin{array}{cc}\[x_i^T,1\]\→x_i^T& \[w^T,b\]\→w^T\end{array}$

Simultaneously in order to obtain more preferably generalization ability, reduce over-fitting, a regular terms w can be increased^TW/2, Thus obtain final optimization aim:

$\underset{w}{min}f\left(w\right)=\frac{1}{2}{w}^{T}w+C\underset{i=1}{\overset{l}{\Σ}}log(1+e^{-y_i{w}^T{x}_i})---\left(5\right)$

It is understood that implement this optimization aim to obtain the method for w, b a lot, such as block cattle The methods of pausing etc., the most specifically introduce.

It is to say, use such a target function model as shown in formula (5), obtain weight w with Constant term b, and use the hypothesis (i.e. formula (4)) of LR model article are given a mark again and arranges Sequence, finally gives and meets pre-conditioned article, and wherein, feature refers to the x in formula (5), and feature is weighed The most i.e. refer to the w in formula (5).

In step S104, by the described Item Information met corresponding to pre-conditioned article to accessing use Family is recommended.

Such as, the recommendation apparatus of Item Information is determined described after meeting pre-conditioned article, first obtains this The Item Information that a little article are corresponding, such as the classification of article, the title of article, the manufacturer of article, article Then these Item Information are shown by size, article tag etc. information, to carry out article to user Recommend.

From the foregoing, the recommendation method of the Item Information of embodiment of the present invention offer, visit when receiving article When asking request, obtain attribute information and the user behavior data accessing user；Then, based on default similar Spend matrix and the attribute information accessing user got and user behavior data, concentrate at candidate item Determining and meet pre-conditioned article, described default similarity matrix is for indicating the phase between candidate item Like the similarity between degree and candidate item and attribute information；Finally, by meet pre-conditioned article to This access user recommends.I.e. the embodiment of the present invention is by similarity matrix, the attribute information of access user and use Family behavioral data, as in the recent period to the click of article difference and consuming behavior etc., realized accessing user anticipating Figure prediction, wherein between similarity moment matrix representation article between similarity and article and user base attribute Similarity, may thereby determine that out the article of applicable user and recommends, and uses relative to based on Linear Model for Prediction Family exposes the mode that conversion ratio score is recommended to article, not only improves the personalization of recommendation results, Improve recommendation effect, also improve the accuracy of recommendation results.

Second embodiment

According to the method described by first embodiment, below citing is described in further detail.

First, the recommendation method of Item Information that the embodiment of the present invention provides, can build at Hadoop and In Spark Distributed Calculation and storage cluster.Secondly, in this embodiment, the recommendation apparatus of Item Information passes through User view is predicted by conduction of heat algorithm, and by LR algorithm, article is carried out screening and sequencing, Obtain eventually being best suitable for the article of user and recommending.Hereinafter will be described in more detail.

Refer to the stream of the recommendation method of the Item Information that Fig. 2 a, Fig. 2 a provides for second embodiment of the invention Journey schematic diagram.Described method includes step S21, obtains data source；Step S22, carry out according to data source Calculated off line；Step S23, online marking and sequence, and export recommendation results.

Step S21, acquisition data source.

The recommendation apparatus of Item Information collects Item Information, the attribute information of user and user behavior data, and From these data source learning LR model and heat conduction model.

Wherein, can to refer specifically to Item Title, goods categories, article head portrait, article big for described Item Information The information such as little, article tag, the attribute information of user refer mainly to the demographics such as sex, age, region and belong to Property information, described user behavior data refer mainly to user browse at PC end or cell-phone customer terminal, click on, Buy, download, the user behaviors log such as installation.

Step S22, carry out calculated off line according to data source.

In the embodiment of the present invention, the recommendation apparatus of Item Information carries out calculated off line according to the data source obtained:

(1) similarity matrix is calculated based on conduction of heat algorithm

User is regarded as the user in conduction of heat algorithm, article and customer attribute information is regarded as item, so Rear use user behavior data and customer attribute information, by conduction of heat computing formula (i.e. aforementioned formula (3)), Can be obtained by between article similarity between similarity and article and customer attribute information, thus constitute phase Like degree matrix S:

$S=\left[\begin{array}{c}{s}_{11},s_{12},....,s_{1(M+N)}\\ s_{21},s_{22},...,s_{2(M+N)}\\ s_{31},s_{32},...,s_{3(M+N)}\\ ...\\ s_{(M+N)1},...,s_{(M+N)(M+N)}\end{array}\right]$

Assume that article Candidate Set has N number of article (N is the integer more than 0), customer attribute information M altogether Dimension (M is the integer more than 0), owing to can regard as article by customer attribute information in this embodiment, then It is believed that be total to M+N article, it is assumed that Item Number 1,2,3 ..., (M+N), then similarity matrix S is one Individual (M+N) * (M+N) square formation, and element s therein_ijExpression article i and the similarity of article j, such as, The article high with male's similarity-rough set, it is meant that, male prefers to consume this kind of article.

