CN103714135A - MapReduce recommendation method and system of second-degree interpersonal relationships of massive users - Google Patents

Abstract

The invention provides a MapReduce recommendation method and system of second-degree interpersonal relationships of massive users. The method includes the steps that in attention relationships of the massive users, the second-degree interpersonal relationships of the users are worked out through the two-step MapReduce calculation method based on first-degree interpersonal relationships of the users, and then the second-degree interpersonal relationships of the users are recommended. By means of the MapReduce recommendation method and system of the second-degree interpersonal relationships of the massive users, repeated querying can be avoided, and recommendation accuracy and computation efficiency can be improved.

Description

Two degree relationship among persons MapReduce recommend method and systems of mass users

Technical field

The present invention relates to mass users relationship among persons technical field, more specifically, relate to a kind of two degree relationship among persons MapReduce recommend method and systems of mass users.

Background technology

Along with SNS(Social Networking Services, social network services) rise of community, the user of magnanimity pays close attention to relation and produces.The mass users that SNS community has is paid close attention to relation, is a valuable data wealth, how user's once human connection is converted into the problem that ShiSNS community faces that is worth.In the process of SNS community operation simultaneously, once human connection tends towards stability, the once relationship among persons based on mass users, and two degree relationship among persons of analysis user, in conjunction with the line business of SNS community, develop more competitive application.So both can bring feeling of freshness to user, guiding user deepens the understanding to user's two degree human connections, can extend the viscosity that user uses community again, extends the time that user uses community, for community brings more flow and profit, make user and value be unified and transform.

In the face of user's concern relation, two degree human connections of digging user are the problems that community need to solve, and Fig. 1 shows the flow process of existing processing user's two degree human connections, as shown in Figure 1,

S110: start;

S120: mass users is paid close attention to relation, A---B, B---C;

S130: search user's S set ET1 that user A pays close attention to;

S140: inquiring user is in conjunction with follower's S set ET2 of SET1;

In S150:SET2, remove the user that user A has paid close attention to;

S160: the two degree relationship among persons set of user A are SET2;

S170: the indirect recommendation number of times of statistics two degree human connections;

S180: finish.

In the face of the calculating of mass users data volume, the processing gimmick conventionally adopting is exactly parallel computation.There is the situation of inquiry repeatedly in the disposal route of above-mentioned Fig. 1, as: user A and user B have paid close attention to user C, and, when the two degree human connection of calculating user A and user B, C is indirect concern, all needs the follower of inquiring user C.Inquiry repeatedly causes the raising that can not walk abreast of the efficiency of parallel computation, and treatment effeciency is also very low.

SNS community, when calculating two degree relationship among persons, not only needs to improve the efficiency that large data are calculated, and when carrying out two degree relationship among persons recommendations, once human connection need to consider to recommend weight simultaneously; If the bean vermicelli number of once human connection is many, user is more active, the recommendation weight of once human connection should improve accordingly.

In order to solve the above problems, a kind of new computing method need to be provided, avoid repeating inquiry, improve treatment effeciency; Simultaneously, when once considering relationship among persons, consider bean vermicelli number, the liveness of once human connection, give corresponding two degree human connections and recommend weights.

Summary of the invention

In view of the above problems, the object of this invention is to provide a kind of two degree relationship among persons MapReduce recommend method and systems of mass users, to solve the problem that repeats inquiry, recommendation accuracy and improve counting yield.

According to an aspect of the present invention, a kind of two degree relationship among persons MapReduce recommend methods of mass users are provided, in the concern relation of mass users, according to user's once relationship among persons, by the computing method of two step MapReduce, obtain user's two degree relationship among persons, and recommended; Wherein,

In the concern relation of mass users, if user A pays close attention to user B, user C pays close attention to user A, and user B is the good friend of user A, and user C is the bean vermicelli of user A; And if user C recommends to pay close attention to user B by user A, user B is the two degree human connections of user C, user A is the indirect follower between user C and user B; User C pays close attention to user B by other users except user A in the concern relation in mass users, and other users are the indirect follower of user C and user B; And,

In the computing method of first step MapReduce in the computing method of two step MapReduce, in the concern relation of mass users, according to the once human connection of user A, the bean vermicelli of the good friend of user A and user A is merged and sorted out, obtain good friend's set of user A and the bean vermicelli set of user A, the quantity of the bean vermicelli of while counting user A; Wherein,

User B is in good friend's set of user A, and user C is in the bean vermicelli set of user A; And,

In the concern relation of mass users, according to other users' once human connection, obtain good friend's set of other users and other user's bean vermicelli set, while the bean vermicelli quantity of adding up other users;

