CN104615631A - Information recommendation method and device - Google Patents

Abstract

The invention provides an information recommendation method and device. The method includes the steps that user evaluation information of each target result in search result sets acquired through search keywords is acquired, it is determined that any two target results is in a forward recommendation relationship or a reverse recommendation relationship according to the user evaluation information, and therefore forward recommendation values or reverse recommendation values of the two target results are calculated; according to the forward recommendation values and reverse recommendation values, attribute values corresponding to all target results are generated; according to the attribute values corresponding to all target results, the search result sets are screened and ranked, so that a recommendation result set containing the screened target results is acquired and sent to user equipment so as to be displayed by the user equipment. By the adoption of the method and device, recommendation result accuracy can be improved and recommendation results further accord with the search intentions of a user.

Description

A kind of method of information recommendation and device

Technical field

The present invention relates to technical field of Internet information, be specifically related to a kind of method and device of information recommendation.

Background technology

Along with the development of Internet information technique, user can touch bulk information quickly and easily, but user is difficult to search from magnanimity information obtains target information, wastes the plenty of time simultaneously and browses irrelevant information.At present, conventional recommend method is content recommendation method, and the record of browsing according to user recommends user not contact the recommendation items of fruit to user, and the target information accuracy of the recommendation results obtained and user is not high, if no user browses record by recommending a large amount of information, cause information overflow.

Summary of the invention

The embodiment of the present invention provides a kind of method and device of information recommendation, can improve the accuracy of recommendation results.

Embodiment of the present invention first aspect provides a kind of method of information recommendation, comprising:

Obtain user's evaluation information that the Search Results obtained by search key concentrates every objective result, determine that any two objective results are that forward is recommended relation or oppositely recommends relation according to described user's evaluation information, to calculate the forward recommendation of any two objective results or reverse recommendation;

Property value corresponding to every objective result is generated according to described forward recommendation and described reverse recommendation;

According to property value corresponding to described every objective result described search result set screened and sort, to obtain the recommendation results collection of the objective result after comprising screening, and described recommendation results collection is sent to subscriber equipment, show described recommendation results collection to make described subscriber equipment.

Embodiment of the present invention second aspect provides a kind of device of information recommendation, comprising:

Acquiring unit, concentrates user's evaluation information of every objective result for obtaining the Search Results obtained by search key;

First computing unit, for determining that according to described user's evaluation information any two objective results are that forward recommends relation or reverse recommendation relation to calculate the forward recommendation of any two objective results or reverse recommendation;

Generation unit, for generating property value corresponding to every objective result according to described forward recommendation and described reverse recommendation;

Screening unit, for screen described search result set according to property value corresponding to described every objective result and sort, to obtain the recommendation results collection of the objective result after comprising screening;

Recommendation unit, for described recommendation results collection is sent to subscriber equipment, shows described recommendation results collection to make described subscriber equipment.

The embodiment of the present invention is by obtaining user to the evaluation information of search result set, determine that any two objective results are that forward is recommended relation or oppositely recommends relation, calculating forward recommendation or oppositely recommendation are to generate property value, thus search result set is screened, the recommendation results that obtains meeting targeted customer's needs of sorting described recommendation results is sent to target UE, improve the accuracy of described recommendation results, search recommendation results is fitted the search intention of user more.

Accompanying drawing explanation

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

The schematic flow sheet of a kind of information recommendation method that Fig. 1 provides for the embodiment of the present invention;

A kind of property value matrix schematic diagram that Fig. 1 a provides for the embodiment of the present invention;

The schematic flow sheet of the method for wherein a kind of generating recommendations result set that Fig. 2 provides for this aspect embodiment;

The schematic flow sheet of the another kind of information recommendation method that Fig. 3 provides for the embodiment of the present invention;

The schematic flow sheet of the another kind of information recommendation method that Fig. 4 provides for the embodiment of the present invention;

The recommendation graph of a relation that Fig. 5 recommends for a kind of finance product that the embodiment of the present invention provides;

The structural representation of a kind of information recommending apparatus that Fig. 6 provides for the embodiment of the present invention;

Fig. 7 is the structural representation of the screening unit of embodiment of the present invention information recommending apparatus;

The structural representation of an embodiment of the ground floor recommendation unit that Fig. 8 provides for the embodiment of the present invention;

The structural representation of an embodiment of the second layer recommendation unit that Fig. 9 provides for the embodiment of the present invention;

The structural representation of another embodiment of the second layer recommendation unit that Figure 10 provides for the embodiment of the present invention.

Embodiment

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

Below in conjunction with accompanying drawing 1-accompanying drawing 4, the information recommendation method that the embodiment of the present invention provides is described in detail.

Referring to Fig. 1, is the process flow diagram of a kind of information recommendation method that the embodiment of the present invention provides; The method can comprise the following steps S101-step S103.

S101, obtain user's evaluation information that the Search Results obtained by search key concentrates every objective result, determine that any two objective results are that forward is recommended relation or oppositely recommends relation according to described user's evaluation information, to calculate the forward recommendation of any two objective results or reverse recommendation.

Concrete, obtain user's evaluation information that the Search Results obtained by search key concentrates every objective result, detect the content relation that described Search Results concentrates user's evaluation information of any two objective results.Described user's evaluation information is the evaluation of user to certain characteristic of objective result, such as to the evaluation of commodity performance, the evaluation of finance product income ratio.Evaluation information according to described user determines content relation, is that same or analogous two objective results are set as that forward recommends relation by described content relation, and will recommend two objective result composition forward result groups of relation for described forward; Be that two contrary objective results are set as oppositely recommending relation by described content relation, and the two objective results for described reverse recommendation relation are formed reverse result group.Suppose that the search result set obtained by search key is S={S1, S2 ..., Sn}, the user's set participating in comment is U={U1, U2 ..Um}, and the comment collection of all users is combined into C={C1, C2 ... Ck}.If user is same or similar to the evaluation information of Si and the Sj of two results in S, then the content relation of result Si and Sj is that forward recommends relation, and result Si and Sj forms forward result group; If user is contrary to the evaluation information of Si and the Sj of two results in S, then the content relation of result Si and Sj is for oppositely to recommend relation, and result Si and Sj forms reverse result group.

Relation is recommended, according to forward recommendation computing formula according to the forward that described user's evaluation information is determined calculate forward recommendation, wherein R _forward(S _i, S _j) be the forward recommendation of result Si and Sj; Ui is the user's weight participating in result Si and Sj comment; To the number of result Si and Sj forward result group in the review information that num (C1|Ui) is described user, the number of each forward result group in comment set C1 is recommended to obtain by the forward of traverse user Ui; To the number of the reverse result group of result Si and Sj in the review information that num (C2|Ui) is described user, the number of being closed each reverse result group in C2 by the reverse recommendation comment collection of traverse user Ui is obtained; Num (C1|U) and num (C|U) is respectively to result Si and the number of Sj forward result group and the number of result group in all user comment information, and the number of described result group comprises the number of forward result group and the number of reverse result group.

In like manner according to the reverse recommendation relation that described user's evaluation information is determined, according to reverse recommendation computing formula calculate reverse recommendation, wherein R _oppositely(S _i, S _j) be the reverse recommendation of result Si and Sj; Ui is the user's weight participating in result Si and Sj comment; To the number of the reverse result group of result Si and Sj in the review information that num (C2|Ui) is described user, the number of being closed each reverse result group in C2 by the reverse recommendation comment collection of traverse user Ui is obtained; To the number of result Si and Sj forward result group in the review information that num (C1|Ui) is described user, the number of each forward result group in comment set C1 is recommended to obtain by the forward of traverse user Ui; Num (C2|U) and num (C|U) is respectively to the number of the reverse result group of result Si and Sj and the number of result group in all user comment information, and the number of described result group comprises the number of forward result group and the number of reverse result group.

S102, generates property value corresponding to every objective result according to described forward recommendation and described reverse recommendation.

