CN105809275A - Item scoring prediction method and apparatus - Google Patents

Abstract

The invention provides an item scoring prediction method and an apparatus which relate to the technical field of data processing to address the problem in the prior art of failing to have items recommended to users accurately due to the fact that a traditional recommendation system cannot tell similar users for a user. The method is performed through the following steps: obtaining the scoring information of a user to one item; based on the scoring information of the user to the item, generating a user-item type scoring matrix; determining targeted users and similar users based on the user-item type scoring matrix; based on the user-item type scoring matrix and the scoring information of the user to the item, calculating the predicted scorings of an to-be-scored item by the similar users of the targeted users and of the to-be-scored item of targeted users by similar users without scoring; and finally, based on the scoring to the to-be-scored item of targeted users by the similar users of the target users, scoring the to-be-scored items. The invention is provided to recommend items.

Description

A kind of article score in predicting method and device

Technical field

The present invention relates to data processing field, particularly relate to a kind of article score in predicting method and device.

Background technology

User is to one of important component part that the scoring of article is article commending system, and it can directly embody user couple The fancy grade of article.General, we can to the scoring of different article or different user be to identical by analyzing user The scoring of article, and combine other because of usually catch unique user hobby or multiple user between hobby similarity, from And realize providing the user good recommendation service.

But, recently as developing rapidly of the Internet and ecommerce, number of users and number of articles all become The hugest numeral, and these two groups of huge numerals are combined into huger user-article rating matrix, but, by The article that can contact in each user are limited, beaten by user and undue can only account for minority, so that these user-article are commented Most numbers in sub matrix present vacancy, and then it is higher openness, so that this user-article rating matrix is had When data recommendation system predicts user to the scoring of a certain article, owing to the scoring overlap between user is less, therefore, pass through The obvious accuracy of scoring that the score data of similar users is predicted for a certain article of user is the highest.Such as, when two use When common article are not marked by family, then similarity between the two cannot calculate the most at all or directly be considered without similar Part, other similarities the most that may be present the most just cannot be caught in, and then causes the article commending system cannot These data are utilized to carry out score in predicting accurately.

Summary of the invention

Embodiments of the invention provide a kind of article score in predicting method and device, solve due to existing user-thing Judge sub matrix and have higher openness, and the inaccurate problem of score in predicting that the user caused is to article.

For reaching above-mentioned purpose, embodiments of the invention adopt the following technical scheme that

First aspect, it is provided that a kind of article score in predicting method, including:

Obtain user's score information to article；

According to described user, the score information of article is generated user-type of items rating matrix；

The similar users of described targeted customer is determined according to described user-type of items rating matrix；

The score information to article according to described user-type of items rating matrix and described user, calculates described mesh Described article mark are existed to be evaluated to described targeted customer of similar users that disappearance is marked by the similar users of mark user Divide the prediction scoring of article；

The scoring of the article to be marked to described targeted customer of the similar users according to described targeted customer, for described to be evaluated Divide article prediction scoring.

Second aspect, it is provided that a kind of article score in predicting device, including:

First acquisition module, for obtaining user's score information to article；

Generation module, generates use for the described user obtained according to described first acquisition module to the score information of article Family-type of items rating matrix；

Determine module, determine institute for the described user-type of items rating matrix generated according to described generation module State the similar users of targeted customer；

Computing module, for described user-type of items rating matrix of generating according to described generation module and described The described user that first acquisition module the obtains score information to article, calculates in the similar users of described targeted customer described There is the prediction scoring of the similar users article to be marked to described targeted customer of disappearance scoring in article to be marked；

Prediction module, for commenting of the similar users article to be marked to described targeted customer according to described targeted customer Point, for the prediction scoring of described article to be marked.

The article score in predicting method and device that embodiments of the invention provide, by believing the scoring of article according to user Breath generates user-type of items rating matrix, then determines the phase of targeted customer according to this user-type of items rating matrix Like user, the score information to article according to this user-type of items rating matrix and user, calculate the similar of targeted customer User treats the prediction scoring that scoring article exist the similar users article to be marked to targeted customer of disappearance scoring, After, according to the scoring of the similar users of this targeted customer article to be marked to targeted customer, for article to be marked prediction scoring. Carry out pre-compared to prior art only according to the disappearance scoring that the user-article rating matrix of shortage of data is targeted customer The method surveyed, this programme is by being aggregated to scattered article in different goods categories in advance, then in conjunction with existing data The user of disappearance-article rating matrix builds the greater concentration of user of data-goods categories rating matrix, thus improves and determine The accuracy of the similar users of targeted customer, and then can be the similar users of targeted customer to be treated scoring article there is disappearance The similar users article to be marked to targeted customer of scoring dope the higher scoring of accuracy, so use according to these targets The scoring of the similar users at family article to be marked to targeted customer, just can be accurately for article the to be marked prediction of targeted customer Go out scoring.

Accompanying drawing explanation

In order to be illustrated more clearly that the technical scheme of the embodiment of the present invention, below will be in embodiment or description of the prior art The required accompanying drawing used is briefly described, it should be apparent that, the accompanying drawing in describing below is only some realities of the present invention Execute example, for those of ordinary skill in the art, on the premise of not paying creative work, it is also possible to according to these accompanying drawings Obtain other accompanying drawing.