In the computing method of second step MapReduce in the computing method of two step MapReduce, in the concern relation of mass users, the bean vermicelli set of user A recommends the good friend who pays close attention to user A to gather by user A, two degree human connections of the bean vermicelli set of user A are good friend's set of user A, wherein, user A is the indirect follower of the concentrated user B of the good friend of user C in the bean vermicelli set of user A and user A; And,

According to the quantity of the bean vermicelli of user A, obtain the recommendation weighted value of user A, according to the quantity of other users' bean vermicelli, obtain other users' recommendation weighted value;

Thereby the recommendation weighted value of user A and other users' recommendation weighted value is merged and obtains maximum recommendation weighted value, and the quantity of the indirect follower between counting user C and user B, wherein, the quantity of the indirect follower between user C and user B is user A and other users' quantity sum;

User B is recommended to described user C, maximum recommendation weighted value and all indirect followers are recommended to user C simultaneously, with two degree relationship among persons of completing user, recommend.

Wherein, in the bean vermicelli set of user A, by user A, recommend to pay close attention in the process of good friend's set of user A,

If have user B in the bean vermicelli set of user A, in good friend's set of user A, also there is user B simultaneously, two of user B degree human connections can not be user B itself;

If the user C of the bean vermicelli set of user A has paid close attention to the user B in the good friend set of user A, two of user C degree human connections can not be user B.

Wherein, user A adopts the mode of LogNormal Function Fitting to obtain the recommendation weighted value of user A according to the bean vermicelli quantity of user A;

LogNormal Function Fitting formula is:

$f(x,\mathrm{\μ},\mathrm{\σ})=\frac{1}{\mathrm{x\σ}\sqrt{2\mathrm{\π}}}{e}^{-\frac{{\left(\ln x\right)}^2}{{2\mathrm{\σ}}^2}}$

Wherein, in LogNormal Function Fitting, X is the array note of all bean vermicelli quantity of mass users, and μ is the mean value of the logarithm of array X, and σ is the data expectation of array X logarithm; Mathematical expectation and the variance of array X are respectively E[X] Var[X];

The fitting formula of parameter is:

$\mathrm{\μ}=\ln \left(E\right[X\left]\right)-0.5*\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

${\mathrm{\σ}}^2=\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

The recommendation weighted value that the directed integration numerical value of LogNormal function is user A.

On the other hand, the present invention also provides a kind of two degree relationship among persons MapReduce commending systems of mass users, commending system, for the concern relation in mass users, according to user's once relationship among persons, by first step MapReduce computing unit and second step MapReduce computing unit two, obtain user's two degree relationship among persons, and recommended; Wherein,

In the concern relation of mass users, if user A pays close attention to user B, user C pays close attention to user A, and user B is the good friend of user A, and user C is the bean vermicelli of user A; And if user C recommends to pay close attention to user B by user A, user B is the two degree human connections of user C, user A is the indirect follower between user C and user B; If user C pays close attention to user B by other users except user A in the concern relation in mass users, other users are the indirect follower of user C and user B;

First step MapReduce computing unit, for the concern relation in mass users, according to the once human connection of user A, the bean vermicelli of the good friend of user A and user A is merged and sorted out, obtain good friend's set of user A and the bean vermicelli set of user A, the quantity of the bean vermicelli of while counting user A; Wherein, user B is in good friend's set of user A, and user C is in the bean vermicelli set of described user A; And,

In the concern relation of mass users, according to other users' once human connection, obtain good friend's set of other users and the bean vermicelli set of the user except user A, while the bean vermicelli quantity of adding up other users;

Second step MapReduce computing unit comprises: recommend weighted value acquiring unit and two degree human connection recommendation unit; Wherein,

Recommend weighted value acquiring unit, for obtaining user's recommendation weighted value; Wherein, in the concern relation of mass users, the bean vermicelli set of user A recommends the good friend who pays close attention to user A to gather by user A, two degree human connections of the bean vermicelli set of user A are good friend's set of user A, wherein, user A is the indirect follower of the concentrated user B of the good friend of user C in the bean vermicelli set of user A and user A; And,

According to the quantity of the bean vermicelli of user A, obtain the recommendation weighted value of user A, according to the quantity of other users' bean vermicelli, obtain other users' recommendation weighted value;

Two degree human connection recommendation unit, recommend for two degree human connections of completing user; Wherein, thereby the recommendation weighted value of user A and other users' recommendation weighted value is merged and obtains maximum recommendation weighted value, and the quantity of the indirect follower between counting user C and described user B, wherein, the quantity of the indirect follower between user C and described user B is user A and other users' quantity sum;

Finally user B is recommended to user C, maximum recommendation weighted value and all indirect followers are recommended to user C simultaneously, with two degree human connections of completing user, recommend.

