CN103870452A - Method and method for recommending data - Google Patents

Abstract

The invention relates to a method and system for recommending data, wherein the method comprises the steps of training weight factors of all users by taking the characteristics of all the users and the serial numbers of pieces of first to-be-recommended data as characteristic factors, obtaining the pieces of second to-be-recommended data from the pieces of first to-be-recommended data and obtaining predicted click rates of the pieces of second to-be-recommended data corresponding to a user requesting recommendation according to the weight factors, the characteristic of the user requesting recommendation and the serial numbers of the pieces of second to-be-recommended data, ranking the predicted click rates from high to low, obtaining first K pieces of the second to-be-recommended data highest in the predicted click rate and recommending the obtained first K pieces of the second to-be-recommended data to the user requesting recommendation, wherein K is a positive integer. The method and the system are capable of realizing personalized recommendation to a new user by thoroughly utilizing the characteristics of the user and the characteristic factors carried in the to-be-recommended data, and the characteristic factors are quite easy to extend, and therefore, new weight factors can be trained quickly according to new characteristic factors.

Description

Data recommendation method and system

Technical field

The present invention relates to a kind of data recommendation method and system.

Background technology

The personalized recommendation of data as the personalized recommendation of video website often according to user and data as the interbehavior of video, recommend the interested data of user as video to user.

But, for new user in the case of without any click, watch or mutual-action behavior, how these new users being carried out to personalized recommendation is a very large problem.For these new users, prior art is only from database, to choose hot data to recommend to it, but this recommend method is not realized the object of personalized recommendation.Therefore, need at present a kind of personalized recommendation method for new user and system badly.

Summary of the invention

The object of the present invention is to provide a kind of data recommendation method and system, feature and the entrained characterization factor of data to be recommended that can make full use of user carry out personalized recommendation to new user, and characterization factor ratio is easier to expansion, the weight factor that can make new advances according to new characterization factor Fast Training.

For addressing the above problem, the invention provides a kind of data recommendation method, comprising:

The weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor;

From the first data to be recommended, obtain the second data to be recommended, and obtain the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation according to user's the feature of described weight factor, request recommendation and the numbering of the second data to be recommended;

Described prediction clicking rate is sorted from big to small, obtain front K the user that the second data recommendation to be recommended is recommended to described request of prediction clicking rate maximum, wherein K is positive integer.

Further, in said method, the quantity of the quantity × data to be recommended of all users' of quantity=1+ of described weight factor feature.

Further, in said method, train all users' weight factor according to following formula:

$Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$

Wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

Further, in said method, obtain the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation according to following formula:

$P=\frac{1}{1+e^{-(W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})}}$

Wherein, P is prediction clicking rate, e=2.71828, W ₀and W _{m × N}represent described weight factor, F _{m × n}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

Further, in said method, user's feature comprises one or the combination in any in browser type, display resolution, types of network equipment, access websites time, location, user website incoming road and user's landing page.

Further, in said method, described the first data to be recommended are the high-quality data of regularly obtaining.

Further, in said method, to described high-quality data sorting, obtain front Q high-quality data as the second recommending data according to one or more users' feature, wherein Q is positive integer.

According to another side of the present invention, a kind of data recommendation system is provided, comprising:

Model module, for the weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor;

Recommended engine module, for obtaining the second data to be recommended from the first data to be recommended, and obtain the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation according to user's the feature of described weight factor, request recommendation and the numbering of the second data to be recommended, and described prediction clicking rate is sorted from big to small, front K the user that the second data recommendation to be recommended is recommended to described request who obtains prediction clicking rate maximum, wherein K is positive integer.