A kind of article score in predicting method flow diagram that Fig. 1 provides for the embodiment of the present invention；

The method flow diagram of the another kind of article score in predicting method that Fig. 2 provides for the embodiment of the present invention；

The structural representation of a kind of article score in predicting device that Fig. 3 provides for the embodiment of the present invention.

Detailed description of the invention

Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, complete Describe, it is clear that described embodiment is only a part of embodiment of the present invention rather than whole embodiments wholely.Based on Embodiment in the present invention, it is every other that those of ordinary skill in the art are obtained under not making creative work premise Embodiment, broadly falls into the scope of protection of the invention.

The executive agent of the article score in predicting method that the embodiment of the present invention provides can be article score in predicting device, or Person is for performing the terminal unit of above-mentioned article score in predicting method.Concrete, this mobile terminal can be intelligent television, intelligence Mobile phone, panel computer, notebook computer, Ultra-Mobile PC's (English: Ultra-mobile Personal Computer, is called for short: UMPC), net book, personal digital assistant (English: Personal Digital Assistant, be called for short: The terminal unit such as PDA).Wherein, article device can be that the central processing unit in above-mentioned terminal unit is (English: Central CPU) or can be the control unit in above-mentioned terminal unit or functional module Processing Unit, be called for short:.

The commodity that article in the present embodiment can be concrete (such as, need to recommend user to buy in online shopping mall Commodity) or shop (for example, it is desired to recommending the retail shop of user), it is also possible to it is many matchmakers such as video, audio frequency, picture, text document Body file.The most all it is illustrated with " article are videos ", and specifically thinks and " use for a certain target Family recommend article " as a example by illustrate.It should be noted that mentioned by it will be apparent to those skilled in the art that hereinafter " depending on Frequently " could alternatively be other any one multi-medium datas such as voice, picture, document.

Additionally, the scoring of article can directly be embodied user's fancy grade to article by user in the embodiment of the present invention, Under normal circumstances, the scoring of article can be showed by user with the numerical value in [0,10] interval, and numerical value is the highest, then it represents that user Scoring to article is the highest, and the lowest expression user is the lowest to the scoring of article.Certainly, the numerical value in above-mentioned [0,10] interval is only It is a kind of example, in actual applications, it is possible to use other mark shows user's scoring to article, does not the most limit Fixed, the scoring of article for convenience of explanation, is illustrated as a example by the numerical value in [0,10] interval by the present embodiment by user.

The terms "and/or", a kind of incidence relation describing affiliated partner, can there are three kinds of passes in expression System, such as, A and/or B, can represent: individualism A, there is A and B, individualism B these three situation simultaneously.It addition, herein Middle character "/", typicallys represent the forward-backward correlation relation to liking a kind of "or".

Embodiments of the invention provide a kind of article score in predicting method, as it is shown in figure 1, the method specifically includes following step Rapid:

101, article score in predicting device obtains user's score information to article.

Exemplary, the article in the present invention can be commodity, shop and multimedia file, and this multimedia file is permissible For multi-medium datas such as video, audio frequency, text documents.In the present embodiment, user includes the mark of user to the score information of article Type of items belonging to knowledge, the mark of article, these article, scoring time article are marked by user by the scoring of article and user Time.Wherein, the mark of above-mentioned user can be the login account of this user or other can uniquely represent the mark of this user Know, the present embodiment uses U1, U2, U3 ... Un form represents the mark of different user；The mark of above-mentioned article can be The title of these article or commercial product code or other can uniquely represent the mark of these article, the present embodiment uses V1, V2, V3 ..., Vm represent the mark of different article.

Each article in the present embodiment may belong to a type of items can also belong simultaneously to multiple type of items, example As, a certain video i.e. belongs to costume film and falls within action movie.The kind of the type of items in the present embodiment can be by technical staff It is set in advance, and determines the type of items belonging to each article.

Preferably, in a step 101, can arrange a update cycle, the length of update cycle can be according to article number Being set according to storehouse update status, for example, it is possible to be set to one month, one week or one day, this was not limited by the present invention, often Obtain user in described each update cycle in the individual update cycle to the score information of article and to be updated.Under the present embodiment State each step all to illustrate as a example by current period.

102, article score in predicting device generates user-type of items scoring square according to user to the score information of article Battle array.

Concrete, article score in predicting device generates user-article rating matrix R according to user to the score information of article With article-type of items relational matrix S, then according to user-article rating matrix R and article-type of items relational matrix S, Generate user-type of items rating matrix Z；Wherein, the row and column of above-mentioned matrix R represents the mark of user and article respectively Mark, the element of this matrix R is for representing user's scoring to article；The row and column of above-mentioned matrix S represents article respectively Mark and the mark of type of items, the element of this matrix S is for representing the type of items belonging to article；The row of above-mentioned matrix Z Represent that the mark of user and the mark of type of items, the element in this matrix Z are used for representing that user is to type of items respectively with row Scoring.