From technical scheme above, two degree relationship among persons MapReduce recommend method and systems of mass users provided by the invention, can obtain following beneficial effect:

1) can avoid the problem that in calculating, process is inquired about repeatedly, improve counting yield, wherein, it is relevant that coefficient is closed in the average concern of the efficiency YuSNS community of raising, pays close attention to network of personal connections more complicated, and efficiency improves more obvious;

2) weight of utilizing once human connection to recommend, it is more accurate to make to recommend;

3) if improve the processing time of calculating or process more data, can carry out easily horizontal extension.

In order to realize above-mentioned and relevant object, one or more aspects of the present invention comprise below by the feature that describes in detail and particularly point out in the claims.Explanation below and accompanying drawing describe some illustrative aspects of the present invention in detail.Yet, the indication of these aspects be only some modes that can use in the variety of way of principle of the present invention.In addition, the present invention is intended to comprise all these aspects and their equivalent.

Accompanying drawing explanation

By reference to the content below in conjunction with the description of the drawings and claims, and along with understanding more comprehensively of the present invention, other object of the present invention and result will be understood and easy to understand more.In the accompanying drawings:

Fig. 1 is the schematic flow sheet of existing processing user's two degree human connections;

Fig. 2 is according to single schematic flow sheet to user's two degree relationship among persons recommend methods in the mass users concern relation of the embodiment of the present invention;

Fig. 3 is according to the schematic flow sheet of two degree relationship among persons MapReduce recommend methods of the mass users of the embodiment of the present invention;

Fig. 4 is the two degree relationship among persons MapReduce commending system logic diagrams according to the mass users of the embodiment of the present invention.

In institute's drawings attached, identical label is indicated similar or corresponding feature or function.

Embodiment

In the following description, for purposes of illustration, for the complete understanding to one or more embodiment is provided, many details have been set forth.Yet, clearly, also can in the situation that there is no these details, realize these embodiment.

In order to solve foregoing problems, the present invention adopts the thought of parallel computation, utilize MapReduce programming framework, write MapReduce program, come analysis and calculation mass users to pay close attention to two degree relationship among persons of relation, Fig. 2 shows according to single flow process to user's two degree relationship among persons recommend methods in the mass users concern relation of the embodiment of the present invention; As shown in Figure 2, its treatment scheme is as follows:

S210: start;

S220: mass users is paid close attention to relation, if A---B, C---A;

S230: user A is as key assignments, and the user C of the user B that user A is paid close attention to and concern user A, does a parallel distribution;

S240: the bean vermicelli of user A and good friend, do one and concentrate merging to sort out;

S250: give the good friend user B of the user C recommendation user A that pays close attention to user A, user A is indirect follower; Using the bean vermicelli user C of user A as key assignments, do a parallel distribution;

S260: do one according to the key assignments of step S250 and merge classification, obtain all two degree human connections of the bean vermicelli user C of user A;

S270: output obtains user's two degree human connections, and add up indirect follower and indirect follower's number;

S280: finish.

In two above-mentioned degree relationship among persons flow processs, the recommendation of the two degree human connections that the user A of take concern user B and user C concern user A carry out as basis.Below with reference to accompanying drawing, specific embodiments of the invention are described in detail.

Fig. 3 shows according to the flow process of two degree relationship among persons MapReduce recommend methods of the mass users of the embodiment of the present invention.As shown in Figure 3, the invention provides a kind of two degree relationship among persons MapReduce recommend methods of mass users, comprise: in the concern relation of mass users, according to user's once relationship among persons, by the computing method of two step MapReduce, obtain user's two degree relationship among persons, and recommended.It should be noted that, the present invention adopts the method for parallel computation, utilize MapReduce programming framework, write MapReduce program, come analysis and calculation mass users to pay close attention to two degree relationship among persons of relation, adopted in the present invention two step MapReduce to realize the recommendation of user's two degree relationship among persons.

In the concern relation of mass users, if user A pays close attention to user B, user C pays close attention to user A, and user B is the good friend of user A, and user C is the bean vermicelli of user A; And if user C recommends to pay close attention to user B by user A, user B is the two degree human connections of user C, user A is the indirect follower between user C and user B.

And if user C pays close attention to user B by other users except user A in mass users, other users are the indirect follower of user C and user B; Other users that mention in the present invention all refer to the indirect follower of user C and user B.