Concrete, the reverse recommendation corresponding according to forward recommendation corresponding to the number of the number of forward result group, oppositely result group, each forward result group, each reverse result group and user's weight calculation formula calculate user's weighted value.R in described user's weight calculation formula _forward(U), R _oppositely(U) be respectively forward recommend number of users and oppositely recommend number of users; R _forward(S _i, S _j), R _oppositely(S _i, S _j) be respectively forward recommendation in step S101 and reverse recommendation.Recommend computing formula, described reverse recommendation computing formula and described user's weight calculation formula construction to become a system of equations described forward, obtain corresponding forward recommendation, oppositely recommendation and user's weighted value by solving equations.

Generate property value corresponding to described objective result according to the number of the number of the forward result group corresponding with described objective result, oppositely result group, forward recommendation, oppositely recommendation, user's weighted value and extended attribute value, described extended attribute at least comprises volumes of searches, click volume, portfolio.

Property value corresponding for every objective result is formed property value matrix, the property value that described in the behavior of described property value matrix, every objective result is corresponding, the property value that first behavior first entry mark result is corresponding, the property value that the second behavior Article 2 objective result is corresponding, by that analogy; Described property value matrix column is the set of kind attributes value, first be classified as forward junction fruit group number, second be classified as reverse junction fruit group number, by that analogy.Described property value matrix refers to Fig. 1 a, and wherein T1 represents Article 1 objective result, and Ri represents certain generic attribute value.

S103, according to property value corresponding to described every objective result described search result set screened and sort, to obtain the recommendation results collection of the objective result after comprising screening, and described recommendation results collection is sent to subscriber equipment, shows described recommendation results collection to make described subscriber equipment.

According to property value corresponding to described every objective result described search result set screened and sort, to obtain the recommendation results collection of the objective result after comprising screening, described recommendation results collection comprises the objective result with Keywords matching, and such as search key " financing " obtains objective result " finance product A "; Also comprise the objective result that keyword attribute is recommended, such as search key " financing " obtains " finance product A ", and " financing is logical " has the attribute of " finance product ", is obtained finance products such as " finance product B " by " finance product attribute ".And these objective results are according to the sequence of correlativity height.

Described recommendation results collection is sent to subscriber equipment, and to make described subscriber equipment show described recommendation results collection, described subscriber equipment can be the query facility in PC, mobile terminal, various field.

The embodiment of the present invention concentrates the user comment information of every objective result by obtaining Search Results, determine forward, oppositely recommend relation, calculate forward, oppositely recommendation and user's weighted value and obtain property value, according to property value, the recommendation results collection that screening and sequencing obtains user's needs is carried out to described search result set, improve the accuracy of described recommendation results collection.

Refer to Fig. 2, the schematic flow sheet of the method for the wherein a kind of generating recommendations result set provided for this aspect embodiment; The method comprises the step S103 that can correspond in the corresponding embodiment of above-mentioned Fig. 1, and the method can comprise the following steps S201-S202.

S201, carries out temperature screening and temperature sequence, to obtain ground floor recommendation results collection according to the property value matrix that the property value corresponding by every objective result is formed to described search result set.

Concrete, according to the property value matrix that the property value corresponding by every objective result is formed, temperature screening and temperature sequence are carried out, to obtain ground floor recommendation results collection to described search result set.The described Search Results obtained by keyword search is concentrated containing many results, comprising correlated results and uncorrelated result, incoherent result can be deleted and sorted by correlated results, obtain ground floor recommendation results collection by temperature screening.

S202, recommends relation, described reverse recommendation relation and described ground floor recommendation results collection to generate second layer recommendation results collection according to described forward.

Relation, described reverse recommendation relation is recommended to carry out recommending again based on user comment information understanding to each objective result that described ground floor recommendation results is concentrated according to described forward, to generate second layer recommendation results collection, described second layer recommendation results collection comprises general recommendations result and personalized recommendation result.General recommendations result is not considered user's weighted value of a certain user but is considered that popular review information is recommended user, can meet the needs of most of user; Personalized recommendation result is the recommendation results a certain user being returned to this user needs, and this recommendation results is based on the evaluation information of this user.

The embodiment of the present invention is by carrying out two-layer recommendation to search result set, and the recommendation results collection obtained can meet the needs of most of user, also can meet the needs of specific user.

Referring to Fig. 3, is the schematic flow sheet of the another kind of information recommendation method that the embodiment of the present invention provides; The method can correspond to the step S201 in the corresponding embodiment of above-mentioned Fig. 2, and the method can comprise the following steps S301-step S305.

S301, regular described property value matrix, with generating recommendations factor matrix.

Concrete, formula W=(num (all)-num (0))/num (all) is utilized to calculate every Column Properties value weighting coefficient of described property value matrix, wherein num (0) for certain property value in all results be the number of 0, for there is the number of this property value in described result in num (all), the weighting coefficient of every Column Properties value constitutes a weighting coefficient matrix, is that the attribute of 0 removes and is optimized weighting coefficient matrix by weighting coefficient.Be multiplied weighting coefficient with the property value of weighting coefficient corresponding row the property value obtaining upgrading.By property value segmentation, segmentation number is n, then the scope of each section is: num (seg)=(max (att)-min (att)/n), the sequence number of section is 1 ~ n, wherein max (att) is the maximal value in the property value upgraded, min (att) is the minimum value in the property value upgraded, and the segment value that num (seg) is each section, using the granularity of division value of segment value as attribute.Find affiliated section according to property value and be mapped to above a certain number in 1 ~ n, the value of such as a certain attribute is 200, the value of section is 100, section number be 10 map after sequence number be 2, computing formula is: [num (att)/num (seg)], and wherein num (att) is the property value upgraded.Upgrade described property value matrix according to mapping result can obtain recommending factor matrix.

S302, according to the set of described recommendation factor matrix generated equivalence relation.

Concrete, linearization is carried out, with generated equivalence relation set to the row and column of described recommendation factor matrix.Recommend the row of factor matrix to represent the associated recommendation factor of result set, i.e. the property value of objective result, factor set is recommended in row representative, i.e. kind attributes value set, relation of equivalence set selects the associated recommendation factor from this matrix.First a submatrix is generated from recommending to choose factor matrix the recommendation factor that the associated recommendation factor and this recommendation factor pair answer, then after recommending the row and column linearization of factor matrix, relation of equivalence set is combined into, there is m the associated recommendation factor at present, suppose that the recommendation factor that associated recommendation factor pair is answered mostly is n most, the relation of equivalence set then generated is 1 ~ m*n element, and now relation of equivalence set is:

R={T1 (R1), T2 (R1) ... .Tm (R1) ... .Tm (factor 1 is recommended in expansion) ...

S303, carries out classification according to relation of equivalence set to described recommendation factor matrix and obtains content recommendation set.

Concrete, the number presetting the first recommendation results is N, concentrates choose 2N bar result as the data source of recommending from described Search Results.Set the granularity of division of recommending the factor in described recommendation factor matrix according to the Update attribute value matrix that described Update attribute value is formed, divide described data source by described granularity of division.Choose properties collection to be recommended according to threshold matrix, detect the number M of described properties collection to be recommended, the initial value of described threshold matrix is zero.When M is less than 2N, described content to be recommended is defined as content recommendation; When M is greater than 2N, selected threshold matrix and properties collection to be recommended are until the number of properties collection to be recommended is less than 2N again, determine that properties collection to be recommended is now content recommendation set.

The process sorted out automatically is run by program and is obtained, and this process is also the basis of attribute loop module, and this process can be different according to the difference of Data Update frequency, directly can utilize threshold matrix within Data Update frequency.

Wherein, if the recommendation number very few illustrated divisions granularity obtained when running this process is excessive, attribute section size can be adjusted like this to reduce granularity of division value, more complicated situation can revise granularity of division value one by one according to the order of property weighing factors, and user can select concrete strategy according to the needs of practical application.

S304, optimizes described content recommendation set and generates optimum recommendation factor matrix, and the result often corresponding to row in described optimum recommendation factor matrix is the objective result after temperature screening.