Exemplary, if as a example by article are as video, it is assumed that user's collection is combined into U={U₁,U₂,...,U_nAnd B={B₁, B₂,...,B_m, video collection is V={v₁,v₂,v₃,...,v_n, if using ID as row matrix, using video labeling as Rectangular array, then the relational matrix between definition user and article is matrixIf all of scoring of video is all by user Integer and time in the range of [0,10], 0 represents that this video is not marked by user.Here illustrate matrix R is contained with table 1 Justice, it should be noted that real matrix R has the dimension of million grades, following table 1 is only that the implication to matrix R is carried out Illustrate, be only a kind of example.

	v1	v2	v3	v4	v5	v6
							u1	0	0	6	2	5	7
u2	3	4	5	0	0	0
							u3	7	6	0	0	1	10
u4	8	0	7	5	0	0

Table 1

Exemplary, if video category set is L={l₁,l₂,l₃,l₄,l₅, when video classification K is 5, here can be with The implication of photopic vision frequency-video classification relational matrix for table 2, wherein, a video may belong to multiple classification, as shown in table 2, If video v_iBelong to classification l_jThen B_i,j=1, otherwise B_i,j=0.It should be noted that real matrix S has the dimension of million grades Degree, following table 2 is only that the implication to matrix S illustrates, and is only a kind of example.

	l1	l2	l3	l4	l5
						v1	1	1	0	0	1
v2	0	1	0	1	0
						v3	1	0	1	0	0
v4	0	0	0	1	0
						v5	1	1	1	0	1
v6	1	0	1	1	1

Table 2

By above-mentioned user-video rating matrix R and audio-video class relations matrix S, we can obtain such as following table User shown in 3-video classification rating matrix Z.It should be noted that real matrix Z has the dimension of million grades, following Table 3 is only that the implication to matrix Z illustrates, and is only a kind of example.

Table 3

Wherein, the element Z in above-mentioned table 3_i,jFor user u_iComment in undue video, belong to classification l_jVideo scoring flat Average.Such as, user u as shown in Table 1₁To video v₃,v₅And v₆All there is scoring, video v the most as shown in Table 2₃,v₅And v₆Belong to together In classification l₁, therefore, C_1,1Value be user u₁Meansigma methods 6 to the scoring of above three video, other are by that analogy.

103, article score in predicting device determines the similar use of targeted customer according to user-type of items rating matrix Family.

Concrete, article score in predicting device according to user-type of items rating matrix, calculate targeted customer and other Similarity between user, and the similarity between each user in targeted customer and other users is ranked up, determine Go out the similar users of targeted customer.Exemplary, article score in predicting device can based on existing calculating formula of similarity or User's calculating formula of similarity (formula one being described below) that the present embodiment is given, calculates targeted customer successively and is somebody's turn to do with removing The similarity between other users beyond targeted customer, and by the similarity between each user in targeted customer and other users It is ranked up, selects the height of number and similarity to determine the phase of targeted customer from high to low according to default similar users Like user.Such as, the similarity of all users calculated with targeted customer is arranged according to order from high to low, selects phase Like the highest front 30 users of property as the neighborhood of user u', it is designated as D_u'。

Exemplary, the embodiment of the present invention one user's calculating formula of similarity of offer:

$sim(u^{\′},u_i)=\frac{min\left(\right|G_{u^{\′},u_i}|,\mathrm{\θ})}{\mathrm{\θ}}\·\frac{\underset{k\∈G_{u^{\′},u_i}}{\Σ}(C_{u^{\′},k}-\stackrel{-}{r_{u^{\′}}})(C_{u_i,k}-\stackrel{-}{r_{u_i}})}{\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u^{\′},k}-\stackrel{-}{r_{u^{\′}}})}^2}\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u_i,k}-\stackrel{-}{r_{u_i}})}^2}}$ (formula one)

Wherein, above-mentioned sim (u', u_i) represent described targeted customer u' and other user u described_iSimilarity,For Targeted customer u' and other user u_iAll play the goods categories set belonging to undue article,For representing targeted customer u' couple The meansigma methods of the scoring of all items classification,For representing other users meansigma methods to the scoring of all items classification, C is User-type of items rating matrix.Additionally, above-mentioned θ is threshold value, the value of θ is the class number that video divides, if with table 2 As a example by understand, owing to the video classification in table 2 has 5 kinds of video classifications, θ takes 5 the most here, consequently, it is possible to as user u' and u_iDuring the video classification number deficiency jointly given a mark, the coefficient value of product will be less than 1, the just fall of the similarity between two users Low.And when the classification number that both gave a mark jointly is abundant, coefficient value is 1, the similarity between two users is constant, so By just alleviating, at threshold θ, the accuracy problems that too high estimation similarity brings, compared to existing calculating formula of similarity, User's similarity that the calculating formula of similarity that this enforcement is provided calculates is more accurate.

104, the scoring of article is believed by article score in predicting device according to user-type of items rating matrix and user Breath, calculates and treats scoring article in the similar users of targeted customer and there is to be evaluated to targeted customer of similar users of disappearance scoring Divide the prediction scoring of article.

Exemplary, due to similar users-article rating matrix openness, in order to use for target of targeted customer More article the most accurately are recommended at family, and the embodiment of the present invention is not marked the pre-of article by the similar users filling targeted customer Test and appraisal point, thus the similar users of completion similar users targeted customer-article rating matrix so that more article become target The similar users of user has been marked and article that targeted customer does not marks.

Concrete, step 104 specifically includes following steps:

A1, article score in predicting device obtain the homologue of the first article to be marked of first similar users of targeted customer Product collection.

Wherein, the similar article that above-mentioned similar article are concentrated be the first similar users marked and and article to be marked between Similarity more than the article of predetermined threshold.The above-mentioned similar users that the first similar users is targeted customer treats scoring thing There is the similar users of disappearance scoring in product.