In a specific embodiment of the present invention, for user's once human connection, pay close attention to relation, as: user A pays close attention to user B, illustrates that user A is user B bean vermicelli, and that user B is the good friend of user A.

User based on above-mentioned pays close attention to relation, need to, according to user's once relationship among persons, to user, recommend two degree human connections: " they have paid close attention to this user's good friend ", MapReduce completes in two steps altogether.

S310: in the computing method of the first step MapReduce in the computing method of two step MapReduce, in the concern relation of mass users, according to the once human connection of user A, the bean vermicelli of the good friend of user A and user A is merged and sorted out, obtain good friend's set of user A and the bean vermicelli set of user A, the quantity of the bean vermicelli of while counting user A; Wherein, user B is in good friend's set of user A, and user C is in the bean vermicelli set of user A; And,

In the concern relation of mass users, according to other users' once human connection, obtain other users' good friend's set and other users' bean vermicelli set, while the bean vermicelli quantity of adding up other users.

Particularly, in the computing method of the first step MapReduce in the computing method of two step MapReduce, be mainly that user's good friend and bean vermicelli are done to a parallel distribution, then user's bean vermicelli and good friend are done to concentrated a classification.In the concern relation of mass users, A---B, C---A; Be that user A pays close attention to user B, user C pays close attention to user A; User B is the good friend of user A, and user C is the bean vermicelli of user A.

That is to say, the bean vermicelli of the good friend of user A and user A is cooked to a parallel distribution, then the bean vermicelli of user A and user's good friend is done to one and concentrate classification.

The computing method of first step MapReduce divide two stages to complete, i.e. Map stage and Reduce stage.

Wherein, at the Map of first step MapReduce in the stage, by the id of user A, as key assignments, the bean vermicelli of the good friend of concurrent dispatch user A and user A; That is to say, by the id of user A, as the concurrent dispatch user B of key assignments and user C.

At the Reduce of first step MapReduce, in the stage, merge bean vermicelli and the good friend of user A, the set of output user A good friend and bean vermicelli; That is to say, merge user B and user C, output user B and user C.

It should be noted that, in the concern relation of mass users, user A, user B and user C are illustrated as concrete user, in fact, in good friend's set of user A, except user B, also have other a lot of users, in the bean vermicelli set of user A, except user C, also have other a lot of users; And, in the concern relation of mass users, can obtain arbitrary user's good friend's set and bean vermicelli set, therefore, according to the once human connection of the user except user A, obtain the good friend's set of the user except user A and the bean vermicelli set of the user except user A, the while is also added up its bean vermicelli quantity.Statistics bean vermicelli quantity is to prepare for the needs of next step computing method.

Therefore, when obtaining good friend's set of user A and the bean vermicelli set of user A, and the quantity of the bean vermicelli of counting user A; And, in the concern relation of mass users, also want the bean vermicelli quantity of other users except user A between counting user C and user B.

S320: in the computing method of the second step MapReduce in the computing method of two step MapReduce, in the concern relation of mass users, the bean vermicelli set of user A recommends the good friend who pays close attention to user A to gather by user A, two degree human connections of the bean vermicelli set of user A are good friend's set of user A, wherein, user A is the indirect follower of the concentrated user B of the good friend of user C in the bean vermicelli set of user A and user A; And,

According to the quantity of the bean vermicelli of user A, obtain the recommendation weighted value of user A, according to the quantity of other users' bean vermicelli, obtain other users' recommendation weighted value;

Thereby the recommendation weighted value of user A and other users' recommendation weighted value is merged and obtains maximum recommendation weighted value, and the quantity of the indirect follower between counting user C and user B, wherein, the quantity of the indirect follower between user C and described user B is user A and other users' quantity sum;

User B is recommended to user C, maximum recommendation weighted value and all indirect followers after merging are recommended to user C simultaneously, with two degree relationship among persons of completing user, recommend.

Particularly, in the computing method of two step MapReduce in the computing method of second step MapReduce, the bean vermicelli of user A be user C as key assignments, the good friend of user A is done to a parallel distribution; Then the indirect follower's of human connection quantity is spent in two of counting user C degree human connections and two.

In the computing method of two step MapReduce, in the computing method of second step MapReduce, being also divided into for two stages completes, and is the Map stage of second step MapReduce and the Reduce stage of second step MapReduce.

The Map stage of second step MapReduce, using the bean vermicelli of user A as key assignments, the good friend of parallel dispatch user A, the indirect follower between the bean vermicelli of user A and the good friend of user A is user A, and obtains the recommendation weighted value of user A.That is to say, using user C as key assignments, parallel dispatch user B, the indirect follower between user B and user C is user A.