Concrete, each goal set in the content recommendation set obtain step S303 carries out linearization according to the method for step S302, obtains the target relation of equivalence that each goal set is corresponding.Target relation of equivalence set corresponding to described each goal set is respectively sorted out, to obtain target content recommendation set corresponding to each goal set.Detecting the Set Status of target content recommendation set corresponding to described each goal set, is that the goal set of non-variable condition is added into optimum recommendation factor matrix by described Set Status.The described optimum recommendation factor matrix result that often row is corresponding is the objective result after temperature screening.

S305, calculates the recommendation that described optimum recommends factor matrix often to go, and recommends the result often corresponding to row in factor matrix to carry out temperature sequence from height to low according to the size of described recommendation, to obtain ground floor recommendation results collection described optimum.

The optimum property value recommending the row of factor matrix to represent objective result, row represent the set of kind attributes value.User it is desired that according to similarity height sequence recommendation results, therefore the present invention proposes a formula calculating temperature recommendation to the described optimum calculating recommending objective result corresponding to factor matrix to carry out recommendation, so that the ground floor recommendation results collection obtained sorts by temperature.The computing formula of this temperature recommendation is: $\mathrm{RS}=(\underset{i=1}{\overset{n}{\mathrm{\Π}}}A\left(T_i\right)-(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}A\left(T_i\right)/n))/\sqrt{(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}A{\left(T_i\right)}^2-({\left(\underset{i=1}{\overset{n}{\mathrm{\Σ}}}A\left(T_i\right)\right)}^2/n))}$

Wherein, RS represents recommendation, and A (Ti) represents the integrated value of objective result Ti all properties value superposition, and n represents the number of objective result in content recommendation set, the line number that namely optimum recommendation factor matrix is corresponding.

The embodiment of the present invention obtains ground floor recommendation results collection after carrying out a series of process by the property value matrix that described search result set is corresponding, and the quantity of described recommendation results collection is obviously less than described search result set, and deletes the result irrelevant with keyword.

Referring to Fig. 4, is the schematic flow sheet of the another kind of information recommendation method that the embodiment of the present invention provides; The method can correspond to the step S202 in the corresponding embodiment of Fig. 2, the method comprising the steps of S401-S402.

S401, general result is recommended.

Concrete, recommend relation, described reverse recommendation relation to recommend the general result that each objective result that described ground floor recommendation results is concentrated carries out understanding based on user comment information according to described forward.

Choose the objective result Si that described ground floor recommendation results is concentrated, obtain the forward result group set of Si, the set of forward target recommendation results group, the oppositely set of result group and oppositely target recommendation results group set.Objective result Si is a certain element that ground floor recommendation results is concentrated, the group set of forward result and the set of forward target recommendation results group of Si is obtained by the review information of traversal ground floor recommendation results collection, the set of described forward result group comprises the forward result group at described objective result, the set of the result composition that the forward result group of described forward target recommendation results set at described objective result is corresponding.In like manner can obtain the set of reverse result group and the set of reverse target recommendation results group of Si.

The forward recommendation that the forward result group of calculating at Si is corresponding, and recommend set to sort according to the size of described forward recommendation to described forward target, obtain S set ' forward.The corresponding forward recommendation of each forward result group of Si, the computing formula according to step S101 calculates forward recommendation, and the size according to forward recommendation sorts to the set of forward target recommendation results, obtains S set ' forward.Such as, the forward result group of Si is (Si, Sm), (Si, Sn), (Si, So), so the target recommendation results group of Si is (Sm, Sn, So), forward result group (Si, Sm) is calculated, (Si, Sn), (Si, So) forward recommendation is carried out sequence to the element of three in (Sm, Sn, So) and is obtained S set ' forward.The reverse recommendation that the reverse result group of calculating at Si is corresponding, and recommend set to sort according to the size of described reverse recommendation to described reverse target, obtain S set ' oppositely.Detailed process and obtain S set ' forward is similar, do not repeat them here.

When described S set being detected, ' forward and described S set ' is oppositely containing identical target recommendation results, and the forward recommendation corresponding with identical target recommendation results is when being less than reverse recommendation, from described S set ' delete described identical target recommendation results forward, obtain general recommendations result.

Assumption set S ' forward is (Sm, Sn, So), S set ' be reversed (Sp, Sq, So), two set are containing identical target recommendation results So, then (Si is compared, So) forward recommendation and the size of reverse recommendation, if the forward recommendation of (Si, So) is less than reverse recommendation, so from S set ' delete target recommendation results So forward and obtain general recommendations result (Sm, Sn).

S402, personalization results is recommended.

Concrete, recommend relation, described reverse recommendation relation to recommend the personalization results that each objective result that described ground floor recommendation results is concentrated carries out understanding based on user comment information according to described forward.

Choose targeted customer and corresponding with described targeted customer objective result Si that described ground floor recommendation results concentrates, obtain the forward result group set participated in containing described targeted customer of Si, the set of forward target recommendation results group, the oppositely set of result group and oppositely target recommendation results group set.Concentrate from described ground floor recommendation results and choose an objective result Si, the comment of targeted customer Uk to Si is comprised in the review information of Si, obtain forward result group set described in the forward result group set of Si and the set of forward target recommendation results group by the review information of traversal ground floor recommendation results collection and comprise the forward result group that the described targeted customer at described objective result participates in, the set of the result composition that the forward result group that the described targeted customer of described forward target recommendation results set at described objective result participates in is corresponding.In like manner can obtain the set of reverse result group and the set of reverse target recommendation results group of Si.The forward recommendation coefficient value that the forward result group of calculating at Si is corresponding, and recommend the size of coefficient value to recommend set to sort to described forward target according to described forward, obtain S set ' forward.

Calculating targeted customer Uk recommends the forward of each element in goal set to recommend coefficient value for objective result Si and forward, and computing formula is: (num (C _just(Si, S _j)) ÷ num (C _instead(Si, S _j))) * (num (C ₁) ÷ num (C))+1, wherein num (C _just(Si, S _j)) be the number of targeted customer Uk to the recommendation of Si and Sj forward in described forward target recommendation set, num (C _instead(Si, S _j) recommend the number of in set, Si and Sj oppositely being recommended, num (C in described reverse target for targeted customer Uk ₁) and num (C) be respectively described forward target and recommend the number of element and target in set to recommend the number of element in set.Recommend the size of coefficient value to recommend set to sort to described forward target according to forward, obtain S set ' forward.The reverse recommendation coefficient value that the reverse result group of calculating at Si is corresponding, and recommend set to sort according to the size of described reverse recommendation coefficient value to described reverse target, obtain S set ' oppositely.Detailed process and obtain S set ' forward is similar, do not repeat them here.

When described S set being detected, ' forward and described S set ' is oppositely containing identical target recommendation results, and the forward recommendation corresponding with identical target recommendation results is when being less than reverse recommendation, from described S set ' delete described identical target recommendation results forward, obtain personalized recommendation result.Similar with acquisition general recommendations result, from S set ' delete identical target recommendation results forward, and the forward recommendation of same target result recommendation results is less than reverse recommendation, obtains the personalized recommendation result for targeted customer.

The embodiment of the present invention both can provide general recommendations result, can provide personalization results again.If user searches for the identity of visitor, the information of server not this user, so will provide general recommendations result to meet the demand of this user to this user.If user has logged in account before searching for, so server can carry out information recommendation according to the evaluation information specific aim of this user to this user, and the recommendation results of feedback is personalized recommendation result.

Referring to Fig. 5, is the recommendation graph of a relation of a kind of finance product recommendation that the embodiment of the present invention provides.

Integrate as S={S1, S2, S3 by keyword in the recommendation results that the vertical search in a certain field obtains,

S4, S5, S6, S7, S8, S9, S10}, user's set of participation is U={U1, U2, U3, U4}.