(2) feature weight is calculated according to LR model.

The recommendation apparatus of Item Information combines Item Information, the attribute information of user and user behavior data, logical Cross the Feature Engineering technology such as feature discretization and characteristic crossover, construct magnanimity feature, use LR model just may be used To learn different characteristic weight, wherein these feature weights are to quantify to weigh to have an impact customer consumption article The influence degree of each factor.

Step S23, online marking and sequence, and export recommendation results.

As shown in Figure 2 a, when the recommendation apparatus of Item Information recommends article for user on line, first take out institute Have a candidate item collection, then according to access the behavior (i.e. user behavior data) of the recent goods for consumption of user with And the attribute information such as user's sex, age, region, constitute a user characteristics vector A, A and similarity Matrix S-phase is taken advantage of, then can obtain the score vector Score of (M+N) individual article.

Score_1×(M+N)=A_1×(M+N)·S_(M+N)×(M+N)

Such as, 27 years old Shenzhen Female downloaded " XX street " App (application program) yesterday, then use Family characteristic vector A i.e. comprises [women, 27 years old, Shenzhen, XX street] such dimension.Then by this Individual user characteristics vector A is multiplied with similarity matrix S, obtains the score vector of (M+N) individual article Score。

It is assumed that certain female user has consumed article 1, then article 1 and this feature similarity of women, female in the recent period Property deflection buy article branch higher；This female consumption article 1, then similar with article 1 simultaneously Article score also can be higher, i.e. similar with article 1 article are regarded as have purchased 1 and buying Article.And the article of comprehensive highest scoring, i.e. speculate it is this women thing of preferring in the recent period or paying close attention to Product.It is to say, first pass through similarity matrix and user characteristics vector A carries out user view prediction, so After to extract one group of article of highest scoring from Score vector be i.e. the one group of article being best suitable for user in the recent period, Use LR model (i.e. feature weight) to be ranked up these group article, obtain final ranking results.

The recommendation apparatus of Item Information, according to the final ranking results of article, first obtains these article corresponding Item Information, such as the classification of article, the title of article, manufacturer etc. the information of article, then by these things Product information shows, to carry out article recommendation to user.

The recommendation method of this Item Information that the embodiment of the present invention provides can be applicable to apply store, such as Fig. 2 b Shown in, some behaviors can browse according to user, click on, buy, downloaded and attribute information, it is recommended that It is best suitable for the application program of user, such that it is able to realize turning the exposure of discovery homepage (or software home page etc.) Rate (ctr) promotes more than 15% to 20%, and exposure conversion ratio refers to that article are in application searches homepage one day Download time and the ratio of exposure frequency.

Comprehensive abovementioned steps understands, and the present invention is by anticipating linear model (i.e. LR model) front increase user Figure prediction interval, i.e. according to customer attribute information and clicks on and consuming behavior article difference in the recent period, uses heat biography Lead the similarity between similarity and article and customer attribute information, then basis between algorithm calculating article User behavior data and user characteristics vector sum similarity matrix are multiplied, obtain article to be recommended marking roughly select beat Point, find the one group of personalization article being best suitable for recommending to user, then in LR model tormulation limit of power Interior limited recommendation set is ranked up.