Wherein, during the recommendation weighted value of the indirect follower user A between considering between user C and user B, adopt the bean vermicelli quantity segmentation weight of indirect follower user A to consider; The bean vermicelli quantity of user A is with the mapping relations that have of the recommendation weight of user A, can be according to the frequency probability distribution situation of the bean vermicelli quantity of user A in SNS community mass users, adopt the mode of LogNormal Function Fitting, simulate the distribution function figure of bean vermicelli quantity, the recommendation weighted value of user A bean vermicelli quantity is the directed integration numerical value of this distribution function.

LogNormal Function Fitting formula is:

$f(x,\mathrm{\μ},\mathrm{\σ})=\frac{1}{\mathrm{x\σ}\sqrt{2\mathrm{\π}}}{e}^{-\frac{{\left(\ln x\right)}^2}{{2\mathrm{\σ}}^2}}$

Wherein, in above-mentioned Function Fitting, if the array of the bean vermicelli quantity that X is all mass users, μ is the mean value of the logarithm of described array X, and σ is the data expectation of described array X logarithm.Mathematical expectation and the variance of array X are respectively E[X] Var[X];

The fitting formula of parameter is:

$\mathrm{\μ}=\ln \left(E\right[X\left]\right)-0.5*\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

${\mathrm{\σ}}^2=\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

The recommendation weight that the directed integration numerical value of LogNormal function is user A.

In the present invention, all indirect followers' recommendation weighted value all obtains by LogNormal Function Fitting according to indirect follower's bean vermicelli quantity.

That is to say, if user C pays close attention to user B by other users except user A in the concern relation in mass users, other users are the indirect follower of user C and user B; In the present invention, other users' recommendation weighted value also obtains by LogNormal Function Fitting according to its bean vermicelli quantity.

In the present invention, the Map stage of second step MapReduce, also have a very important duplicate removal task, there is situation below:

In the bean vermicelli set of user A, by user A, recommend to pay close attention in the process of good friend's set of user A, if having user B in the bean vermicelli set of user A, also have user B in good friend's set of user A simultaneously, two of user B degree human connections can not be user B itself; Meanwhile,

The user B during if the good friend that the bean vermicelli of user A concentrates the user C closing to pay close attention to user A gathers, two of user C degree human connections can not be user B.

Illustrate two kinds of sights:

Sight one: if if user A and user B pay close attention to mutually, user B is that the bean vermicelli of user A is also the good friend of user A; But at this time can not recommend user B by user B.

Sight two: if user A pays close attention to user B, user C pays close attention to user A, and the bean vermicelli of user A is user C,, when recommending to pay close attention to B to user C, need to judge that user C does not pay close attention to user B, if user C has paid close attention to user B, will not recommend user B.

In the present invention, consider the once recommendation weighted value of relationship among persons, can improve the accuracy that user recommends.

In the Reduce stage of second step MapReduce, merge user's recommendation weights, adds up indirect follower's number.

That is to say, the Reduce stage of second step MapReduce, user A and other users' recommendation weighted value is being merged and sorted out, and add up indirect follower and quantity thereof, after merging classification, draw maximum recommendation weighted value, maximum recommendation weighted value and all indirect follower and quantity are recommended to user C in the lump, and two degree human connections of completing user are recommended.

Last Output rusults is: A W6.898B, D, E, F4

The recommendation of two degree human connections is explanatoryly: user W is recommended to user A, and recommending weights is 6.898, and rationale for the recommendation is that user A your 4 good friend B, D, E, F have also paid close attention to W.

All indirect followers of output user, can improve the interpretation of two degree human connections, and such rationale for the recommendation user ratio is easier to understand, and avoids user's information anxiety problem, contribute to improve the conversion ratio of recommending.

The detailed process of the above-mentioned two degree relationship among persons recommendations for mass users of the present invention; The present invention has adopted parallel calculating method that two degree relationship among persons are calculated and recommended, and can, according to the scale of deal with data, increase computing node and carry out horizontal extension; That is to say, the once relationship among persons of mass users is calculated.

In order to verify counting yield, adopt two kinds of computing method to contrast, control methods is as follows:

Method one: user's once relationship among persons deposits in relevant database, adopts the SQL query method in relevant database, does not consider in the situation of once human connection recommendation weight, calculates user's two degree relationship among persons and recommends.

Method two: adopt method of the present invention, user's once relationship among persons deposits in distributed file system HDFS, adopts the method for MapReduce of the present invention to calculate.Consider the once recommendation weight of relationship among persons, the maximum Reduce number that Reduce number is single node, can guarantee to adopt single node to calculate like this.