Forward recommends system of equations as follows:

R _just(S1, S2)=U1*1/2

R _just(S1, S9)=U2*2+U3

R _just(S2, S3)=U2*3+U4*2+U3*4

R _just(S2, S4)=U1*1/3

R _just(S2, S6)=U1+U3*2

R _just(S2, S8)=U1*1/2

R _just(S2, S10)=U3+U4*2

R _just(S3, S4)=U2*2+U4

R _just(S3, S9)=U1/3

R _just(S3, S10)=U2*2+U4

Reverse recommendation system of equations is as follows:

R _instead(S1, S2)=U2*2*2/3

R _instead(S2, S4)=U3*2*2/3

R _instead(S2, S8)=U2*1/2

R _instead(S3, S9)=U3*2*2/3

User's weighted value system of equations is as follows:

U1=R _just(S1, S2)+R _just(S2, S4)+R _just(S2, S6) * 1/2+R _just(S3, S9)

+ R _just(S2, S8)-1

U2=R _instead(S1, S2)+R _just(S2, S3) * 1/3+R _just(S3, S4) * 1/2+R _instead(S2, S8)+R _just(S3, S10) * 1/2+R _just(S1, S9) * 1/2-1

U3=R _just(S1, S9) * 1/2+R _instead(S2, S4)+R _just(S2, S10) * 1/2+R _instead(S3, S9)+R _just(S1, S9) * 1/2+R _just(S2, S3) * 1/3-1

U4=R _just(S2, S10) * 1/2+R _just(S3, S4) * 1/2+R _just(S2, S3) * 1/3+R _just(S3, S10) more than * 1/2-1 is the computing formula of forward recommendation, oppositely recommendation, user's weighted value, wherein user's weighted value subtract 1 be in order to solution of equations convergence and on the occasion of, and by user's weighted value by add a positive integer such as 1 be modified on the occasion of, it is as follows that solving equations obtains each value:

Forward recommendation:

R _just(S1, S2)=0.93

R _just(S1, S9)=1.74

R _just(S2, S3)=10.59

R _just(S2, S4)=0.62

R _just(S2, S6)=3.86

R _just(S2, S8)=0.93

R _just(S2, S10)=6.48

R _just(S3, S4)=3.48

R _just(S3, S9)=0.62

R _just(S3, S10)=3.48

Reverse recommendation:

R _instead(S1, S2)=0.49

R _instead(S2, S4)=1.33

R _instead(S2, S8)=0.19

R _instead(S3, S9)=1.33

User's weighted value:

U1＝1.86

U2＝0.37

U3＝1

U4＝2.74

The property value obtained by flow process of the present invention is as follows:

Each objective result has the number R1 of recommending data:

R1(S1)＝3 R1(S2)＝6 R1(S3)＝ 4 R1(S4)＝2 R1(S5)＝0

R1(S6)＝1 R1(S7)＝0 R1(S8)＝1 R1(S9)＝2 R1(S10)＝2

The recommendation summation R2 of each objective result and other target recommendation results:

R1(S1)＝3.16 R1(S2)＝24.93 R1(S3)＝ 19.5

R1(S4)＝5.43 R1(S5)＝0 R1(S6)＝3.86

R1(S7)＝0 R1(S8)＝1.12 R1(S9)＝3.69 R1(S10)＝9.96

The forward recommendation summation R3 of each objective result and other target recommendation results:

R1(S1)＝2.67 R1(S2)＝22.92 R1(S3)＝ 18.17

R1(S4)＝4.1 R1(S5)＝0 R1(S6)＝3.86

R1(S7)＝0 R1(S8)＝0.93 R1(S9)＝3.69 R1(S10)＝9.96

The reverse recommendation summation R4 of each objective result and other target recommendation results:

R1(S1)＝0.49 R1(S2)＝2.01 R1(S3)＝ 1.33

R1(S4)＝1.33 R1(S5)＝0 R1(S6)＝0

R1(S7)＝0 R1(S8)＝0.19 R1(S9)＝2.36 R1(S10)＝0

Extended attribute is searching times, is obtained by search daily record:

{100,5000,400,300,40,200,50，140,230,350}

Recommend according to above data and temperature, the ground floor recommendation results collection that obtains of sorting be as follows:

{, by recommendation results collection S and ground floor recommendation results set pair ratio, can find out, incoherent S5 and S7 is deleted for S2, S3, S10, S4, S6, S9, S1, S8}, and the result obtained is the result by temperature sequence, and first S2 meets the search intention of user most.

By carrying out general recommendations to ground floor recommendation results collection and personalized recommendation obtains following result:

(1) general recommendations result:

R _just(S1)={ S9, S2} R _just(S2)={ S3, S10, S6, S8} R _just(S3)={ S2, S4, S10}

R _just(S4)={ S3} R _just(S6)={ S2} R _just(S8)={ S2}

R _just(S9)={ S1} R _just(S10)={ S3}

(2) personalized recommendation is recommended:

For user U1 recommendation results:

R _just(S1|U1)={ S2} R _just(S2|U1)={ S6, S4} R _just(S3|U1)={ S9}

R _just(S4)={ S2} R _just(S6|U1)={ S2} R _just(S8|U1)={ S2}

R _just(S9|U1)={ S1}

For user U2 recommendation results:

R _just(S1|U2)={ S9} R _just(S3|U2)={ S2, S4} R _just(S4|U2)={ S3}

R _just(S6|U2)={ S2} R _just(S9|U2)={ S1} R _just(S10|U2)={ S3}

For user U3 recommendation results:

R _just(S2|U3)={ S3, S10} R _just(S3|U3)={ S2, S4, S10} R _just(S6|U3)={ S2}

R _just(S9|U3)={ S1} R _just(S10|U3)={ S3}

For user U4 recommendation results:

R _just(S2|U4)={ S3, S10, S6, S8} R _just(S3|U4)={ S2, S4, S10}

R _just(S4|U4)={ S3} R _just(S9|U4)={ S1} R _just(S10|U4)={ S3}

By recommendation results collection S and ground floor recommendation results set pair ratio, can find out, incoherent S5 and S7 is deleted, and the result obtained is the result by temperature sequence, and first S2 meets the search intention of user most.Second layer recommendation results collection is based on ground floor recommendation results collection, and the search intention of being close to the users further, user browses other irrelevant informations without the need to losing time simultaneously.Such as, obtained by search keyword " financing " to the relevant search result set of managing money matters, recommend the ground floor recommendation results collection obtained to be through ground floor: finance product A, finance product B, finance product C, finance product D etc.The general recommendations result set obtained is recommended to be through the second layer: finance product A, finance product B, finance product C.If user is the fan of finance product A, delivered the evaluation logical to financing, and to the recommendation results of this user be: finance product A.

The embodiment of the present invention has set forth flow process of the present invention by specific embodiment, and the recommendation results collection obtained can meet the demand of most of user, can meet again the demand of a certain user targetedly.

Referring to Fig. 6, is the structural representation of a kind of information recommending apparatus that the embodiment of the present invention provides; This device can comprise: the first acquiring unit 101, first computing unit 102, generation unit 103, screening unit 104, recommendation unit 105.

First acquiring unit 101, concentrates user's evaluation information of every objective result for obtaining the Search Results obtained by search keyword.

Described user's evaluation information is the evaluation of user to certain characteristic of objective result, such as to the evaluation of commodity performance, the evaluation of finance product income ratio.Evaluation information according to described user determines content relation, is that same or analogous two objective results are set as that forward recommends relation by described content relation, and will recommend two objective result composition forward result groups of relation for described forward; Be that two contrary objective results are set as oppositely recommending relation by described content relation, and the two objective results for described reverse recommendation relation are formed reverse result group.

First computing unit 102, for determining that according to described user's evaluation information any two objective results are that forward recommends relation or reverse recommendation relation to calculate the forward recommendation of any two objective results or reverse recommendation.

Recommend relation according to the forward that described user's evaluation information is determined, oppositely recommend relation, calculate forward recommendation, oppositely recommendation according to forward recommendation computing formula, oppositely recommendation computing formula.Circular has detailed elaboration in step S101, does not repeat them here.

Generation unit 103, for generating property value corresponding to every objective result according to described forward recommendation and described reverse recommendation.

The reverse recommendation corresponding according to forward recommendation corresponding to the number of the number of forward result group, oppositely result group, each forward result group, each reverse result group and user's weight calculation formula calculate user's weighted value.