From the foregoing, the recommendation method of the Item Information of embodiment of the present invention offer, visit when receiving article When asking request, obtain attribute information and the user behavior data accessing user；Then, based on default similar Spend matrix and the attribute information accessing user got and user behavior data, concentrate at candidate item Determining and meet pre-conditioned article, described default similarity matrix is for indicating the phase between candidate item Like the similarity between degree and candidate item and attribute information；Finally, by meet pre-conditioned article to This access user recommends.I.e. the embodiment of the present invention is by similarity matrix, the attribute information of access user and use Family behavioral data, as in the recent period to the click of article difference and consuming behavior etc., realized accessing user anticipating Figure prediction, say, that increase by one layer of user view prediction before LR model, use heat conduction model first Candidate Set is carried out individualized selection, then transfers to LR, LR to play it select one group of personalization article Measurement ability on the factor that various impacts are recommended, carries out essence row, may thereby determine that out the thing of applicable user Product are also recommended, and recommend the exposure conversion ratio score of article relative to based on Linear Model for Prediction user Mode, not only improve the personalization of recommendation results, improve recommendation effect, also improve the standard of recommendation results Really property.

3rd embodiment

For ease of preferably implementing the recommendation method of the Item Information that the embodiment of the present invention provides, the present invention implements Example also provides for the device of a kind of recommendation method based on above-mentioned Item Information.The wherein implication of noun and above-mentioned thing In the method for the recommendation of product information identical, implement the explanation that details is referred in embodiment of the method.

The structure of the recommendation apparatus referring to the Item Information that Fig. 3 a, Fig. 3 a provides for the embodiment of the present invention is shown Being intended to, the recommendation apparatus of wherein said Item Information can include that the first acquiring unit 301, second obtains single Unit 302, determine unit 303 and recommendation unit 304.

Wherein said first acquiring unit 301, for when receiving article access request, obtains to access and uses The attribute information at family and user behavior data；Second acquisition unit 302, is used for obtaining corresponding candidate item Collection.

Such as, accessing user can be by cell-phone customer terminal or PC (PC) end etc. to Item Information Recommendation apparatus send article access request, the recommendation apparatus of Item Information, according to this article access request, obtains Take attribute information and the user behavior data of this access user.When by terminal applies (as mobile phone terminal apply business City) when receiving article access request, this terminal applies corresponding item information data storehouse can be got, And the Item Information in this item information data storehouse is set as candidate item collection.

In the embodiment of the present invention, described article can refer specifically to the application program in mobile phone terminal application store, purchase Article provided on thing platform etc., described Item Information can refer specifically to Item Title, goods categories, thing The information such as product size, article tag, are not especially limited herein.

The described attribute information accessing user is primarily referred to as the demographic attributes such as user's sex, age, region； Described user behavior data refer mainly to user browse at PC end or cell-phone customer terminal, click on, buy, under The user behaviors logs such as load, installation.

Determine unit 303, for based on default similarity matrix, the attribute information of described access user and User behavior data, determines in described candidate item concentration and meets pre-conditioned article, described default phase Similar for indicate between the similarity between candidate item and candidate item and attribute information like degree matrix Degree；Recommendation unit 304, is used for the described Item Information met corresponding to pre-conditioned article to access User recommends.

The recommendation apparatus of the Item Information provided for the embodiment of the present invention please also refer to Fig. 3 b, Fig. 3 b another One structural representation；The recommendation apparatus of described Item Information can also include collector unit 305 and generate single Unit 306, is used for previously generating similarity matrix and storing:

Wherein said collector unit 305, for collecting Item Information, the attribute information of user and user behavior Data；Described signal generating unit 306, is used for based on conduction of heat algorithm, and according to indicated by described Item Information Article, the attribute information of user and user behavior data, generate preset similarity matrix.

It is to say, the recommendation apparatus of Item Information need first to collect Item Information, the attribute information of user and The data sources such as user behavior data, and learn heat conduction model based on these data sources, thus generate similarity Matrix.

It is understood that the thought of conduction of heat algorithm is that the relation of user and article analogizes to one two Figure, i.e. user is a category node, and article are another kind of nodes, there is even limit between user and article, but Even limit is there is not between user and user and between article and article.

The behavior of certain article of customer consumption is it can be understood as energy (Energy) or heat (Heat). As shown in (a) in Fig. 1 d, it is assumed that the recommendation ability of user's goods for consumption information is 1, non-goods for consumption information Recommendation ability be 0, as a example by the user (user) being labelled with *, for material spread, article first Its energy is divided equally to its user of post-consumer, and as Suo Shi (b), then oneself energy is divided equally to article by user, As Suo Shi (c), it is possible to obtain in the article that * user does not consumes, the highest scoring of the 3rd article, connect down Recommend the 3rd article to * user；Conducting for heat, first one user's temperature is bought equal to him Crossing the meansigma methods of article temperature, as Suo Shi (d), next each article temperature bought its equal to all User's temperature averages, as Suo Shi (e), it can be seen that in the article that * user does not consumes, it should push away to * user Recommend the 4th article.Then, according to material diffusion and two processes of heat conduction, can obtain calculating thing The aggregative formula of similarity between product:

Further, the recommendation apparatus of described Item Information can also include the first structural unit 307, is used for According to the described attribute information accessing user and user behavior data, structuring user's characteristic vector.