Between each similar article that A2, article score in predicting device are concentrated to described similar article according to article to be marked Obtained by the scoring of each similar article that similar article are concentrated by similarity, the first similar users, the first article to be marked The meansigma methods of the scoring obtained by each similar article that the meansigma methods of scoring and similar article are concentrated, it was predicted that go out first similar User's scoring to the first article to be marked.

Understand based on hereinbefore described content, if article are as a example by video, when trying to achieve similar use for targeted customer u' Family collection, such as, D_u', and when similar users treats the problem that scoring video v' equally exists scoring disappearance, use similar users number It is predicted that targeted customer is the most inappropriate to the scoring of video v'.It would therefore be desirable to the similar users data tried to achieve It is filled with, obtains new similar users data E_u'.In advance, during we obtain video collection V most like with video v' before 30 videos, constitute set S_v', wherein, for any video v_k(v_k∈S_v'), its similarity with video v' with sim (v', v_k) represent.

Then, our similar users u to user u'_j(u_j∈ D) video v' is predicted that scoring can be based on following public affairs Formula two calculates, and the prediction calculated scoring is filled with.

$r_{u_j,v^{\′}}=\stackrel{-}{r_{v^{\′}}}+\frac{{\mathrm{\Σ}}_{v_k\∈S_{v^{\′}}}sim(v^{\′},v_k)\·(r_{u_j,v_k}-\stackrel{-}{r_{v_k}})}{{\mathrm{\Σ}}_{v_k\∈S_{v^{\′}}}\left|sim(v^{\′},v_k)\right|}$ (formula two)

Wherein, in above-mentioned formula twoWithIt is respectively video v and video v_kObtained by scoring meansigma methods (It it is exactly institute Have in user to have beats undue to video v', seeks the meansigma methods of these points,In like manner),Represent user u_jTo video v_kComment Point.For any video v_k(v_k∈S_v′), itself and similarity sim (v', the v of video v'_k) represent, and sim (v', v_k) permissible According to video v' and video v_kBelonging to video type and other feature score information (such as, area, language, keyword, drill Member, director etc.) calculate, the calculation of the similarity between calculating video is a lot (such as, Euclidean distance method) at present, this This is not limited by inventive embodiments.

Through above-mentioned steps, this article score in predicting device is by thing of not marking each similar users of this targeted customer After the scoring of product is predicted so that user-article rating matrix that each similar users of this targeted customer is corresponding is filled out Fill, and then the article that this targeted customer is not marked all have scoring at the similar users of this targeted customer, improve article Score in predicting device is the accuracy that targeted customer recommends article.

105, article score in predicting device commenting according to the similar users article to be marked to targeted customer of targeted customer Point, for article to be marked prediction scoring.

Exemplary, article score in predicting device is according to the similar users article to be marked to targeted customer of targeted customer Scoring, targeted customer is to the meansigma methods of the scoring of all items, each similar users meansigma methods to the scoring of all items And the similarity between targeted customer and described each similar users, calculate targeted customer and treat the prediction scoring of scoring article. Concrete calculating can be predicted based on following score in predicting formula.

Wherein, above-mentioned score in predicting formula is:

$r_{u^{\′},v^{\′}}=\stackrel{-}{r_{u^{\′}}}+\frac{{\mathrm{\Σ}}_{u_j\∈E_{u^{\′}}}sim(u^{\′},u_j)\·(r_{u_j,v^{\′}}-\stackrel{-}{r_{u_j}})}{{\mathrm{\Σ}}_{u_j\∈E_{u^{\′}}}\left|sim(u^{\′},u_j)\right|}$ (formula three)

Wherein, above-mentioned user u' score in predicting value r to video v'_u',v'Meansigma methods for user u' scoring adds arest neighbors The scoring impact of user.For the meansigma methods of all scorings of user u', E_u'Neighborhood after filling for user u', sim (u',u_j) it is the user u' and user u calculated according to formula one_j(u_j∈E_u') similarity,For user u_jTo video v''s Scoring,For user u_jThe meansigma methods of all scorings.

Through above-mentioned steps, this article score in predicting device is by carrying out the scoring of the article that this targeted customer does not marks After prediction so that user-article rating matrix corresponding to this targeted customer is filled, due to dope this targeted customer couple Not marking the scoring of article, therefore, this article score in predicting device can directly select article of not marking from this targeted customer Select the highest several article of scoring and recommend this targeted customer, it is ensured that article score in predicting device is that targeted customer recommends The accuracy of article, improves recommendation efficiency.

The article score in predicting method and device that embodiments of the invention provide, by believing the scoring of article according to user Breath generates user-type of items rating matrix, then determines the phase of targeted customer according to this user-type of items rating matrix Like user, the score information to article according to this user-type of items rating matrix and user, calculate the similar of targeted customer User treats the prediction scoring that scoring article exist the similar users article to be marked to targeted customer of disappearance scoring, After, according to the scoring of the similar users of this targeted customer article to be marked to targeted customer, for article to be marked prediction scoring. Carry out pre-compared to prior art only according to the disappearance scoring that the user-article rating matrix of shortage of data is targeted customer The method surveyed, this programme is by being aggregated to scattered article in different goods categories in advance, then in conjunction with existing data The user of disappearance-article rating matrix builds the greater concentration of user of data-goods categories rating matrix, thus improves and determine The accuracy of the similar users of targeted customer, and then can be the similar users of targeted customer to be treated scoring article there is disappearance The similar users article to be marked to targeted customer of scoring dope the higher scoring of accuracy, so use according to these targets The scoring of the similar users at family article to be marked to targeted customer, just can be accurately for article the to be marked prediction of targeted customer Go out scoring.