Generate property value corresponding to described objective result according to the number of the number of the forward result group corresponding with described objective result, oppositely result group, forward recommendation, oppositely recommendation, user's weighted value and extended attribute value, described extended attribute at least comprises volumes of searches, click volume, portfolio.

Property value corresponding for every objective result is formed property value matrix, the property value that described in the behavior of described property value matrix, every objective result is corresponding, the property value that first behavior first entry mark result is corresponding, the property value that the second behavior Article 2 objective result is corresponding, by that analogy; Described property value matrix column is the set of kind attributes value, first be classified as forward junction fruit group number, second be classified as reverse junction fruit group number, by that analogy.Described property value matrix refers to Fig. 1 a, and wherein T1 represents Article 1 objective result, and Ri represents certain generic attribute value.

Screening unit 104, for screen described search result set according to property value corresponding to described every objective result and sort, to obtain the recommendation results collection of the objective result after comprising screening.Screening unit 104 can comprise ground floor recommendation unit 1401 and second layer recommendation unit 1402, refers to Fig. 7.

Ground floor recommendation unit 1401, carries out temperature screening and temperature sequence, to obtain ground floor recommendation results collection according to the property value matrix that the property value corresponding by every objective result is formed to described search result set.The described Search Results obtained by keyword search is concentrated containing many results, comprising correlated results and uncorrelated result, incoherent result can be deleted and sorted by correlated results, obtain ground floor recommendation results collection by temperature screening.

Second layer recommendation unit 1402, relation, described reverse recommendation relation is recommended to carry out recommending again based on user comment information understanding to each objective result that described ground floor recommendation results is concentrated according to described forward, to generate second layer recommendation results collection, described second layer recommendation results collection comprises general recommendations result and personalized recommendation result.General recommendations result is not considered user's weighted value of a certain user but is considered that popular review information is recommended user, can meet the needs of most of user; Personalized recommendation result is the recommendation results a certain user being returned to this user needs, and this recommendation results is based on the evaluation information of this user.

Recommendation unit 105, for described recommendation results collection is sent to subscriber equipment, shows described recommendation results collection to make described subscriber equipment.

Described subscriber equipment can be PC, mobile terminal, query facility.

The embodiment of the present invention concentrates the user comment information of every objective result by obtaining Search Results, determine forward, oppositely recommend relation, calculate forward, oppositely recommendation and user's weighted value and obtain property value, according to property value, the recommendation results collection that screening and sequencing obtains user's needs is carried out to described search result set, described recommendation results collection can meet the needs of most of user, also can meet the needs of specific user.

Refer to Fig. 8, the structural representation of an embodiment of the ground floor recommendation unit provided for the embodiment of the present invention; Ground floor recommendation unit can comprise: regular unit 1411, relation of equivalence generation unit 1421, classification unit 1431, optimization unit 1441, second computing unit 1451, first sequencing unit 1461.

Regular unit 1411, for regular property value matrix, with generating recommendations factor matrix.

Formula W=(num (all)-num (0))/num (all) is utilized to calculate every Column Properties value weighting coefficient of described property value matrix, wherein num (0) for certain property value in all results be the number of 0, for there is the number of this property value in described result in num (all), the weighting coefficient of every Column Properties value constitutes a weighting coefficient matrix, is that the attribute of 0 removes and is optimized weighting coefficient matrix by weighting coefficient.Be multiplied weighting coefficient with the property value of weighting coefficient corresponding row the property value obtaining upgrading.By property value segmentation, segmentation number is n, then the scope of each section is: num (seg)=(max (att)-min (att)/n), the sequence number of section is 1 ~ n, wherein max (att) is the maximal value in the property value upgraded, min (att) is the minimum value in the property value upgraded, and the segment value that num (seg) is each section, using the granularity of division value of segment value as attribute.Find affiliated section according to property value and be mapped to above a certain number in 1 ~ n, the value of such as a certain attribute is 200, the value of section is 100, section number be 10 map after sequence number be 2, computing formula is: [num (att)/num (seg)], and wherein num (att) is the property value upgraded.Upgrade described property value matrix according to mapping result can obtain recommending factor matrix.

Relation of equivalence generation unit 1421, obtains content recommendation set for carrying out classification according to relation of equivalence set to described recommendation factor matrix.

Linearization is carried out, with generated equivalence relation set to the row and column of described recommendation factor matrix.Recommend the row of factor matrix to represent the associated recommendation factor of result set, i.e. the property value of objective result, factor set is recommended in row representative, i.e. kind attributes value set, relation of equivalence set selects the associated recommendation factor from this matrix.First a submatrix is generated from recommending to choose factor matrix the recommendation factor that the associated recommendation factor and this recommendation factor pair answer, then after recommending the row and column linearization of factor matrix, relation of equivalence set is combined into, there is m the associated recommendation factor at present, suppose that the recommendation factor that associated recommendation factor pair is answered mostly is n most, the relation of equivalence set then generated is 1 ~ m*n element, and now relation of equivalence set is:

R={T1 (R1), T2 (R1) ... .Tm (R1) ... .Tm (factor 1 is recommended in expansion) ...

Sorting out unit 1431, obtaining content recommendation set for carrying out classification according to relation of equivalence set to described recommendation factor matrix.

The number presetting the first recommendation results is N, concentrates choose 2N bar result as the data source of recommending from described Search Results.Set the granularity of division of recommending the factor in described recommendation factor matrix according to the Update attribute value matrix that described Update attribute value is formed, divide described data source by described granularity of division.Choose properties collection to be recommended according to threshold matrix, detect the number M of described properties collection to be recommended, the initial value of described threshold matrix is zero.When M is less than 2N, described content to be recommended is defined as content recommendation; When M is greater than 2N, selected threshold matrix and properties collection to be recommended are until the number of properties collection to be recommended is less than 2N again, determine that properties collection to be recommended is now content recommendation set.

The process sorted out automatically is run by program and is obtained, and this process is also the basis of attribute loop module, and this process can be different according to the difference of Data Update frequency, directly can utilize threshold matrix within Data Update frequency.

Wherein, if the recommendation number very few illustrated divisions granularity obtained when running this process is excessive, attribute section size can be adjusted like this to reduce granularity of division value, more complicated situation can revise granularity of division value one by one according to the order of property weighing factors, and user can select concrete strategy according to the needs of practical application.

Optimization unit 1441, generates optimum recommendation factor matrix for optimizing described content recommendation set, and the result often corresponding to row in described optimum recommendation factor matrix is the objective result after temperature screening.

Linearization is carried out in each goal set in the content recommendation set obtain classification unit, obtains the target relation of equivalence that each goal set is corresponding.Target relation of equivalence set corresponding to described each goal set is respectively sorted out, to obtain target content recommendation set corresponding to each goal set.Detecting the Set Status of target content recommendation set corresponding to described each goal set, is that the goal set of non-variable condition is added into optimum recommendation factor matrix by described Set Status.The described optimum recommendation factor matrix result that often row is corresponding is the objective result after temperature screening.

Second computing unit 1451, for calculating the recommendation that described optimum recommends factor matrix often to go.

Optimum recommendation of recommending factor matrix often to go is calculated according to temperature recommendation computing formula.

First sequencing unit 1461, for recommending the result often corresponding to row in factor matrix to carry out temperature sequence from height to low according to the size of described recommendation, to obtain ground floor recommendation results collection described optimum.

The embodiment of the present invention obtains ground floor recommendation results collection after carrying out a series of process by the property value matrix that described search result set is corresponding, and the quantity of described recommendation results collection is obviously less than described search result set, and deletes the result irrelevant with keyword.

Refer to Fig. 9, the structural representation of an embodiment of the second layer recommendation unit provided for the embodiment of the present invention; Second layer recommendation unit can comprise: second acquisition unit 1412, second sequencing unit 1422, the 3rd sequencing unit 1432, the 4th detecting unit 1442, delete cells 1452.