Can be concrete, the recommendation apparatus of Item Information, when recommending article for access user on line, first obtains Take candidate item collection, then according to behavior and user's sex, age, the region of the recent goods for consumption of user Deng attribute information, constitute a user characteristics vector A.Such as, 27 years old Shenzhen Female was downloaded yesterday " XX street " App (application program), then user characteristics vector A i.e. comprises [women, 27 years old, deeply Ditch between fields, XX street] such dimension.

Based on this, described determine unit 303 for: based on default similarity matrix and described user characteristics Vector, determines in described candidate item concentration and meets pre-conditioned article.

In some embodiments, described determine unit 303 can include give a mark subelement 3031 and first true Stator unit 3032, for based on default similarity matrix and described user characteristics vector, described candidate Article are concentrated and are determined and meet pre-conditioned article.

Can be concrete, described marking subelement 3031, for according to described similarity matrix and described user Characteristic vector, the article concentrating described candidate item are given a mark, and obtain result of giving a mark；

Described first determines subelement 3032, for according to described marking result, determines target item, by institute State target item and be defined as meeting pre-conditioned article, described target item for concentrate at described candidate item, Corresponding marking result exceedes the article of preset fraction threshold value.

Further, in this embodiment, described marking subelement 3031 can be specifically for: by described Similarity matrix and described user characteristics vector are multiplied, and generate score vector, by described score vector As marking result.

In some embodiments, described determine unit 303 can include give a mark subelement 3031, second true Stator unit 3033 and sequence subelement 3034, for based on default similarity matrix and described user Characteristic vector, determines in described candidate item concentration and meets pre-conditioned article.

Described marking subelement 3031, for vectorial according to described similarity matrix and described user characteristics, The article concentrating described candidate item are given a mark, and obtain result of giving a mark；Can be concrete, by described similar Degree matrix and described user characteristics vector be multiplied, generate score vector, using described score vector as Marking result.

Described second determines subelement 3033, for according to described marking result, determines target item, described Target item is for concentrate at described candidate item, and corresponding marking result exceedes the article of preset fraction threshold value； Described sequence subelement 3034, for by this special regression model of default logic, entering described target item Row sequence, the target item after being sorted；

Described second determines that subelement 3033 is additionally operable to: be defined as the target item after described sequence meeting in advance If the article of condition.

It is to say, after determining target item, by this special regression model of default logic to target item Carry out essence row, so that Item Information corresponding for the article being best suitable for user is recommended user.

It is understood that in the embodiment of the present invention, this special regression model of this logic can be set in advance in thing In the recommendation apparatus of product information, the recommendation apparatus of the most described Item Information can also include the second structural unit 308, for according to described Item Information, the attribute information of user and user behavior data, structural feature, Described feature includes the cross feature between article characteristics and article and user.

Based on this, described sequence subelement 3034 may be used for: by this special regression model of default logic, Described feature is learnt, generates feature weight, based on described feature weight, described target item is entered Row sequence, the target item after being sorted.

The i.e. recommendation apparatus of Item Information can be in conjunction with Item Information, the attribute information of user and user behavior number According to, by Feature Engineering technology such as feature discretization and characteristic crossover, use this special regression model (LR of logic Model) learn, such that it is able to learn different characteristic weight, wherein these feature weights are to quantify weighing apparatus Measure the influence degree of each factor influential on customer consumption article.

It is understood that use the such a target function model as shown in formula (5), obtain weight W and constant term b, and use the hypothesis (i.e. formula (4)) of LR model article are given a mark and arranges Sequence, finally gives and meets pre-conditioned article, and wherein, feature refers to the x in formula (5), and feature is weighed The most i.e. refer to the w in formula (5).

After determining and meeting pre-conditioned article, it is recommended that unit 304, for default bar described will be met Item Information corresponding to the article of part is recommended to accessing user.