Optionally, generally, each user different times can because of by around many and diverse influences and Produce the change of interest, capture, accordingly, it is capable to no, the interest that user is recent exactly, and by interested for current for user most probable Video recommendations is to a bit that user is non-the normally off key.And the present invention is by following the tracks of the interest in user's nearest a period of time, according to The distance of scoring time gap current time when user marks for article, so that it is determined that go out user to different goods categories The preference value shown.

Concrete, as in figure 2 it is shown, the following institute of determination process that user is to the preference value that different goods categories are shown Show:

201, article score in predicting device obtains the target item collection of user and the target item that target item collection is corresponding Set of types.

The target item that target item in the present embodiment is concentrated is that user is more than scoring time when having marked and marked The article of predetermined threshold, contain the property category belonging to target item that target item is concentrated in this target item set of types Not.Exemplary, if these article are as a example by video, targeted customer can be commented undue and scoring point by article score in predicting device The value video not less than 7 carries out inverted order arrangement according to the scoring time, then picks out front 30 videos and forms this user currently Video set (i.e. target item collection) interestedMeanwhile, to video setIn the generic duplicate removal of each video, enter one Step obtains the goods categories set (i.e. target item classification collection) that user is most interested in recentlyIt should be noted that here User does not limit, and can be targeted customer, it is also possible to be the similar users of targeted customer.

202, article score in predicting device is according to the target item collection of user and target item set of types, calculates user Preference value to every kind of goods categories.

Exemplary, belong to set for every kindClassification k, this method be its calculate a corresponding weightI.e. at user u_iThe video collection likedIn belong to the ratio shared by video of classification k.This weighted value is used for representing use Family u_iFor the preference of classification k in a period of time recently.It is worth the biggest, then it represents that user u_iWithin nearest a period of time, More like watching the video belonging to classification k.

Optionally, in order to improve the accuracy of the article that article score in predicting device is recommended by targeted customer, this enforcement The preference value of every kind of goods categories is come to use for target by the targeted customer that example is also based on step 201 and step 202 calculates Article the to be marked prediction scoring at family, i.e. based on step 201 and step 202, step 105 specifically includes following content:

105a1, article score in predicting device are according to the similar users article to be marked to targeted customer of targeted customer Scoring, targeted customer to the meansigma methods of the scoring of all items, each similar users to the meansigma methods of the scoring of all items, mesh Similarity between mark user and described each similar users and the similar users of the targeted customer preference to every kind of type of items Value, calculates targeted customer and marks the prediction of described article to be marked.

After getting the similar users of the targeted customer preference value to every kind of goods categories, belong to object for each Product collectionVideo v_i, this method is that it calculates a corresponding weightIt is used for representing user u_iA period of time recently Interior for video v_iPreference value, first, willIt is initialized as 1, then, belongs to set for every kindClassification k, if work as Front video v_iBelong to classification k, then its classification is liked coefficientAdd up, i.e.

Exemplary, article score in predicting device is at the thing to be marked to targeted customer of the similar users according to targeted customer The scoring of product, targeted customer are average to the scoring of all items to the meansigma methods of the scoring of all items, each similar users Similarity between value, targeted customer and described each similar users and the similar users of targeted customer are to every kind of type of items Preference value, when the described prediction wait article of marking is marked by calculating targeted customer, can enter based on following score in predicting formula Row prediction.

Wherein, above-mentioned score in predicting formula is:

$r_{u^{\′},v^{\′}}=\stackrel{-}{r_{u^{\′}}}+\frac{{\mathrm{\Σ}}_{u_j\∈E_{u^{\′}}}sim(u^{\′},u_j)\·(r_{u_j,v^{\′}}-\stackrel{-}{r_{u_j}})\·o_{u_j,v^{\′}}}{{\mathrm{\Σ}}_{u_j\∈E_{u^{\′}}}\left|sim(u^{\′},u_j)\right|\·o_{u_j,v^{\′}}}$ (formula four)

Wherein, above-mentioned user u' score in predicting value r to video v'_u',v'Meansigma methods for user u' scoring adds arest neighbors The scoring impact of user.For the meansigma methods of all scorings of user u', E_u'Neighborhood after filling for user u', sim (u',u_j) it is the user u' and user u calculated according to formula one_j(u_j∈E_u') similarity,For user u_jTo video v''s Scoring,For user u_jThe meansigma methods of all scorings,For the user u calculated in step B1_jPreference value to video v', i.e. $o_{u_j,v^{\′}}={\Σ}_{k\∈L_{u_j}^{\′},v^{\′}\∈k}{g}_{u_j,k}.$

Additionally, article score in predicting device can also by by targeted customer to the preference value of each article to be recommended according to It is ranked up from high to low order, and the article several to be recommended choosing preference value the highest recommend targeted customer.

The dynamic interest that this sample plan is recent by catching user, adds the access customer preference value to goods categories, comes Weigh user's preference to article.The prediction scoring of the article in combination with, neighboring user not marked, and target The prediction scoring of the article that user does not marks, thus ensure that the accuracy of the article recommended into targeted customer.