Second acquisition unit 1412, for according to choosing a concentrated objective result Si of described ground floor recommendation results, obtains the forward result group set of Si, the set of forward target recommendation results group, the oppositely set of result group and oppositely target recommendation results group set.

Objective result Si is a certain element that ground floor recommendation results is concentrated, the group set of forward result and the set of forward target recommendation results group of Si is obtained by the review information of traversal ground floor recommendation results collection, the set of described forward result group comprises the forward result group at described objective result, the set of the result composition that the forward result group of described forward target recommendation results set at described objective result is corresponding.In like manner can obtain the set of reverse result group and the set of reverse target recommendation results group of Si.

Second acquisition unit 1412, also for obtaining forward recommendation corresponding to forward result group at Si and reverse recommendation that oppositely result group is corresponding.

Second sequencing unit 1422, recommends set to sort for the size according to described forward recommendation to described forward target, obtains S set ' forward.

Such as, the forward result group of Si is (Si, Sm), (Si, Sn), (Si, So), so the target recommendation results group of Si is (Sm, Sn, So), forward result group (Si, Sm) is calculated, (Si, Sn), (Si, So) forward recommendation is carried out sequence to the element of three in (Sm, Sn, So) and is obtained S set ' forward.

3rd sequencing unit 1432, recommends set to sort for the size according to described reverse recommendation to described reverse target, obtains S set ' oppositely.

Whether the 4th detecting unit 1442, for detecting S set ' forward and S set ' oppositely containing identical target recommendation results.

Assumption set S ' forward is (Sm, Sn, So), S set ' be reversed (Sp, Sq, So), so the 4th detecting unit detects that two set are containing identical target recommendation results So.

Delete cells 1452, for S set being detected when described 4th detecting unit, ' forward and S set ' is oppositely containing identical target recommendation results, and the forward recommendation corresponding with identical target recommendation results is when being less than reverse recommendation, from described S set ' delete described identical target recommendation results forward, obtain general recommendations result.

Relatively (Si, So) forward recommendation and the size of reverse recommendation, if the forward recommendation of (Si, So) is less than reverse recommendation, so from S set ' delete target recommendation results So forward and obtain general recommendations result (Sm, Sn).

The result that the embodiment of the present invention obtains is general recommendations result, can meet the demand of most of user, particularly carries out the user searched for visitor's identity.

Refer to Figure 10, the structural representation of another embodiment of the second layer recommendation unit provided for the embodiment of the present invention; Second layer recommendation unit can comprise: second acquisition unit 1412, the 4th sequencing unit 1462, the 5th sequencing unit 1472, the 5th detecting unit 1482, delete cells 1452.

Second acquisition unit 1412, also for according to choosing the targeted customer and corresponding with described targeted customer objective result Si that described ground floor recommendation results concentrates, obtain the forward result group set participated in containing described targeted customer of Si, the set of forward target recommendation results group, the oppositely set of result group and oppositely target recommendation results group set.

Concentrate from described ground floor recommendation results and choose an objective result Si, the comment of targeted customer Uk to Si is comprised in the review information of Si, obtain forward result group set described in the forward result group set of Si and the set of forward target recommendation results group by the review information of traversal ground floor recommendation results collection and comprise the forward result group that the described targeted customer at described objective result participates in, the set of the result composition that the forward result group that the described targeted customer of described forward target recommendation results set at described objective result participates in is corresponding.In like manner can obtain the set of reverse result group and the set of reverse target recommendation results group of Si.

Second acquisition unit 1412, forward corresponding to the forward result group also participated in for the described targeted customer obtained at Si recommends coefficient value and reverse recommendation coefficient value that oppositely result group is corresponding.

The forward of each element in goal set is recommended to recommend the computing formula (num (C of coefficient value according to targeted customer Uk for objective result Si and forward _just(Si, S _j)) ÷ num (C _instead(Si, S _j))) * (num (C ₁) ÷ num (C))+1 acquisition forward recommendation coefficient value.In like manner oppositely can be recommended coefficient value.

4th sequencing unit 1462, for recommending the size of coefficient value to recommend set to sort to described forward target according to forward, obtains S set ' forward.

5th sequencing unit 1472, for according to oppositely recommending the size of coefficient value to recommend set to sort to described reverse target, obtains S set ' oppositely.

Whether the 5th detecting unit 1482, for detecting S set ' forward and S set ' oppositely containing identical target recommendation results.

Delete cells 1452, for S set being detected when described 5th detecting unit, ' forward and S set ' is oppositely containing identical target recommendation results, and the forward corresponding with identical target recommendation results is when recommending coefficient value to be less than oppositely to recommend coefficient value, from described S set ' delete described identical target recommendation results forward, obtain personalized recommendation result.

The result that the embodiment of the present invention obtains is personalized recommendation result, is to carry out recommending for the evaluation information of targeted customer, makes recommendation results have more specific aim.

One of ordinary skill in the art will appreciate that all or part of flow process realized in above-described embodiment method, that the hardware that can carry out instruction relevant by computer program has come, described program can be stored in a computer read/write memory medium, this program, when performing, can comprise the flow process of the embodiment as above-mentioned each side method.Wherein, described storage medium can be magnetic disc, CD, read-only store-memory body (Read-Only Memory, ROM) or random store-memory body (Random Access Memory, RAM) etc.

Above disclosedly be only present pre-ferred embodiments, certainly can not limit the interest field of the present invention with this, therefore according to the equivalent variations that the claims in the present invention are done, still belong to the scope that the present invention is contained.

Claims (24)

1. a method for information recommendation, is characterized in that, comprising:

Obtain user's evaluation information that the Search Results obtained by search keyword concentrates every objective result, determine that any two objective results are that forward is recommended relation or oppositely recommends relation according to described user's evaluation information, to calculate the forward recommendation of any two objective results or reverse recommendation;

Property value corresponding to every objective result is generated according to described forward recommendation and described reverse recommendation;

According to property value corresponding to described every objective result described search result set screened and sort, to obtain the recommendation results collection of the objective result after comprising screening, and described recommendation results collection is sent to subscriber equipment, show described recommendation results collection to make described subscriber equipment.

2. method according to claim 1, it is characterized in that, the Search Results that described acquisition is obtained by search key concentrates user's evaluation information of every objective result, determine that any two objective results are that forward is recommended relation or oppositely recommends relation according to described user's evaluation information, to calculate the forward recommendation of any two objective results or reverse recommendation, comprising:

Obtain user's evaluation information that the Search Results obtained by search key concentrates every objective result;

Detect the content relation of user's evaluation information of any two objective results;

Be that same or analogous two objective results are set as that forward recommends relation by described content relation, and two objective result composition forward result groups of relation will be recommended for described forward, to calculate forward recommendation corresponding to described forward result group;

Be that two contrary objective results are set as oppositely recommending relation by described content relation, and the two objective results for described reverse recommendation relation are formed reverse result group, to calculate reverse recommendation corresponding to described reverse result group.

3. method according to claim 2, is characterized in that, describedly generates property value corresponding to every objective result according to described forward recommendation and described reverse recommendation, comprising:

The forward recommendation corresponding according to the number of the number of forward result group, oppositely result group, each forward result group and reverse recommendation corresponding to each reverse result group calculate user's weighted value;

Generate property value corresponding to described objective result according to the number of the number of the forward result group corresponding with described objective result, oppositely result group, forward recommendation, oppositely recommendation, user's weighted value and extended attribute value, described extended attribute at least comprises volumes of searches, click volume, portfolio;

Property value corresponding for every objective result is formed property value matrix, and the property value that described in the behavior of described property value matrix, every objective result is corresponding, is classified as the set of kind attributes value.

4. method according to claim 3, it is characterized in that, the described property value corresponding according to described every objective result screens described search result set and sorts, to obtain the recommendation results collection of the objective result after comprising screening, and described recommendation results collection is sent to subscriber equipment, to make described subscriber equipment show described recommendation results collection, comprising:

According to the property value matrix that the property value corresponding by every objective result is formed, temperature screening and temperature sequence are carried out, to obtain ground floor recommendation results collection to described search result set;

Relation, described reverse recommendation relation and described ground floor recommendation results collection is recommended to generate second layer recommendation results collection according to described forward;

Described second layer recommendation results collection comprises general recommendations result and personalized recommendation result;

Described second layer recommendation results collection is sent to subscriber equipment, shows described recommendation results collection to make described subscriber equipment.