Further, in said system, train all users' weight factor according to following formula:

$Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$

Wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

Further, in said system, obtain the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation according to following formula:

$P=\frac{1}{1+e^{-(W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})}}$

Wherein, P is prediction clicking rate, e=2.71828, W ₀and W _{m × N}represent described weight factor, F _{m × n}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

Compared with prior art, the present invention is by the weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor, from the first data to be recommended, obtain the second data to be recommended, and according to described weight factor, user's the feature that request is recommended and the numbering of the second data to be recommended are obtained the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation, described prediction clicking rate is sorted from big to small, obtain front K the user that the second data recommendation to be recommended is recommended to described request of prediction clicking rate maximum, feature and the entrained characterization factor of data to be recommended that can make full use of user carry out new user to carry out personalized recommendation.

In addition, according to formula

train all users' weight factor, wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, and according to formula

obtain the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation, wherein, P is prediction clicking rate, e=2.71828, W ₀and W _{m × N}represent described weight factor, F _{m × N}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, makes characterization factor ratio be easier to expansion, the weight factor that can make new advances according to new characterization factor Fast Training.By above-mentioned intelligent recommendation scheme, can greatly improve the efficiency of user's fast searching content of interest, reduce the invalid browsing time of user, especially use in the environment that network traffics expense is higher at mobile device, save a large amount of communication flowss, for user saves campus network, improve user's reading satisfaction.

Accompanying drawing explanation

Fig. 1 is the process flow diagram of the data recommendation method of the embodiment of the present invention one;

Fig. 2 is the personalized recommendation result schematic diagram of the embodiment of the present invention one;

Fig. 3 is the data recommendation method flow diagram of the embodiment of the present invention two;

Fig. 4 is the module diagram of the data recommendation system of the embodiment of the present invention three;

Fig. 5 is the structural drawing of the data recommendation system of the embodiment of the present invention three.

Embodiment

For above-mentioned purpose of the present invention, feature and advantage can be become apparent more, below in conjunction with the drawings and specific embodiments, the present invention is further detailed explanation.

Embodiment mono-

As shown in Figure 1, the invention provides a kind of data recommendation method, comprising:

Step S11, the weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor.Wherein, described data can be videos, and the numbering of described the first data to be recommended can be video ID, have represented the difference hobby for the treatment of recommending data without user's feature.

Preferably, described user's feature can be obtained from a User action log, described user's feature comprises one or the combination in any in browser type, display resolution, types of network equipment, access websites time, location, user website incoming road (referer) and user's landing page (landing page), described user's feature comprises one or more characterization factors, and certain described user's feature also can comprise the data that user is demonstrated, the data that user clicked or watched.

Concrete, each user's feature specifically can comprise:

1) browser type can comprise: 360, IE7, IE6, IE8, Sogou, Chome, FireFox or other;

2) display resolution can comprise: 1440*900,1024*768,1280*800,800*680 or other;

3) types of network equipment can comprise: pc, movement or other;

4) the access websites time can comprise various hourages;

5) location can be accurate to province;

6) various user website incoming roads, user website incoming road (referrer) is that visitor enters any approach in website, HTTP referer is a part of header, in the time that browser sends request to web server, generally can bring referer to tell server user to connect and come from that page, server can obtain some information term processing by this

7) various user's landing page, user's landing page (Landing Page is sometimes called as and primarily catches user page) is the webpage that is shown to user after potential user clicks advertisement or utilizes search engine search.General this page can show and the advertisement of clicking or Search Results link relevant expansion content, and this page should be to do search engine optimization for certain key word (or phrase).

Preferably, the quantity N of the quantity M × data to be recommended of all users' of quantity=1+ of described weight factor feature.

Preferably, can train according to following formula all users' weight factor:

$Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$

Wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, can carry out LR modeling to user's characterization factor by this formula, and the weight generation factor.

The feature of supposing the user who only uses 1 types of network equipment comprises pc, mobile and other, and the numbering that has 2 data comprises video1 and video2 as video ID, and such one has 6 characterization factors comprises: F _{pc × video1}, F _{mobile × video1}, F _{other × video1}, F _{pc × viedo2}, F _{mobile × video2}and F _{other × video2}, need to train altogether 1+3*2=7 weight factor to comprise: W ₀, W _{pc × video1}, W _{mobile × video1}, W _{other × video1}, W _{pc × viedo2}, W _{mobile × video2}and W _{other × video2}.