From above-mentioned contrast, can learn, the present invention can avoid the problem that in calculating, process is inquired about repeatedly, improves counting yield, and has considered the weight that once human connection is recommended, and user is recommended more accurate.

Corresponding with said method, the present invention also provides a kind of two degree relationship among persons MapReduce commending systems of mass users.Fig. 4 shows the two degree relationship among persons MapReduce commending system logical organizations according to the mass users of the embodiment of the present invention.

As shown in Figure 4, two degree relationship among persons MapReduce commending systems 400 of mass users provided by the invention are for the concern relation in mass users, according to user's once relationship among persons, by first step MapReduce computing unit and second step MapReduce computing unit, obtain user's two degree relationship among persons, and recommended; Wherein, first step MapReduce computing unit and second step MapReduce computing unit are respectively used to carry out the first step and the second step in the computing method of aforementioned two step MapReduce.Wherein,

In the concern relation of mass users, if user A pays close attention to user B, user C pays close attention to user A, and user B is the good friend of user A, and user C is the bean vermicelli of user A; And,

If user C recommends to pay close attention to user B by user A, user B is the two degree human connections of user C, and user A is the indirect follower between user C and user B;

If user C recommends to pay close attention to user B by other users except user A in mass users, other users are the indirect follower of user C and user B.

Commending system 400 comprises first step MapReduce computing unit and second step MapReduce computing unit.

Wherein, first step MapReduce computing unit 410 is in the computing method of first step MapReduce, in the concern relation of mass users, according to the once human connection of user A, the bean vermicelli of the good friend of user A and user A is merged and sorted out, obtain good friend's set of user A and the bean vermicelli set of user A, the quantity of the bean vermicelli of while counting user A; Wherein,

User B is in good friend's set of user A, and user C is in the bean vermicelli set of described user A; And,

In the concern relation of mass users, according to other users' once human connection, obtain other users' good friend's set and other users' bean vermicelli set, while the bean vermicelli quantity of adding up other users.

Second step MapReduce computing unit 420 comprises recommends weighted value acquiring unit 421 and two degree human connection recommendation unit 422.

Recommend weighted value acquiring unit 421 for obtaining user's recommendation weighted value; Wherein, in the concern relation of mass users, the bean vermicelli set of user A recommends the good friend who pays close attention to user A to gather by user A, and two degree human connections of the bean vermicelli set of user A are good friend's set of user A, wherein,

User A is the indirect follower of the concentrated user B of the good friend of user C in the bean vermicelli set of user A and user A; And,

According to the quantity of the bean vermicelli of user A, obtain the recommendation weighted value of user A,

According to the quantity of other users' bean vermicelli, obtain other users' recommendation weighted value.

Two degree human connection recommendation unit 422, recommend for two degree human connections of completing user; Wherein, thereby the recommendation weighted value of user A and other users' recommendation weighted value is merged and obtains maximum recommendation weighted value, and the quantity of the indirect follower between counting user C and described user B, wherein,

The quantity of indirect follower between user C and user B is user A and other users' quantity sum;

Finally user B is recommended to user C, maximum recommendation weighted value and all indirect followers are recommended to user C simultaneously, with two degree human connections of completing user, recommend.

Wherein, recommend weighted value acquiring unit 422 by user A, to recommend to pay close attention in the process of good friend's set of user A in the bean vermicelli set of user A;

If have user B in the bean vermicelli set of user A, in good friend's set of user A, also there is described user B simultaneously, two of user B degree human connections can not be user B itself; Meanwhile,

The user B during if the good friend that the bean vermicelli of user A concentrates the user C closing to pay close attention to user A gathers, two of user C degree human connections can not be user B.

In recommending weighted value acquiring unit 421, user A adopts the mode of LogNormal Function Fitting to obtain the recommendation weighted value of user A according to the bean vermicelli quantity of user A;

LogNormal Function Fitting formula is:

$f(x,\mathrm{\μ},\mathrm{\σ})=\frac{1}{\mathrm{x\σ}\sqrt{2\mathrm{\π}}}{e}^{-\frac{{\left(\ln x\right)}^2}{{2\mathrm{\σ}}^2}}$

Wherein, in LogNormal Function Fitting, if the array of the bean vermicelli quantity that X is all mass users, μ is the mean value of the logarithm of described array X, and σ is the data expectation of described array X logarithm.Mathematical expectation and the variance of array X are respectively E[X] Var[X];

The fitting formula of parameter is:

$\mathrm{\μ}=\ln \left(E\right[X\left]\right)-0.5*\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

${\mathrm{\σ}}^2=\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

The recommendation weighted value that the directed integration numerical value of LogNormal function is user A.