Embodiments of the invention provide a kind of article score in predicting device, and this device is for realizing above-mentioned article recommendation side Formula, as it is shown on figure 3, this device includes: the first acquisition module 21, generation module 22, determine module 23, computing module 24 and pre- Survey module 25, wherein:

First acquisition module 21, for obtaining user's score information to article.

Generation module 22, for the user that obtains according to the first acquisition module 21 score information of article generated user- Type of items rating matrix.

Determine module 23, determine that target is used for the user-type of items rating matrix generated according to generation module 22 The similar users at family.

Computing module 24, obtains for the user-type of items rating matrix generated according to generation module 22 and first The user that module 21 the obtains score information to article, calculates and treats scoring article existence disappearance in the similar users of targeted customer The prediction scoring of the similar users article to be marked to targeted customer of scoring.

Prediction module 25, is used for the scoring of the similar users article to be marked to targeted customer according to targeted customer, for Article to be marked prediction scoring.

Optionally, computing module 24 specifically for:

From the article that first similar users of targeted customer has been marked obtain targeted customer article to be marked first Similar article collection, first similar article concentrate similar article be the first similar users marked and with described article phase to be marked As article；First similar users be targeted customer similar users in treat scoring article exist disappearance scoring similar use Family；

According to the similarity between each similar article that article to be marked are concentrated to described similar article, the first similar users The meansigma methods of the scoring obtained by the scoring of each similar article that similar article are concentrated, the first article to be marked and similar The meansigma methods of the scoring obtained by each similar article that article are concentrated, it was predicted that go out the first similar users to the first article to be marked Scoring.

Optionally, it was predicted that module 25 specifically for:

The scoring of the article to be marked to targeted customer of the similar users according to targeted customer, targeted customer are to all items The meansigma methods of scoring, each similar users each similar to described to meansigma methods and the targeted customer of the scoring of all items Similarity between user, calculates targeted customer and treats the prediction scoring of scoring article.

Optionally, score information also includes scoring time when article are marked by user, as it is shown on figure 3, this device Also include: the second acquisition module 26, wherein:

Second acquisition module 26, for obtaining the target item collection of user and the object category that target item collection is corresponding Type collection, the target item that target item is concentrated is user's scoring time when having marked and marked article more than predetermined threshold, Target item set of types contains all items classification belonging to target item that target item is concentrated.

Computing module 24, is additionally operable to the target item collection of user and the target item obtained according to the second acquisition module 26 Set of types, calculates user's preference value to every kind of goods categories.

Further alternative, it was predicted that module 25 specifically for:

The scoring of the article to be marked to targeted customer of the similar users according to targeted customer, targeted customer are to all items The meansigma methods of scoring, each similar users meansigma methods, targeted customer and described each similar use to the scoring of all items Similarity between family and the similar users of the targeted customer preference value to every kind of type of items, calculate targeted customer and treat described The prediction scoring of scoring article.

Optionally, determine module 23 specifically for:

According to calculating formula of similarity and described user-type of items rating matrix, calculate targeted customer and other Similarity between user；

Wherein, calculating formula of similarity is: $sim(u^{\′},u_i)=\frac{min\left(\right|G_{u^{\′},u_i}|,\mathrm{\θ})}{\mathrm{\θ}}\·\frac{\underset{k\∈G_{u^{\′},u_i}}{\Σ}(C_{u^{\′},k}-\stackrel{-}{r_{u^{\′}}})(C_{u_i,k}-\stackrel{-}{r_{u_i}})}{\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u^{\′},k}-\stackrel{-}{r_{u^{\′}}})}^2}\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u_i,k}-\stackrel{-}{r_{u_i}})}^2}},$ sim(u',u_i) represent targeted customer u' and other user u_iSimilarity,For targeted customer u' and other user u_iAll beat Goods categories set belonging to undue article,For representing average to the scoring of all items classification of targeted customer u' Value,For representing other users meansigma methods to the scoring of all items classification, C is user-type of items rating matrix, θ For threshold value.

The article score in predicting device that embodiments of the invention provide, by generating the score information of article according to user User-type of items rating matrix, then determines the similar use of targeted customer according to this user-type of items rating matrix Family, the score information to article according to this user-type of items rating matrix and user, calculate the similar users of targeted customer In treat scoring article exist disappearance scoring similar users article mark to targeted customer prediction mark, finally, root According to the scoring of the similar users of this targeted customer article to be marked to targeted customer, for article to be marked prediction scoring.Compare It is predicted only according to the disappearance scoring that the user-article rating matrix of shortage of data is targeted customer in prior art Method, this programme is by being aggregated to scattered article in different goods categories in advance, then in conjunction with existing shortage of data User-article rating matrix build the greater concentration of user of data-goods categories rating matrix, thus improve and determine target The accuracy of the similar users of user, and then can be the similar users of targeted customer to be treated scoring article there is disappearance scoring Similar users article to be marked to targeted customer dope the higher scoring of accuracy, so according to these targeted customers' The scoring of similar users article to be marked to targeted customer, just can dope for the article to be marked of targeted customer accurately and comment Point.

Additionally, the dynamic interest that this programme is recent also by catching user, add the access customer preference value to goods categories, Weigh user's preference to article.The prediction scoring of the article in combination with, neighboring user not marked, and mesh The prediction scoring of the article that mark user does not marks, thus ensure that the accuracy of the article recommended into targeted customer.