Such as, the recommendation apparatus of Item Information is determined described after meeting pre-conditioned article, first obtains this The Item Information that a little article are corresponding, such as the classification of article, the title of article, manufacturer etc. the information of article, Then these Item Information are shown, to carry out article recommendation to user.

It should be noted that conduction of heat algorithm and this special regression model principle of logic refer to first embodiment Related content, here is omitted.

When being embodied as, above unit can realize as independent entity, it is also possible to carries out arbitrarily Combination, realizes as same or several entities, and being embodied as of above unit can be found in above Embodiment of the method, does not repeats them here.

The recommendation apparatus of this Item Information specifically can be integrated in the network equipment such as server or gateway.

From the foregoing, the recommendation apparatus of the Item Information of embodiment of the present invention offer, visit when receiving article When asking request, obtain attribute information and the user behavior data accessing user；Then, based on default similar Spend matrix and the attribute information accessing user got and user behavior data, concentrate at candidate item Determining and meet pre-conditioned article, described default similarity matrix is for indicating the phase between candidate item Like the similarity between degree and candidate item and attribute information；Finally, by meet pre-conditioned article to This access user recommends.I.e. the embodiment of the present invention is by similarity matrix, the attribute information of access user and use Family behavioral data, as in the recent period to the click of article difference and consuming behavior etc., realized accessing user anticipating Figure prediction, wherein between similarity moment matrix representation article between similarity and article and user base attribute Similarity, may thereby determine that out the article of applicable user and recommends, and uses relative to based on Linear Model for Prediction Family exposes the mode that conversion ratio score is recommended to article, not only improves the personalization of recommendation results, Improve recommendation effect, also improve the accuracy of recommendation results.

4th embodiment

The embodiment of the present invention also provides for a kind of server, wherein can be with the Item Information of the integrated embodiment of the present invention Recommendation apparatus, as shown in Figure 4, it illustrates the structural representation of server involved by the embodiment of the present invention Figure, specifically:

This server can include one or the processor 401, or of more than one process core The memorizer 402 of above computer-readable recording medium, radio frequency (Radio Frequency, RF) circuit 403, The parts such as power supply 404, input block 405 and display unit 406.It will be understood by those skilled in the art that Server architecture shown in Fig. 4 is not intended that the restriction to server, can include more more or more than diagram Few parts, or combine some parts, or different parts are arranged.Wherein:

Processor 401 is the control centre of this server, utilizes various interface and the whole server of connection Various piece, by run or perform be stored in the software program in memorizer 402 and/or module, and Call the data being stored in memorizer 402, perform the various functions of server and process data, thus right Server carries out integral monitoring.Optionally, processor 401 can include one or more process core；Preferably , processor 401 can integrated application processor and modem processor, wherein, application processor is main Processing operating system, user interface and application program etc., modem processor mainly processes radio communication. It is understood that above-mentioned modem processor can not also be integrated in processor 401.

Memorizer 402 can be used for storing software program and module, and processor 401 is stored in by operation The software program of reservoir 402 and module, thus perform the application of various function and data process.Memorizer 402 can mainly include store program area and storage data field, wherein, storage program area can store operating system, Application program (such as sound-playing function, image player function etc.) etc. needed at least one function；Deposit Storage data field can store the data etc. that the use according to server is created.Additionally, memorizer 402 can wrap Include high-speed random access memory, it is also possible to include nonvolatile memory, for example, at least one disk storage Device, flush memory device or other volatile solid-state parts.Correspondingly, memorizer 402 can also wrap Include Memory Controller, to provide the processor 401 access to memorizer 402.

During RF circuit 403 can be used for receiving and sending messages, the reception of signal and transmission, especially, by base station Downlink information receive after, transfer to one or more than one processor 401 process；It addition, will relate to The data of row are sent to base station.Generally, RF circuit 403 include but not limited to antenna, at least one amplifier, Tuner, one or more agitator, subscriber identity module (SIM) card, transceiver, bonder, Low-noise amplifier (LNA, Low Noise Amplifier), duplexer etc..Additionally, RF circuit 403 Can also be communicated with network and other equipment by radio communication.Described radio communication can use arbitrary communication Standard or agreement, include but not limited to global system for mobile communications (GSM, Global System of Mobile Communication), general packet radio service (GPRS, General Packet Radio Service), CDMA (CDMA, Code Division Multiple Access), WCDMA (WCDMA, Wideband Code Division Multiple Access), Long Term Evolution (LTE, Long Term Evolution), Email, Short Message Service (SMS, Short Messaging Service) etc..