By above-mentioned embodiment, can find out, two degree relationship among persons MapReduce recommend method and systems of mass users provided by the invention, can avoid the problem that in calculating, process is inquired about repeatedly, improve counting yield, it is relevant that coefficient is closed in the average concern of the efficiency YuSNS community of improving, pay close attention to network of personal connections more complicated, efficiency improves more obvious; The weight of utilizing once human connection to recommend, it is more accurate to make to recommend; In mass users, if improve the processing time of calculating or process more data, can carry out easily horizontal extension.

Two degree relationship among persons MapReduce recommend method and systems of the mass users proposing according to the present invention have been described in the mode of example above with reference to accompanying drawing.But, it will be appreciated by those skilled in the art that two degree relationship among persons MapReduce recommend method and systems of the mass users that proposes for the invention described above, can also on the basis that does not depart from content of the present invention, make various improvement.Therefore, protection scope of the present invention should be determined by the content of appending claims.

Claims (6)

1. two of mass users spend relationship among persons MapReduce recommend methods, comprising:

In the concern relation of mass users, according to user's once relationship among persons, by the computing method of two step MapReduce, obtain user's two degree relationship among persons, and recommended; Wherein,

In the concern relation of described mass users, if user A pays close attention to user B, user C pays close attention to user A, and user B is the good friend of user A, and user C is the bean vermicelli of user A; And if user C recommends to pay close attention to user B by user A, user B is the two degree human connections of user C, user A is the indirect follower between user C and user B; If user C pays close attention to user B by other users except user A in the concern relation in described mass users, other users are the indirect follower of user C and user B; And,

In the computing method of first step MapReduce in the computing method of described two step MapReduce, in the concern relation of described mass users, according to the once human connection of described user A, the bean vermicelli of the good friend of user A and user A is merged and sorted out, obtain good friend's set of user A and the bean vermicelli set of user A, the quantity of the bean vermicelli of while counting user A;

Wherein, described user B is in good friend's set of described user A, and described user C is in the bean vermicelli set of described user A; And, in the concern relation of described mass users, according to described other users' once human connection, obtain good friend's set of described other users and described other users' bean vermicelli set, while described other users' of statistics bean vermicelli quantity;

In the computing method of second step MapReduce in the computing method of described two step MapReduce, in the concern relation of described mass users, the bean vermicelli set of described user A recommends the good friend who pays close attention to described user A to gather by described user A, two degree human connections of the bean vermicelli set of described user A are good friend's set of described user A, wherein, the indirect follower of the described user B that the described user C in the bean vermicelli set that described user A is described user A and the good friend of described user A concentrate; And,

According to the quantity of the bean vermicelli of described user A, obtain the recommendation weighted value of described user A, according to the quantity of described other users' bean vermicelli, obtain described other users' recommendation weighted value;

Thereby the recommendation weighted value of described user A and described other users' recommendation weighted value is merged and obtains maximum recommendation weighted value, and the quantity of adding up the indirect follower between described user C and described user B, wherein, the quantity of the indirect follower between described user C and described user B is described user A and described other users' quantity sum;

Described user B is recommended to described user C, the recommendation weighted value of described maximum and all indirect followers are recommended to described user C simultaneously, to complete described user's two degree relationship among persons, recommend.

2. two of mass users as claimed in claim 1 spend relationship among persons MapReduce recommend methods, wherein, and in the process that the good friend who pays close attention to described user A by described user A recommendation in the bean vermicelli set of described user A gathers,

If there is described user B in the bean vermicelli set of described user A, in good friend's set of described user A, also there is described user B simultaneously, two of described user B degree human connections can not be described user B itself;

If the described user C of the bean vermicelli set of described user A has paid close attention to the described user B in the good friend set of described user A, two of described user C degree human connections can not be described user B.

3. two of mass users as claimed in claim 1 spend relationship among persons MapReduce recommend methods, wherein,

Described user A adopts the mode of LogNormal Function Fitting to obtain the recommendation weighted value of described user A according to the bean vermicelli quantity of described user A;

Described LogNormal Function Fitting formula is:

$f(x,\mathrm{\μ},\mathrm{\σ})=\frac{1}{\mathrm{x\σ}\sqrt{2\mathrm{\π}}}{e}^{-\frac{{\left(\ln x\right)}^2}{{2\mathrm{\σ}}^2}}$

Wherein, in described LogNormal Function Fitting, the array of the bean vermicelli quantity that X is all mass users, μ is the mean value of the logarithm of described array X, σ is the data expectation of described array X logarithm;

Mathematical expectation and the variance of described array X are respectively E[X] Var[X];

The fitting formula of parameter is:

$\mathrm{\μ}=\ln \left(E\right[X\left]\right)-0.5*\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

${\mathrm{\σ}}^2=\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

The recommendation weighted value that the directed integration numerical value of described LogNormal function is described user A.