5. method according to claim 4, is characterized in that, the property value matrix that described basis is made up of the property value that every objective result is corresponding carries out temperature screening and temperature sequence to described search result set, to obtain ground floor recommendation results collection, comprising:

Regular described property value matrix, with generating recommendations factor matrix;

According to the set of described recommendation factor matrix generated equivalence relation;

According to relation of equivalence set, classification is carried out to described recommendation factor matrix and obtain content recommendation set;

Optimize described content recommendation set and generate optimum recommendation factor matrix, the result often corresponding to row in described optimum recommendation factor matrix is the objective result after temperature screening;

Calculate the recommendation that described optimum recommends factor matrix often to go, and recommend the result often corresponding to row in factor matrix to carry out temperature sequence from height to low according to the size of described recommendation, to obtain ground floor recommendation results collection described optimum.

6. method according to claim 5, is characterized in that, described regular described property value matrix, with generating recommendations factor matrix, comprising:

Calculate the weighting coefficient of every Column Properties value of described property value matrix, being multiplied with the property value of described weighting coefficient corresponding row by described weighting coefficient obtains Update attribute value;

By the segmentation of described Update attribute value, and the number of the section of determination and segment value;

Described Update attribute value is mapped to described segment value, to generate mapping result;

Described property value matrix is upgraded, to obtain recommending factor matrix according to described mapping result.

7. method according to claim 6, is characterized in that, described according to the set of described recommendation factor matrix generated equivalence relation, specifically comprises:

The row and column of factor matrix is recommended, with generated equivalence relation set described in linearization.

8. method according to claim 7, is characterized in that, describedly carries out classification according to relation of equivalence set to described recommendation factor matrix and obtains content recommendation set, comprising:

The number of the data source needing to recommend is determined according to the ground floor recommendation results number threshold value preset;

Set the granularity of division of recommending the factor in described recommendation factor matrix according to the Update attribute value matrix that described Update attribute value is formed, divide described data source by described granularity of division;

Choose properties collection to be recommended according to threshold matrix, detect the number of described properties collection to be recommended, the initial value of described threshold matrix is zero;

When the number of described properties collection to be recommended is greater than the number of the data source that described needs are recommended, described threshold matrix is upgraded, again properties collection to be recommended is chosen, so that the number of properties collection to be recommended after again choosing is less than the number of the properties collection to be recommended before again choosing according to the threshold matrix after described renewal;

When the number of described properties collection to be recommended is less than the number of the data source that described needs are recommended, described properties collection to be recommended is defined as content recommendation set.

9. method according to claim 8, is characterized in that, the described content recommendation set of described optimization generates optimum recommendation factor matrix, and the result often corresponding to row in described optimum recommendation factor matrix is the objective result after temperature screening, comprising:

By goal set linearization each in described content recommendation set, to generate target relation of equivalence set corresponding to each goal set;

Target relation of equivalence set corresponding to described each goal set is respectively sorted out, to obtain target content recommendation set corresponding to each goal set;

Detect the Set Status of target content recommendation set corresponding to described each goal set;

Be that the goal set of non-variable condition is added into and optimumly recommends factor matrix by described Set Status.

10. method according to claim 4, is characterized in that, described according to described forward recommendation relation, described reverse recommendation relation and described ground floor recommendation results collection generation second layer recommendation results collection, comprising:

According to described forward recommendation relation, described reverse recommendation relation, general screening and generalized sort are carried out, to obtain general recommendations result to described ground floor recommendation results collection;

Relation, described reverse recommendation relation is recommended to carry out general personalization screening and personalized ordering, to obtain personalized recommendation result to described ground floor recommendation results collection according to described forward.

11. methods according to claim 10, is characterized in that, describedly recommend relation according to described forward, described reverse recommendation relation carries out general screening and generalized sort to described ground floor recommendation results collection, to obtain general recommendations result, comprising:

Choose the objective result that described ground floor recommendation results is concentrated, obtain the group set of forward result and the set of forward target recommendation results of described objective result, the set of described forward result group comprises the forward result group at described objective result, the set of the result composition that the forward result group of described forward target recommendation results set at described objective result is corresponding;

Obtain the forward recommendation that forward result group at described objective result is corresponding, and recommend set to sort according to the size of described forward recommendation to described forward target;

Obtain the set of reverse result group and the set of reverse target recommendation results of described objective result, the set of described reverse result group comprises the reverse result group at described objective result, the set of the result composition that the reverse result group of described reverse target recommendation results set at described objective result is corresponding;

Obtain the reverse recommendation that reverse result group at described objective result is corresponding, and recommend set to sort according to the size of described reverse recommendation to described reverse target;

When detecting that the target recommendation results in the set of described forward target recommendation results is identical with the target recommendation results in the set of described reverse target recommendation results, and the forward recommendation corresponding with identical target recommendation results is when being less than reverse recommendation, delete described identical target recommendation results, to obtain general recommendations result, described general recommendations result comprises the target recommendation results in the set of described forward target recommendation results.

12. methods according to claim 10, it is characterized in that, describedly recommend relation, described reverse recommendation relation to carry out general personalization screening and personalized ordering to described ground floor recommendation results collection according to described forward, to obtain personalized recommendation result, comprising:

Choose targeted customer and corresponding with described targeted customer objective result that described ground floor recommendation results concentrates, obtain the group set of forward result and the set of forward target recommendation results of described objective result, the set of described forward result group comprises the forward result group that the described targeted customer at described objective result participates in, the set of the result composition that the forward result group that the described targeted customer of described forward target recommendation results set at described objective result participates in is corresponding;

Obtain the forward recommendation coefficient value that the forward result group of the described targeted customer's participation at described objective result is corresponding, and recommend the size of coefficient value to recommend set to sort to described forward target according to described forward;

Obtain the set of reverse result group and the set of reverse target recommendation results of described objective result, the set of described reverse result group comprises the reverse result group that the described targeted customer at described objective result participates in, the set of the result composition that the reverse result group that the described targeted customer of described reverse target recommendation results set at described objective result participates in is corresponding;

Obtain the reverse recommendation coefficient value that the reverse result group of the described targeted customer's participation at described objective result is corresponding, and recommend set to sort according to the size of described reverse recommendation coefficient value to described reverse target;

When detecting that the target recommendation results in the set of described forward target recommendation results is identical with the target recommendation results in the set of described reverse target recommendation results, and the forward recommendation corresponding with identical target recommendation results is when being less than reverse recommendation, delete described identical target recommendation results, to obtain personalized recommendation result, described personalized recommendation result comprises the target recommendation results in the set of described forward target recommendation results.

The device of 13. 1 kinds of information recommendations, is characterized in that, comprising:

Acquiring unit, concentrates user's evaluation information of every objective result for obtaining the Search Results obtained by search keyword;

First computing unit, for determining that according to described user's evaluation information any two objective results are that forward recommends relation or reverse recommendation relation to calculate the forward recommendation of any two objective results or reverse recommendation;

Generation unit, for generating property value corresponding to every objective result according to described forward recommendation and described reverse recommendation;

Screening unit, for screen described search result set according to property value corresponding to described every objective result and sort, to obtain the recommendation results collection of the objective result after comprising screening;

Recommendation unit, for described recommendation results collection is sent to subscriber equipment, shows described recommendation results collection to make described subscriber equipment.

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

First detecting unit, for detecting the content relation of user's evaluation information of any two objective results;

Setup unit, for by described content relation be same or analogous two objective results be set as forward recommend relation, and two objective result composition forward result groups of relation will be recommended for described forward, calculate forward recommendation corresponding to described forward result group with described computing unit;

Described setup unit, also for being that two contrary objective results are set as oppositely recommending relation by described content relation, and the two objective results for described reverse recommendation relation are formed reverse result group, calculate reverse recommendation corresponding to described reverse result group with described computing unit.