Server also includes the power supply 404 (such as battery) powered to all parts, it is preferred that power supply can With logically contiguous with processor 401 by power-supply management system, thus realize management by power-supply management system The functions such as charging, electric discharge and power managed.Power supply 404 can also include one or more directly Stream or alternating current power supply, recharging system, power failure detection circuit, power supply changeover device or inverter, electricity The random component such as source positioning indicator.

This server may also include input block 405, and this input block 405 can be used for receiving the numeral of input Or character information, and produce the keyboard relevant with user setup and function control, mouse, action bars, Optics or the input of trace ball signal.

This server may also include display unit 406, and this display unit 406 can be used for display and inputted by user Information or be supplied to the information of user and the various graphical user interface of server, these graphical users connect Mouth can be made up of figure, text, icon, video and its combination in any.Display unit 406 can include Display floater, optionally, can use liquid crystal display (LCD, Liquid Crystal Display), The forms such as Organic Light Emitting Diode (OLED, Organic Light-Emitting Diode) configure display surface Plate.

Concrete the most in the present embodiment, the processor 401 in server can according to following instruction, by one or The executable file that the process of more than one application program is corresponding is loaded in memorizer 402, and by processing Device 401 runs the application program being stored in memorizer 402, thus realizes various function, as follows:

When receiving article access request, obtain attribute information and the user behavior data accessing user；Obtain Take corresponding candidate item collection；Based on default similarity matrix, the attribute information of described access user and use Family behavioral data, concentrates at described candidate item and determines and meet pre-conditioned article, described default similar Degree matrix is similar for indicate between the similarity between candidate item and candidate item and attribute information Degree；The described Item Information met corresponding to pre-conditioned article is recommended to accessing user.

Preferably, described processor 401 can be also used for, collect Item Information, the attribute information of user and User behavior data；Based on conduction of heat algorithm, and according to the article indicated by described Item Information, user Attribute information and user behavior data, generate the similarity matrix preset.

Preferably, described processor 401 can be also used for, according to the described attribute information accessing user and use Family behavioral data, structuring user's characteristic vector；Based on default similarity matrix and described user characteristics vector, Determine in described candidate item concentration and meet pre-conditioned article.

Preferably, described processor 401 can be also used for, according to described similarity matrix and described user Characteristic vector, the article concentrating described candidate item are given a mark, and obtain result of giving a mark；Beat according to described Point result, determines target item, described target item for concentrate at described candidate item, knot of giving a mark accordingly Fruit exceedes the article of preset fraction threshold value；It is defined as meeting pre-conditioned article by described target item.

Preferably, described processor 401 can be also used for, according to described similarity matrix and described user Characteristic vector, the article concentrating described candidate item are given a mark, and obtain result of giving a mark；Beat according to described Point result, determines target item, described target item for concentrate at described candidate item, knot of giving a mark accordingly Fruit exceedes the article of preset fraction threshold value；By this special regression model of default logic, to described target item It is ranked up, the target item after being sorted；It is defined as the target item after described sequence meeting and presets The article of condition.

Preferably, described processor 401 can be also used for, and believes according to the attribute of described Item Information, user Breath and user behavior data, structural feature, described feature includes the friendship between article characteristics and article and user Fork feature；By this special regression model of default logic, described feature is learnt, generates feature weight, Based on described feature weight, described target item is ranked up, the target item after being sorted.

Preferably, described processor 401 can be also used for, by special to described similarity matrix and described user Levy vector to be multiplied, generate score vector, using described score vector as marking result.

From the foregoing, in the server of embodiment of the present invention offer, when receiving article access request, Obtain attribute information and the user behavior data accessing user；Then, based on default similarity matrix, with And get access the attribute information of user and user behavior data, candidate item concentrate determine meet pre- If the article of condition, described default similarity matrix is for indicating the similarity between candidate item and time Select the similarity between article and attribute information；Finally, pre-conditioned article will be met to this access user Recommend.I.e. the embodiment of the present invention is by similarity matrix, the attribute information of access user and user behavior data, As in the recent period to the click of article difference and consuming behavior etc., realized user is accessed carrying out Intention Anticipation, wherein Similarity between similarity and article and user base attribute between similarity moment matrix representation article, thus Can determine that the article of applicable user and recommend, relative to based on the Linear Model for Prediction user exposure to article Light conversion ratio score carries out the mode recommended, and not only improves the personalization of recommendation results, improves and recommends effect Really, the accuracy of recommendation results is also improved.