4. two of mass users spend relationship among persons MapReduce commending systems,

Described commending system, for the concern relation in mass users, obtains user's two degree relationship among persons, and is recommended by first step MapReduce computing unit and second step MapReduce computing unit according to user's once relationship among persons; Wherein,

In the concern relation of described mass users, if user A pays close attention to user B, user C pays close attention to user A, and user B is the good friend of user A, and user C is the bean vermicelli of user A; And if user C recommends to pay close attention to user B by user A, user B is the two degree human connections of user C, user A is the indirect follower between user C and user B; If user C recommends to pay close attention to user B by other users except user A in the concern relation in described mass users, described other users are the indirect follower of described user C and described user B;

Described first step MapReduce computing unit, for the concern relation in described mass users, according to the once human connection of described user A, the bean vermicelli of the good friend of user A and user A is merged and sorted out, obtain good friend's set of user A and the bean vermicelli set of user A, the quantity of the bean vermicelli of while counting user A; Wherein,

Described user B is in good friend's set of described user A, and described user C is in the bean vermicelli set of described user A; And, in the concern relation of described mass users, according to described other users' once human connection, obtain good friend's set of described other users and described other users' bean vermicelli set, while described other users' of statistics bean vermicelli quantity;

Described second step MapReduce computing unit comprises recommends weighted value acquiring unit and two degree human connection recommendation unit; Wherein,

Described in Reduce, recommend weighted value acquiring unit, for obtaining user's recommendation weighted value; Wherein, in the concern relation of described mass users, the bean vermicelli set of described user A recommends the good friend who pays close attention to described user A to gather by described user A, two degree human connections of the bean vermicelli set of described user A are good friend's set of described user A, wherein, the indirect follower of the described user B that the described user C in the bean vermicelli set that described user A is described user A and the good friend of described user A concentrate; And,

According to the quantity of the bean vermicelli of described user A, obtain the recommendation weighted value of described user A, according to the quantity of described other users' bean vermicelli, obtain described other users' recommendation weighted value;

Described two degree human connection recommendation unit, recommend for two degree human connections of completing user; Wherein, thereby the recommendation weighted value of described user A and described other users' recommendation weighted value is merged and obtains maximum recommendation weighted value, and the quantity of adding up the indirect follower between described user C and described user B, wherein, the quantity of the indirect follower between described user C and described user B is described user A and described other users' quantity sum;

Described user B is recommended to described user C, the recommendation weighted value of described maximum and all indirect followers are recommended to described user C simultaneously, to complete described user's two degree human connections, recommend.

5. two of mass users as claimed in claim 4 spend relationship among persons MapReduce commending systems, wherein, and in the process that the good friend that described recommendation weighted value acquiring unit is paid close attention to described user A in the bean vermicelli set of described user A by described user A recommendation gathers,

If there is described user B in the bean vermicelli set of described user A, in good friend's set of described user A, also there is described user B simultaneously, two of described user B degree human connections can not be described user B itself;

If the described user C of the bean vermicelli set of described user A has paid close attention to the described user B in the good friend set of described user A, two of described user C degree human connections can not be described user B.

6. two of mass users as claimed in claim 4 spend relationship among persons MapReduce commending systems, wherein, and in described recommendation Weight Acquisition unit,

Described user A adopts the mode of LogNormal Function Fitting to obtain the recommendation weighted value of described user A according to the bean vermicelli quantity of described user A;

Described LogNormal Function Fitting formula is:

$f(x,\mathrm{\μ},\mathrm{\σ})=\frac{1}{\mathrm{x\σ}\sqrt{2\mathrm{\π}}}{e}^{-\frac{{\left(\ln x\right)}^2}{{2\mathrm{\σ}}^2}}$

Wherein, in described LogNormal Function Fitting, the array of the bean vermicelli quantity that X is all mass users note, μ is the mean value of the logarithm of described array X, σ is the data expectation of described array X logarithm.Mathematical expectation and the variance of described array X are respectively E[X] Var[X];

The fitting formula of parameter is:

$\mathrm{\μ}=\ln \left(E\right[X\left]\right)-0.5*\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

${\mathrm{\σ}}^2=\ln (1+\frac{\mathrm{Var}\left[X\right]}{{\left(E\right[X\left]\right)}^2})$

The recommendation weighted value that the directed integration numerical value of described LogNormal function is described user A.

Images