In several embodiments provided herein, it should be understood that disclosed terminal and method, can be passed through it Its mode realizes.Such as, device embodiment described above is only schematically, such as, and the division of described unit, only Being only a kind of logic function to divide, actual can have other dividing mode, the most multiple unit or assembly to tie when realizing Close or be desirably integrated into another system, or some features can be ignored, or not performing.Another point, shown or discussed Coupling each other or direct-coupling or communication connection can be the INDIRECT COUPLING by some interfaces, device or unit or logical Letter connects, and can be electrical, machinery or other form.

The described unit illustrated as separating component can be or may not be physically separate, shows as unit The parts shown can be or may not be physical location, i.e. may be located at a place, or can also be distributed to multiple On NE.Some or all of unit therein can be selected according to the actual needs to realize the mesh of the present embodiment scheme 's.

It addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it is also possible to It is that the independent physics of unit includes, it is also possible to two or more unit are integrated in a unit.Above-mentioned integrated list Unit both can realize to use the form of hardware, it would however also be possible to employ hardware adds the form of SFU software functional unit and realizes.

The above-mentioned integrated unit realized with the form of SFU software functional unit, can be stored in an embodied on computer readable and deposit In storage media.Above-mentioned SFU software functional unit is stored in a storage medium, including some instructions with so that a computer Equipment (can be personal computer, server, or the network equipment etc.) performs the portion of method described in each embodiment of the present invention Step by step.And aforesaid storage medium includes: (Read-Only Memory is called for short for USB flash disk, portable hard drive, read only memory ROM), random access memory (Random Access Memory is called for short RAM), magnetic disc or CD etc. are various can store The medium of program code.

Last it is noted that above example is only in order to illustrate technical scheme, it is not intended to limit；Although With reference to previous embodiment, the present invention is described in detail, it will be understood by those within the art that: it still may be used So that the technical scheme described in foregoing embodiments to be modified, or wherein portion of techniques feature is carried out equivalent； And these amendment or replace, do not make appropriate technical solution essence depart from various embodiments of the present invention technical scheme spirit and Scope.

Claims (12)

1. an article score in predicting method, it is characterised in that including:

Obtain user's score information to article；

According to described user, the score information of article is generated user-type of items rating matrix；

The similar users of described targeted customer is determined according to described user-type of items rating matrix；

The score information to article according to described user-type of items rating matrix and described user, calculates described target and uses Described article to be marked are existed the similar users thing to be marked to described targeted customer of disappearance scoring by the similar users at family The prediction scoring of product；

The scoring of the article to be marked to described targeted customer of the similar users according to described targeted customer, for described thing to be marked Product prediction scoring.

Method the most according to claim 1, it is characterised in that described according to described user-type of items rating matrix with And the score information that described user is to article, calculate described article to be marked are existed by the similar users of described targeted customer and lack The prediction scoring of the similar users article to be marked to described targeted customer losing scoring specifically includes:

The article to be marked of described targeted customer are obtained from the article that first similar users of described targeted customer has been marked First similar article collection, the similar article that described first similar article are concentrated are that described first similar users has been marked and with described The article that article to be marked are similar；Described first similar users be described targeted customer similar users in described thing to be marked There is the similar users of disappearance scoring in product；

According to the similarity between each similar article that described article to be marked are concentrated to described similar article, described first similar User's putting down the scoring obtained by the scoring of each similar article that described similar article are concentrated, described first article to be marked The meansigma methods of the scoring obtained by each similar article that average and described similar article are concentrated, it was predicted that go out described first similar User's scoring to described first article to be marked.

Method the most according to claim 1 and 2, it is characterised in that the described similar users pair according to described targeted customer The scoring of the article to be marked of described targeted customer, specifically includes for the prediction scoring of described article to be marked:

The scoring of the article to be marked to described targeted customer of the similar users according to described targeted customer, described targeted customer couple Meansigma methods and the described target of the scoring of all items are used by the meansigma methods of the scoring of all items, described each similar users Similarity between family and described each similar users, calculates described targeted customer and marks the prediction of described article to be marked.

Method the most according to claim 1, it is characterised in that institute's scoring information also includes that article are carried out by described user Scoring time during scoring, the described article to be marked to described targeted customer of each similar users according to described targeted customer Prediction scoring, for described article mark prediction mark before, described method also includes:

Obtain target item collection and the target item set of types that described target item collection is corresponding, the described object of described user The target item that product are concentrated is described user scoring time when having marked and the marked article more than predetermined threshold, described target Type of items is concentrated and is contained all items classification belonging to target item that described target item is concentrated；

Target item collection according to described user and target item set of types, calculate described user to every kind of goods categories Preference value.

Method the most according to claim 4, it is characterised in that the described similar users according to described targeted customer is to described The scoring of the article to be marked of targeted customer, specifically includes for the prediction scoring of described article to be marked:

The scoring of the article to be marked to described targeted customer of the similar users according to described targeted customer, described targeted customer couple The meansigma methods of the scoring of all items, described each similar users are to the meansigma methods of the scoring of all items, described targeted customer And the similarity between described each similar users and the similar users of the described targeted customer preference value to every kind of type of items, Calculate described targeted customer the prediction of described article to be marked is marked.