So, formula $Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$ Be converted into:

Z=W ₀+W _pc×video1×F _pc×video1+W _{mobile×video1}×F _{mobile×video1}+W _{other×video1}×F _{other× video1}+W _pc×viedo2×F _pc×viedo2+W _{mobile×video2}×F _{mobile×video2}+W _{other×video2}×F _{other×video2}

If certain user sees to it and shown video Video1 by PC, but do not click Video1, Z is 0, owing to only having F _{pc × video1}be 1, remaining F _{mobile × video1}, F _{other × video1}, F _{pc × viedo2}, F _{mobile × video2}and F _{other × video2}be 0, above-mentioned formula is further converted to:

0=W ₀+W _pc×video1

Can train above-mentioned 7 weight factors according to above-mentioned same mode.

Step S12, from the first data to be recommended, obtain the second data to be recommended, and obtain the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation according to user's the feature of described weight factor, request recommendation and the numbering of the second data to be recommended.

Preferably, can obtain the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation according to following formula:

$P=\frac{1}{1+e^{-(W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})}}$

Wherein, P is prediction clicking rate, and e is natural logarithm, e=2.71828, and, W ₀and W _{m × N}represent described weight factor, F _{m × N}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is that the quantity of the first data to be recommended can be carried out ctr to the second data to be recommended by this formula and estimated to obtain prediction clicking rate.

Step S13, described prediction clicking rate is sorted from big to small, obtain front K the user that the second data recommendation to be recommended is recommended to described request of prediction clicking rate maximum, wherein K is positive integer, concrete, for without logging in, history information as click, watch or mutual-action behavior etc. without or little new user, in the time of its request personalized recommendation, still can recommend out data accurately, Fig. 2 is the personalized recommendation result that the present embodiment is shown to new user.

To sum up, the present embodiment can make characterization factor ratio be easier to expansion, the weight factor that can make new advances according to new characterization factor Fast Training, thus the feature and the entrained characterization factor of data to be recommended that make full use of user carry out personalized recommendation to new user.By above-mentioned intelligent recommendation scheme, can greatly improve the efficiency of user's fast searching content of interest, reduce the invalid browsing time of user, especially use in the environment that network traffics expense is higher at mobile device, save a large amount of communication flowss, for user saves campus network, improve user's reading satisfaction.

Embodiment bis-

As shown in Figure 2, the invention provides another kind of data recommendation method, the difference of the present embodiment and embodiment is that described the first data to be recommended are the high-quality data of regularly obtaining, in addition according to one or more users' feature to described high-quality data sorting, obtain front Q high-quality data as the second recommending data, thereby make recommendation results more accurate, described method comprises:

Step S21, the high-quality data of regularly obtaining are as the first data to be recommended; Concrete, described data can be video, for example can regularly from the whole network video library, obtain high-quality video and upgrade and store in a high-quality video storehouse, because the video of the whole network video library may have several necessarily even several hundred million, data volume is too large, train all users' the weight factor workload can be very large to all videos of the whole network video library, also there is no need, cross first to screen the video in the whole network video library and obtain high-quality video as the first data to be recommended, specifically can screen according to dimensions such as the broadcasting number of video, comment number, video qualities.

Step S22, the weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor.Wherein, described data can be videos, and the numbering of described the first data to be recommended can be video ID.

Preferably, described user's feature can be obtained from a User action log, described user's feature comprises one or the combination in any in browser type, display resolution, types of network equipment, access websites time, location, user website incoming road (referer) and user's landing page (landing page), described user's feature comprises one or more characterization factors, and certain described user's feature also can comprise the data that user is demonstrated, the data that user clicked or watched.