In the above-described embodiments, the description to each embodiment all emphasizes particularly on different fields, in certain embodiment the most in detail The part stated, may refer to the detailed description of the recommendation method above with respect to Item Information, and here is omitted.

The recommendation apparatus of described Item Information that the embodiment of the present invention provides, be such as computer, panel computer, Mobile phone with touch function etc., the recommendation apparatus of described Item Information and the article letter in foregoing embodiments The recommendation method of breath belongs to same design, can run described article on the recommendation apparatus of described Item Information The either method provided in the recommendation embodiment of the method for information, it implements process and refers to described Item Information Recommendation embodiment of the method, here is omitted.

It should be noted that for the recommendation method of Item Information of the present invention, this area common test Personnel are appreciated that all or part of flow process of the recommendation method realizing Item Information described in the embodiment of the present invention, Can be by the hardware that computer program controls to be correlated with to complete, described computer program can be stored in one In computer read/write memory medium, as being stored in the memorizer of terminal, and by least in this terminal Individual processor performs, and can include the stream of the embodiment of the recommendation method such as described Item Information in the process of implementation Journey.Wherein, described storage medium can be magnetic disc, CD, read only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory) etc..

For the recommendation apparatus of the described Item Information of the embodiment of the present invention, its each functional module can be integrated Process in chip at one, it is also possible to be that modules is individually physically present, it is also possible to two or more Module is integrated in a module.Above-mentioned integrated module both can realize to use the form of hardware, it is also possible to The form using software function module realizes.If described integrated module is real with the form of software function module Now and as independent production marketing or use time, it is also possible to be stored in a computer read/write memory medium In, described storage medium is such as read only memory, disk or CD etc..

The recommendation method and device of a kind of Item Information provided the embodiment of the present invention above has been carried out in detail Introducing, principle and the embodiment of the present invention are set forth by specific case used herein, above reality The explanation executing example is only intended to help to understand method and the core concept thereof of the present invention；Simultaneously for this area Technical staff, according to the thought of the present invention, the most all can change it Place, in sum, this specification content should not be construed as limitation of the present invention.

Claims (14)

1. the recommendation method of an Item Information, it is characterised in that including:

When receiving article access request, obtain attribute information and the user behavior data accessing user；

Obtain corresponding candidate item collection；

Based on default similarity matrix, the attribute information of described access user and user behavior data, in institute Stating candidate item to concentrate and determine and meet pre-conditioned article, described default similarity matrix is used for indicating time Select the similarity between the similarity between article and candidate item and attribute information；

The described Item Information met corresponding to pre-conditioned article is recommended to accessing user.

The recommendation method of Item Information the most according to claim 1, it is characterised in that described when receiving During to article access request, before obtaining attribute information and the user behavior data accessing user, also include:

Collect Item Information, the attribute information of user and user behavior data；

Based on conduction of heat algorithm, and according to the article indicated by described Item Information, the attribute information of user and User behavior data, generates the similarity matrix preset.

The recommendation method of Item Information the most according to claim 1, it is characterised in that described based in advance If similarity matrix, described attribute information and user behavior data, described candidate item concentrate determine full Before the article that foot is pre-conditioned, also include:

According to the described attribute information accessing user and user behavior data, structuring user's characteristic vector；

Described based on default similarity matrix, described attribute information and user behavior data, described candidate Article are concentrated and are determined that meeting pre-conditioned article includes: special based on default similarity matrix and described user Levy vector, determine in described candidate item concentration and meet pre-conditioned article.

The recommendation method of Item Information the most according to claim 3, it is characterised in that described based in advance If similarity matrix and described user characteristics vector, described candidate item concentrate determine meet pre-conditioned Article include:

According to described similarity matrix and described user characteristics vector, the article that described candidate item is concentrated Give a mark, obtain result of giving a mark；

According to described marking result, determine target item, described target item for concentrate at described candidate item, Corresponding marking result exceedes the article of preset fraction threshold value；

It is defined as meeting pre-conditioned article by described target item.