15. devices according to claim 13, is characterized in that, described generation unit comprises:

Weighted value computing unit, the forward recommendation corresponding for the number of the number according to forward result group, oppositely result group, each forward result group and reverse recommendation corresponding to each reverse result group calculate user's weighted value;

Attribute generation unit, generate property value corresponding to described objective result for the number of the number according to the forward result group corresponding with described objective result, oppositely result group, forward recommendation, oppositely recommendation, user's weighted value and extended attribute value, described extended attribute at least comprises volumes of searches, click volume, portfolio;

Matrix construction unit, for property value corresponding for every objective result is formed property value matrix, the property value that described in the behavior of described property value matrix, every objective result is corresponding, is classified as the set of kind attributes value.

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

Ground floor recommendation unit, for carrying out temperature screening and temperature sequence, to obtain ground floor recommendation results collection according to the property value matrix be made up of the property value that every objective result is corresponding to described search result set;

Second layer recommendation unit, for recommending relation, described reverse recommendation relation and described ground floor recommendation results collection to generate second layer recommendation results collection according to described forward, described second layer recommendation results collection comprises general recommendations result and personalized recommendation result.

17. devices according to claim 16, is characterized in that, described ground floor recommendation unit comprises:

Regular unit, for regular property value matrix, with generating recommendations factor matrix;

Relation of equivalence generation unit, for according to the set of described recommendation factor matrix generated equivalence relation;

Sorting out unit, obtaining content recommendation set for carrying out classification according to relation of equivalence set to described recommendation factor matrix;

Optimization unit, generates optimum recommendation factor matrix for optimizing described content recommendation set, and the result often corresponding to row in described optimum recommendation factor matrix is the objective result after temperature screening;

Second computing unit, for calculating the recommendation that described optimum recommends factor matrix often to go;

First sequencing unit, for recommending the result often corresponding to row in factor matrix to carry out temperature sequence from height to low according to the size of described recommendation, to obtain ground floor recommendation results collection described optimum.

18. devices according to claim 17, is characterized in that, described regular unit comprises:

3rd computing unit, for calculating the weighting coefficient of every Column Properties value of described property value matrix, being multiplied described weighting coefficient with the property value of described weighting coefficient corresponding row and obtaining Update attribute value;

Segmenting unit, for by the segmentation of described Update attribute value, and the number of the section of determination and segment value;

Map unit, for described Update attribute value is mapped to described segment value, to generate mapping result, upgrades described property value matrix according to described mapping result, to obtain recommending factor matrix.

19. devices according to claim 17, is characterized in that, described relation of equivalence generation unit comprises:

Linearizer, for recommending the row and column of factor matrix described in linearization, with generated equivalence relation set.

20. devices according to claim 17, is characterized in that, described classification unit comprises:

Quantity determining unit, for determining the number needing the data source of recommending according to the ground floor recommendation results number threshold value preset;

Granularity of division unit, sets the granularity of division of recommending the factor in described recommendation factor matrix for the Update attribute value matrix formed according to described Update attribute value, divides described data source by described granularity of division;

Second detecting unit, for detecting the number of the properties collection to be recommended chosen according to threshold matrix, the initial value of described threshold matrix is zero;

Updating block, for when being greater than the number of the data source that described needs are recommended wait the number recommending properties collection described in described second detecting unit detection, described threshold matrix is upgraded, again properties collection to be recommended is chosen, so that the number of properties collection to be recommended after again choosing is less than the number of the properties collection to be recommended before again choosing according to the threshold matrix after described renewal;

When being less than the number of the data source that described needs are recommended wait the number recommending properties collection described in described second detecting unit detects, described properties collection to be recommended is defined as content recommendation set.

21. devices according to claim 17, is characterized in that, described optimization unit comprises:

Described generation unit also for by goal set linearization each in described content recommendation set, to generate target relation of equivalence set corresponding to each goal set;

Described classification unit is also sorted out for target relation of equivalence set corresponding to described each goal set respectively, to obtain target content recommendation set corresponding to each goal set;

Described Set Status, for detecting the Set Status of target content recommendation set corresponding to described each goal set, is that the goal set of non-variable condition is added into optimum recommendation factor matrix by the 3rd detecting unit.

22. devices according to claim 16, is characterized in that, described second layer recommendation unit comprises:

General recommendations unit, for carrying out general screening and generalized sort, to obtain general recommendations result according to described forward recommendation relation, described reverse recommendation relation to described ground floor recommendation results collection;

Personalized recommendation unit, carries out general personalization screening and personalized ordering, to obtain personalized recommendation result for recommending relation, described reverse recommendation relation according to described forward to described ground floor recommendation results collection.

23. devices according to claim 22, is characterized in that, described general recommendations unit comprises:

Described acquiring unit also chooses a concentrated objective result of described ground floor recommendation results for basis, obtain the group set of forward result and the set of forward target recommendation results of described objective result, the set of described forward result group comprises the forward result group at described objective result, the set of the result composition that the forward result group of described forward target recommendation results set at described objective result is corresponding;

Described acquiring unit is also for obtaining forward recommendation corresponding to forward result group at described objective result;

Second sequencing unit, recommends set to sort for the size according to described forward recommendation to described forward target;

Described acquiring unit is also for obtaining the set of reverse result group and the set of reverse target recommendation results of described objective result, the set of described reverse result group comprises the reverse result group at described objective result, the set of the result composition that the reverse result group of described reverse target recommendation results set at described objective result is corresponding;

Described acquiring unit is also for obtaining reverse recommendation corresponding to reverse result group at described objective result;

3rd sequencing unit, recommends set to sort for the size according to described reverse recommendation to described reverse target;

Whether the 4th detecting unit is identical with the target recommendation results in the set of described reverse target recommendation results for the target recommendation results detected in the set of described forward target recommendation results;

Delete cells, for detecting that the recommendation objective result in the set of described forward target recommendation results is identical with the target recommendation results in the set of described reverse target recommendation results when described 4th detecting unit, and the forward recommendation corresponding with identical target recommendation results is when being less than reverse recommendation, delete described identical target recommendation results, to obtain general recommendations result, described general recommendations result comprises the target recommendation results in the set of described forward target recommendation results.

24. devices according to claim 22, is characterized in that, described personalized recommendation unit comprises:

Described acquiring unit is also for according to choosing the targeted customer and the objective result corresponding with described targeted customer that described ground floor recommendation results concentrates, obtain the group set of forward result and the set of forward target recommendation results of described objective result, the set of described forward result group comprises the forward result group that the described targeted customer at described objective result participates in, the set of the result composition that the forward result group that the described targeted customer of described forward target recommendation results set at described objective result participates in is corresponding;

Described acquiring unit also recommends coefficient value for the forward that the forward result group obtaining the described targeted customer's participation at described objective result is corresponding;

4th sequencing unit, recommends set to sort for recommending the size of coefficient value according to described forward to described forward target;

Described acquiring unit is also for obtaining the set of reverse result group and the set of reverse target recommendation results of described objective result, the set of described reverse result group comprises the reverse result group that the described targeted customer at described objective result participates in, the set of the result composition that the reverse result group that the described targeted customer of described reverse target recommendation results set at described objective result participates in is corresponding;

The reverse recommendation coefficient value that the reverse result group that described acquiring unit also participates in for the described targeted customer obtained at described objective result is corresponding;

5th sequencing unit, recommends set to sort for the size according to described reverse recommendation coefficient value to described reverse target;

5th detecting unit, whether identical with the target recommendation results in the set of described reverse target recommendation results for the target recommendation results in the set of described forward target recommendation results,

Described delete cells, also for detecting that the target recommendation results in the set of described forward target recommendation results is identical with the target recommendation results in the set of described reverse target recommendation results when the 5th detecting unit, and the forward recommendation corresponding with identical target recommendation results is when being less than reverse recommendation, delete described identical target recommendation results, to obtain personalized recommendation result, described personalized recommendation result comprises the target recommendation results in the set of described forward target recommendation results.