Concrete, each user's feature specifically can comprise:

1) browser type can comprise: 360, IE7, IE6, IE8, Sogou, Chome, FireFox or other;

2) display resolution can comprise: 1440*900,1024*768,1280*800,800*680 or other;

3) types of network equipment can comprise: pc, movement or other;

4) the access websites time can comprise various hourages;

5) location can be accurate to province;

6) various user website incoming roads, user website incoming road (referrer) is that visitor enters any approach in website, HTTP referer is a part of header, in the time that browser sends request to web server, generally can bring referer to tell server user to connect and come from that page, server can obtain some information term processing by this

7) various user's landing page, user's landing page (Landing Page is sometimes called as and primarily catches user page) is the webpage that is shown to user after potential user clicks advertisement or utilizes search engine search.General this page can show and the advertisement of clicking or Search Results link relevant expansion content, and this page should be to do search engine optimization for certain key word (or phrase).

Preferably, the quantity N of the quantity M × data to be recommended of all users' of quantity=1+ of described weight factor feature.

Preferably, can train according to following formula all users' weight factor:

$Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$

Wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, can carry out LR modeling to user's characterization factor by this formula, and the weight generation factor.

The feature of supposing the user who only uses 1 types of network equipment comprises pc, mobile and other, and the numbering that has 2 data comprises video1 and video2 as video ID, and such one has 6 characterization factors comprises: F _{pc × video1}, F _{mobile × video1}, F _{other × video1}, F _{pc × viedo2}, F _{mobile × video2}and F _{other × video2}, need to train altogether 1+3*2=7 weight factor to comprise: W ₀, W _{pc × video1}, W _{mobile × video1}, W _{other × video1}, W _{pc × viedo2}, W _{mobile × video2}and W _{other × video2}.

So, formula $Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$ Be converted into:

Z=W ₀+W _pc×video1×F _pc×video1+W _{mobile×video1}×F _{mobile×video1}+W _{other×video1}×F _{other× video1}+W _pc×viedo2×F _pc×viedo2+W _{mobile×video2}×F _{mobile×video2}+W _{other×video2}×F _{other×video2}

If certain user sees to it and shown video Video1 by PC, but do not click Video1, Z is 0, owing to only having F _{pc × video1}be 1, remaining F _{mobile × video1}, F _{other × video1}, F _{pc × viedo2}, F _{mobile × video2}and F _{other × video2}be 0, above-mentioned formula is further converted to:

0=W ₀+W _pc×video1

Can train above-mentioned 7 weight factors according to above-mentioned same mode.

Step S23, according to one or more users' feature to described high-quality data sorting, obtain front Q high-quality data as the second recommending data, wherein Q is positive integer, concrete, in the time that data are video, can pass through browser type, display resolution, types of network equipment, access websites time, location, user website incoming road (referer) and user's landing page (landing page) described high-quality video is sorted, obtain a front Q high-quality video as 1024 more the data of high-quality as the second recommending data.

Step S24, from the first data to be recommended, obtain the second data to be recommended, and obtain the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation according to user's the feature of described weight factor, request recommendation and the numbering of the second data to be recommended.

Preferably, can obtain the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation according to following formula:

$P=\frac{1}{1+e^{-(W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})}}$

Wherein, P is prediction clicking rate, and e is natural logarithm, e=2.71828, and, W ₀and W _{m × N}represent described weight factor, F _{m × N}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is that the quantity of the first data to be recommended can be carried out ctr to the second data to be recommended by this formula and estimated to obtain prediction clicking rate.

Step S25, described prediction clicking rate is sorted from big to small, obtain front K the user that the second data recommendation to be recommended is recommended to described request of prediction clicking rate maximum, wherein K is positive integer, concrete, for without logging in, history information as click, watch or mutual-action behavior etc. without or little new user, in the time of its request personalized recommendation, still can recommend out data accurately.