The recommendation method of Item Information the most according to claim 3, it is characterised in that described based in advance If similarity matrix and described user characteristics vector, described candidate item concentrate determine meet pre-conditioned Article include:

According to described similarity matrix and described user characteristics vector, the article that described candidate item is concentrated Give a mark, obtain result of giving a mark；

According to described marking result, determine target item, described target item for concentrate at described candidate item, Corresponding marking result exceedes the article of preset fraction threshold value；

By this special regression model of default logic, described target item is ranked up, after being sorted Target item；

It is defined as meeting pre-conditioned article by the target item after described sequence.

The recommendation method of Item Information the most according to claim 5, it is characterised in that described by advance If this special regression model of logic, described target item is ranked up, the target item after being sorted it Before, also include:

According to described Item Information, the attribute information of user and user behavior data, structural feature, described spy Levy the cross feature included between article characteristics and article and user；

Described by this special regression model of default logic, described target item is ranked up, is sorted After target item include: by this special regression model of default logic, described feature is learnt, raw Become feature weight, based on described feature weight, described target item is ranked up, the mesh after being sorted Mark article.

7. according to the recommendation method of the Item Information described in claim 4 or 5, it is characterised in that described According to described similarity matrix and described user characteristics vector, the article concentrating described candidate item carry out beating Point, obtain result of giving a mark, including:

Described similarity matrix and described user characteristics vector are multiplied, generate score vector, by institute State score vector as marking result.

8. the recommendation apparatus of an Item Information, it is characterised in that including:

First acquiring unit, for when receiving article access request, obtains the attribute information accessing user And user behavior data；

Second acquisition unit, is used for obtaining corresponding candidate item collection；

Determine unit, for based on default similarity matrix, the attribute information of described access user and user Behavioral data, determines in described candidate item concentration and meets pre-conditioned article, described default similarity Matrix is for indicating the similarity between the similarity between candidate item and candidate item and attribute information；

Recommendation unit, is used for the described Item Information met corresponding to pre-conditioned article to accessing user Recommend.

The recommendation apparatus of Item Information the most according to claim 8, it is characterised in that described device is also Including:

Collector unit, for collecting Item Information, the attribute information of user and user behavior data；

Signal generating unit, is used for based on conduction of heat algorithm, and according to the article indicated by described Item Information, use The attribute information at family and user behavior data, generate the similarity matrix preset.

The recommendation apparatus of Item Information the most according to claim 8, it is characterised in that described device Also include:

First structural unit, for according to the described attribute information accessing user and user behavior data, structure User characteristics vector；

Described determine unit for: based on default similarity matrix and described user characteristics vector, described Candidate item is concentrated and is determined and meet pre-conditioned article.

The recommendation apparatus of 11. Item Information according to claim 10, it is characterised in that described determine Unit includes:

Marking subelement, for according to described similarity matrix and described user characteristics vector, to described time The article selecting article to concentrate are given a mark, and obtain result of giving a mark；

First determines subelement, for according to described marking result, determines target item, by described object Product are defined as meeting pre-conditioned article, and described target item is for concentrate at described candidate item, accordingly Marking result exceedes the article of preset fraction threshold value.

The recommendation apparatus of 12. Item Information according to claim 10, it is characterised in that described determine Unit includes:

Marking subelement, for according to described similarity matrix and described user characteristics vector, to described time The article selecting article to concentrate are given a mark, and obtain result of giving a mark；

Second determines subelement, for according to described marking result, determines target item, described target item For concentrating at described candidate item, corresponding marking result exceedes the article of preset fraction threshold value；

Sequence subelement, for by this special regression model of default logic, arranging described target item Sequence, the target item after being sorted；

Described second determines that subelement is additionally operable to: is defined as the target item after described sequence meeting and presets bar The article of part.

The recommendation apparatus of 13. Item Information according to claim 12, it is characterised in that described device Also include:

Second structural unit, is used for according to described Item Information, the attribute information of user and user behavior data, Structural feature, described feature includes the cross feature between article characteristics and article and user；

Described sequence subelement is used for: by this special regression model of default logic, to described feature Practise, generate feature weight, based on described feature weight, described target item is ranked up, is sorted After target item.

14. according to the recommendation apparatus of the Item Information described in claim 11 or 12, it is characterised in that institute State marking subelement for: described similarity matrix and described user characteristics vector are multiplied, generate Score vector, using described score vector as marking result.