Method the most according to claim 1, it is characterised in that described true according to described user-type of items rating matrix The similar users making described targeted customer specifically includes:

According to calculating formula of similarity and described user-type of items rating matrix, calculate described targeted customer and other Similarity between user；

It is ranked up for the similarity between described targeted customer and other users, determines the similar users of described targeted customer；

Wherein, described calculating formula of similarity is: $sim(u^{\′},u_i)=\frac{min\left(\right|G_{u^{\′},u_i}|,\mathrm{\θ})}{\mathrm{\θ}}\·\frac{\underset{k\∈G_{u^{\′},u_i}}{\Σ}(C_{u^{\′},k}-\stackrel{\‾}{r_{u^{\′}}})(C_{u_i,k}-\stackrel{\‾}{r_{u_{{}_i}}})}{\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u^{\′},k}-\stackrel{\‾}{r_{u^{\′}}})}^2}\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u_i,k}-\stackrel{\‾}{r_{u_i}})}^2}},$ sim(u',u_i) represent described targeted customer u' and other user u described_iSimilarity,For described targeted customer u' and institute State other user u_iAll play the goods categories set belonging to undue article,For representing that described targeted customer u' is to property The meansigma methods of the other scoring of category,For representing described other users meansigma methods to the scoring of all items classification, C is institute Stating user-type of items rating matrix, θ is threshold value.

7. an article score in predicting device, it is characterised in that including:

First acquisition module, for obtaining user's score information to article；

Generation module, for the described user that obtains according to described first acquisition module the score information of article generated user- Type of items rating matrix；

Determine module, determine described mesh for the described user-type of items rating matrix generated according to described generation module The similar users of mark user；

Computing module, for the described user-type of items rating matrix and described first generated according to described generation module The described user that acquisition module the obtains score information to article, calculates in the similar users of described targeted customer described to be evaluated Divide the prediction scoring that article exist the similar users article to be marked to described targeted customer that disappearance is marked；

Prediction module, for the scoring of the similar users article to be marked to described targeted customer according to described targeted customer, For the prediction scoring of described article to be marked.

Device the most according to claim 7, it is characterised in that described computing module specifically for:

The article to be marked of described targeted customer are obtained from the article that first similar users of described targeted customer has been marked First similar article collection, the similar article that described first similar article are concentrated are that described first similar users has been marked and with described The article that article to be marked are similar；Described first similar users be described targeted customer similar users in described thing to be marked There is the similar users of disappearance scoring in product；

According to the similarity between each similar article that described article to be marked are concentrated to described similar article, described first similar User's putting down the scoring obtained by the scoring of each similar article that described similar article are concentrated, described first article to be marked The meansigma methods of the scoring obtained by each similar article that average and described similar article are concentrated, it was predicted that go out described first similar User's scoring to described first article to be marked.

9. according to the device described in claim 7 or 8, it is characterised in that described prediction module specifically for:

The scoring of the article to be marked to described targeted customer of the similar users according to described targeted customer, described targeted customer couple Meansigma methods and the described target of the scoring of all items are used by the meansigma methods of the scoring of all items, described each similar users Similarity between family and described each similar users, calculates described targeted customer and marks the prediction of described article to be marked.

Device the most according to claim 7, it is characterised in that institute's scoring information also includes that article are entered by described user Scoring time during row scoring, described device also includes:

Second acquisition module, for obtaining the target item collection of described user and the target item that described target item collection is corresponding Set of types, the target item that described target item is concentrated is that described user scoring time when having marked and marked is more than predetermined threshold The article of value, contain the property category belonging to target item that described target item is concentrated in described target item set of types Not；

Described computing module, is additionally operable to the target item collection of described user and the target obtained according to described second acquisition module Type of items collection, calculates the described user preference value to every kind of goods categories.

11. devices according to claim 10, it is characterised in that described prediction module specifically for:

The scoring of the article to be marked to described targeted customer of the similar users according to described targeted customer, described targeted customer couple The meansigma methods of the scoring of all items, described each similar users are to the meansigma methods of the scoring of all items, described targeted customer And the similarity between described each similar users and the similar users of the described targeted customer preference value to every kind of type of items, Calculate described targeted customer the prediction of described article to be marked is marked.

12. devices according to claim 7, it is characterised in that described determine module specifically for:

According to calculating formula of similarity and described user-type of items rating matrix, calculate targeted customer and other users Between similarity；

Wherein, described calculating formula of similarity is: $sim(u^{\′},u_i)=\frac{min\left(\right|G_{u^{\′},u_i}|,\mathrm{\θ})}{\mathrm{\θ}}\·\frac{\underset{k\∈G_{u^{\′},u_i}}{\Σ}(C_{u^{\′},k}-\stackrel{\‾}{r_{u^{\′}}})(C_{u_i,k}-\stackrel{\‾}{r_{u_{{}_i}}})}{\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u^{\′},k}-\stackrel{\‾}{r_{u^{\′}}})}^2}\sqrt{\underset{k\∈G_{u^{\′},u_i}}{\Σ}{(C_{u_i,k}-\stackrel{\‾}{r_{u_i}})}^2}},$ sim(u',u_i) represent described targeted customer u' and other user u described_iSimilarity,For described targeted customer u' and institute State other user u_iAll play the goods categories set belonging to undue article,For representing that described targeted customer u' is to property The meansigma methods of the other scoring of category,For representing described other users meansigma methods to the scoring of all items classification, C is institute Stating user-type of items rating matrix, θ is threshold value.