To sum up, the present embodiment can make characterization factor ratio be easier to expansion, the weight factor that can make new advances according to new characterization factor Fast Training, thereby the feature and the entrained characterization factor of data to be recommended that make full use of user carry out personalized recommendation to new user, and be the high-quality data of regularly obtaining by described the first data to be recommended, and according to one or more users' feature to described high-quality data sorting, obtain front Q high-quality data as the second recommending data, thereby make recommendation results more accurate.By above-mentioned intelligent recommendation scheme, can greatly improve the efficiency of user's fast searching content of interest, reduce the invalid browsing time of user, especially use in the environment that network traffics expense is higher at mobile device, save a large amount of communication flowss, for user saves campus network, improve user's reading satisfaction.

Embodiment tri-

As shown in Figure 4, the present invention also provides a kind of data recommendation system, comprises model module 1 and recommended engine module 2.Wherein, described data can be videos, and the numbering of described the first data to be recommended can be video ID.

The weight factor of model module 1 for the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor.

Preferably, the quantity of the quantity × data to be recommended of all users' of quantity=1+ of described weight factor feature.

Preferably, described user's feature can be obtained from a User action log, described user's feature comprises one or the combination in any in browser type, display resolution, types of network equipment, access websites time, location, user website incoming road (referer) and user's landing page (landing page), described user's feature comprises one or more characterization factors, and certain described user's feature also can comprise the data that user is demonstrated, the data that user clicked or watched.

Concrete, each user's feature specifically can comprise:

1) browser type can comprise: 360, IE7, IE6, IE8, Sogou, Chome, FireFox or other;

2) display resolution can comprise: 1440*900,1024*768,1280*800,800*680 or other;

3) types of network equipment can comprise: pc, movement or other;

4) the access websites time can comprise various hourages;

5) location can be accurate to province;

6) various user website incoming roads, user website incoming road (referrer) is that visitor enters any approach in website, HTTP referer is a part of header, in the time that browser sends request to web server, generally can bring referer to tell server user to connect and come from that page, server can obtain some information term processing by this

7) various user's landing page, user's landing page (Landing Page is sometimes called as and primarily catches user page) is the webpage that is shown to user after potential user clicks advertisement or utilizes search engine search.General this page can show and the advertisement of clicking or Search Results link relevant expansion content, and this page should be to do search engine optimization for certain key word (or phrase).

Preferably, can train according to following formula all users' weight factor:

$Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$

Wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, can carry out LR modeling to user's characterization factor by this formula, and the weight generation factor.

The feature of supposing the user who only uses 1 types of network equipment comprises pc, mobile and other, and the numbering that has 2 data comprises video1 and video2 as video ID, and such one has 6 characterization factors comprises: F _{pc × video1}, F _{mobile × video1}, F _{other × video1}, F _{pc × viedo2}, F _{mobile × video2}and F _{other × video2}, need to train altogether 1+3*2=7 weight factor to comprise: W ₀, W _{pc × video1}, W _{mobile × video1}, W _{other × video1}, W _{pc × viedo2}, W _{mobile × video2}and W _{other × video2}.Formula

be converted into:

Z=W ₀+W _pc×video1×F _pc×video1+W _{mobile×video1}×F _{mobile×video1}+W _{other×video1}×F _{other× video1}+W _pc×viedo2×F _pc×viedo2+W _{mobile×video2}×F _{mobile×video2}+W _{other×video2}×F _{other×video2}

If certain user sees to it and shown video Video1 by PC, but do not click Video1, Z is 0, owing to only having F _{pc × video1}be 1, remaining F _{mobile × video1}, F _{other × video1}, F _{pc × viedo2}, F _{mobile × video2}and F _{other × video2}be 0, above-mentioned formula is further converted to:

0=W ₀+W _pc×video1

Can train above-mentioned 7 weight factors according to above-mentioned same mode.

Preferably, described the first data to be recommended are the high-quality data of regularly obtaining.Concrete, described data can be video, for example can regularly from the whole network video library, obtain high-quality video and upgrade and store in a high-quality video storehouse, because the video of the whole network video library may have several necessarily even several hundred million, data volume is too large, train all users' the weight factor workload can be very large to all videos of the whole network video library, also there is no need, cross first to screen the video in the whole network video library and obtain high-quality video as the first data to be recommended, specifically can screen according to dimensions such as the broadcasting number of video, comment number, video qualities.

Recommended engine module 2 is for obtaining the second data to be recommended from the first data to be recommended, and obtain the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation according to user's the feature of described weight factor, request recommendation and the numbering of the second data to be recommended, and described prediction clicking rate is sorted from big to small, front K the user that the second data recommendation to be recommended is recommended to described request who obtains prediction clicking rate maximum, wherein K is positive integer.Concrete, in the time that data are video, can pass through browser type, display resolution, types of network equipment, access websites time, location, user website incoming road (referer) and user's landing page (landing page) sorts to described high-quality video, obtain a front Q high-quality video as 1024 more high-quality video as the second recommending data, for without logging in, history information as click, watch or mutual-action behavior etc. without or little new user, in the time of its request personalized recommendation, still can recommend out data accurately.

Preferably, can obtain the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation according to following formula:

$P=\frac{1}{1+e^{-(W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})}}$

Wherein, P is prediction clicking rate, and e is natural logarithm, e=2.71828, W ₀and W _{m × N}represent described weight factor, F _{m × N}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, can carry out ctr to the second data to be recommended estimate to obtain prediction clicking rate by this formula.

Preferably, can be according to one or more users' feature to described high-quality data sorting, obtain front Q high-quality data as the second recommending data, wherein Q is positive integer.

As shown in Figure 5, in the time that data are video, can from the whole network video library 3, filter out high-quality video by a VideoSelector module 4 and deposit a high-quality video storehouse 5 in, model module 1 is according to the feature of the user in user's row daily record 6 and the high-quality video weight generation factor 8, then recommended engine 2 sends to front end according to the partial video of the further screening in high-quality video storehouse 5 and weight factor 8 generating recommendations results, realizes to new user and carries out personalized video recommendation.

To sum up, the present embodiment can make characterization factor ratio be easier to expansion, the weight factor that can make new advances according to new characterization factor Fast Training, thereby the feature and the entrained characterization factor of data to be recommended that make full use of user carry out personalized recommendation to new user, and be the high-quality data of regularly obtaining by described the first data to be recommended, and according to one or more users' feature to described high-quality data sorting, obtain front Q high-quality data as the second recommending data, thereby make recommendation results more accurate.

Compared with prior art, the present invention is by the weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor, from the first data to be recommended, obtain the second data to be recommended, and according to described weight factor, user's the feature that request is recommended and the numbering of the second data to be recommended are obtained the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation, described prediction clicking rate is sorted from big to small, obtain front K the user that the second data recommendation to be recommended is recommended to described request of prediction clicking rate maximum, feature and the entrained characterization factor of data to be recommended that can make full use of user carry out new user to carry out personalized recommendation.

In addition, according to formula train all users' weight factor, wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, and according to formula

obtain the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation, wherein, P is prediction clicking rate, and e is natural logarithm, e=2.71828, W ₀and W _{m × N}represent described weight factor, F _{m × N}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended, makes characterization factor ratio be easier to expansion, the weight factor that can make new advances according to new characterization factor Fast Training.By above-mentioned intelligent recommendation scheme, can greatly improve the efficiency of user's fast searching content of interest, reduce the invalid browsing time of user, especially use in the environment that network traffics expense is higher at mobile device, save a large amount of communication flowss, for user saves campus network, improve user's reading satisfaction.

In this instructions, each embodiment adopts the mode of going forward one by one to describe, and what each embodiment stressed is and the difference of other embodiment, between each embodiment identical similar part mutually referring to.For the disclosed system of embodiment, owing to corresponding to the method disclosed in Example, so description is fairly simple, relevant part illustrates referring to method part.

Professional can also further recognize, unit and the algorithm steps of each example of describing in conjunction with embodiment disclosed herein, can realize with electronic hardware, computer software or the combination of the two, for the interchangeability of hardware and software is clearly described, composition and the step of each example described according to function in the above description in general manner.These functions are carried out with hardware or software mode actually, depend on application-specific and the design constraint of technical scheme.Professional and technical personnel can realize described function with distinct methods to each specifically should being used for, but this realization should not thought and exceeds scope of the present invention.

Obviously, those skilled in the art can carry out various changes and modification and not depart from the spirit and scope of the present invention invention.Like this, if within of the present invention these are revised and modification belongs to the scope of the claims in the present invention and equivalent technologies thereof, the present invention is also intended to including these changes and modification.

Claims (10)

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

The weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor;

From the first data to be recommended, obtain the second data to be recommended, and obtain the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation according to user's the feature of described weight factor, request recommendation and the numbering of the second data to be recommended;

Described prediction clicking rate is sorted from big to small, obtain front K the user that the second data recommendation to be recommended is recommended to described request of prediction clicking rate maximum, wherein K is positive integer.

2. data recommendation method as claimed in claim 1, is characterized in that, trains all users' weight factor according to following formula:

$Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$

Wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

3. data recommendation method as claimed in claim 2, is characterized in that, obtains the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation according to following formula:

$P=\frac{1}{1+e^{-(W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})}}$

Wherein, P is prediction clicking rate, e=2.71828, W ₀and W _{m × N}represent described weight factor, F _{m × n}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

4. data recommendation method as claimed in claim 1, is characterized in that, the quantity of the quantity × data to be recommended of all users' of quantity=1+ of described weight factor feature.

5. data recommendation method as claimed in claim 1, it is characterized in that, user's feature comprises one or the combination in any in browser type, display resolution, types of network equipment, access websites time, location, user website incoming road and user's landing page.

6. data recommendation method as claimed in claim 1, is characterized in that, described the first data to be recommended are the high-quality data of regularly obtaining.

7. data recommendation method as claimed in claim 6, is characterized in that, according to one or more users' feature, described high-quality data is sorted, and obtains front Q high-quality data as the second recommending data, and wherein Q is positive integer.

8. a data recommendation system, is characterized in that, comprising:

Model module, for the weight factor that the numbering of all users' feature and the first data to be recommended is trained to all users as characterization factor;

Recommended engine module, for obtaining the second data to be recommended from the first data to be recommended, and obtain the prediction clicking rate of described the second data to be recommended with respect to the user of request recommendation according to user's the feature of described weight factor, request recommendation and the numbering of the second data to be recommended, and described prediction clicking rate is sorted from big to small, front K the user that the second data recommendation to be recommended is recommended to described request who obtains prediction clicking rate maximum, wherein K is positive integer.

9. data recommendation system as claimed in claim 8, is characterized in that, trains all users' weight factor according to following formula:

$Z=W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})$

Wherein, the click situation of Z representative of consumer to a certain the first data to be recommended, when click, Z is 1, while click, Z is 0, W ₀and W _{m × N}represent described weight factor, F _{m × N}the representative feature factor is the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

10. data recommendation system as claimed in claim 8, is characterized in that, obtains the prediction clicking rate of the second data to be recommended with respect to the user of request recommendation according to following formula:

$P=\frac{1}{1+e^{-(W_0+\underset{M\×N}{\mathrm{\Σ}}(W_{M\×N}\×F_{M\×N})}}$

Wherein, P is prediction clicking rate, e=2.71828, W ₀and W _{m × N}represent described weight factor, F _{m × n}represent the displaying situation of a certain the first data to be recommended under a certain user's feature, F when displaying _{m × N}be 1, F while displaying _{m × N}be that 0, M is the quantity of user's feature, N is the quantity of the first data to be recommended.

Images