CN102999540A - Method and system for determining user features on Internet - Google Patents
Method and system for determining user features on Internet Download PDFInfo
- Publication number
- CN102999540A CN102999540A CN2011102809074A CN201110280907A CN102999540A CN 102999540 A CN102999540 A CN 102999540A CN 2011102809074 A CN2011102809074 A CN 2011102809074A CN 201110280907 A CN201110280907 A CN 201110280907A CN 102999540 A CN102999540 A CN 102999540A
- Authority
- CN
- China
- Prior art keywords
- user
- document
- vector
- characteristics vector
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Abstract
The invention provides a method and a system for determining user features on the Internet. According to the method, personalized features of documents and users are automatically updated according to signals for visiting the documents by the users and signals for contacting other users by the users. When the users visit the documents, the personalized features of the users are updated according to the personalized features of the documents, and the personalized features of the documents are updated by the personalized features of the users. When the users contact other users, the personalized features of the users are updated by the personalized features of other users. The personalized features of a plurality of users and documents are obtained through using the method for many times. According to the personalized features of the users and the documents, the webpage information obtained by a search engine is further filtered and screened for improving the search accuracy rate; and user groups with the specific features can be found in the Internet according to the personalized features of the users.
Description
Technical field
The present invention relates to internet arena, relate in particular to a kind of method and system of determining on the internet user characteristics.
Background technology
On the internet, search engine and social networks are to use the more instrument that obtains the network information.But all there are some shortcomings in these two kinds of instruments.
For search engine, when user entered keyword carried out Webpage search, the web page interlinkage quantity that search engine returns often reached several ten thousand even several ten million, and this brings very large puzzlement for the user search target web.The Search Results that has statistical study to show that the user on average checks is no more than two pages, and most Search Results users that this explanation search engine obtains do not see.Even the part that the user sees, Search Results also usually comprise the lower or basic incoherent webpage of a lot of degrees of correlation.
For social networks, the information of social networks issue every day has reached several hundred million.Although the user can filter and filter information by adding personal relationship's network in social networks, for example obtain other people information or share information that friend " likes (like) " etc. by plusing good friend by adding " pay close attention to (follow) ", but in the social networks such as microblogging and face book (Facebook), the information overload phenomenon occurred at present.Owing to worry to have important or interesting information to omit, the user can add too much relational network usually in social networks, for example pay close attention to more people or add more good friend etc.This is with regard to so that social networks becomes a kind of service of the user being carried out " INFORMATION BOMB " gradually.U.S. San Jose state university studies show that, the result of information overload is the proportion that has strengthened scanning input and skimmed, and it is to browse and scanning that 82% interviewee represents mostly, and 85% reader represents to carry out more " non-linear reading ".
A common ground of the problems referred to above is effects that the individualized feature of not considering the user plays in information filtering and screening.For example, the Search Results that obtains when different users uses identical key word to carry out Webpage search in search engine is identical, and is irrelevant with the user's who submits search inquiry to individualized feature.And in social networks, the information that each user obtains is only relevant with its relational network, and irrelevant with user's self individualized feature.The information that the user obtains is all information of each user's issue in its relational network, and the user can not receive these information selectively.For example you are as long as pay close attention to a people, even you are only interested in his certain category information of issue, you also have to receive the full detail from this people.
A thinking that addresses the above problem is to obtain user's customized information in computer system, and according to user's customized information the network information of obtaining is effectively filtered and screened, and reduces the user to the scanning of invalid information and browses.
But in obtaining the process of user personalized information, there is following problem:
The firstth, the accuracy problem of customized information.The user is unwilling to provide on the internet accurately customized information usually.Although certain customers provide the personal information such as age of user, education degree, previous graduate college, geographic position, professional domain and preference in the social networks such as face book (Facebook), but considerable user is to providing personal information to suspect, a lot of users use false personal information at social networks, so that the personal information that system obtains is not accurate enough.The secondth, the comprehensive problem of customized information.The user normally is difficult to express all sidedly its individualized feature, for example in the social networks, the common description in consumer taste one hurdle is to like Mozart, play baseball or reading etc., and these Partial Feature of a representative of consumer often are difficult and require each user to fill in all sidedly its individualized feature.The 3rd is the structuring expression problem of customized information.Different but the semantic identical feature of character express, be difficult on the internet they are carried out the structuring classification, fill in such as the user who has and to like Mozart, filling in of having to like classical music, possible these two users' hobby is identical, but because therefore the difference of literal expression is difficult to they are effectively sorted out.The 4th is the replacement problem of customized information.As time goes on, user's personal information and hobby may change, but it is difficult requiring all users dynamically to upgrade these information.
Obtain the user individual feature many useful application are arranged.For example in conjunction with user's individualized feature and webpage individualized feature, can the webpage that search be filtered and screen.Utilize user's individualized feature, can realize user's cluster analysis to determine to have the customer group of special characteristic, comprise on the internet the expert that seeks individual with same interest hobby and group, searching and have a certain ability, seek that customer group with special characteristic is sought by the businessman that sells certain product and businessman in order to directedly throw in advertisement etc.
In sum, how to obtain user's individualized feature, and in " noise " of magnanimity, filter out Useful Information according to these individualized features, and according to user's individualized feature suitable information is sent to suitable people in the suitable time, be the problem that current internet needs to be resolved hurrily.
Summary of the invention
Problem in view of above-mentioned prior art existence, the object of the present invention is to provide a kind of on the internet method and system of definite user characteristics to come the individualized feature of automatic acquisition user and document, and accordingly the information content that obtains in search engine and social networks is effectively filtered and screened.
Another object of the present invention is to provide a kind of and determines that on the internet the method and system of user characteristics comes automatic acquisition user's individualized feature, and seeks the customer group with given feature according to the user individual feature in social networks.
According to above-described purpose, the present invention proposes a kind of method of determining on the internet user characteristics, it is characterized in that, storage document sets I={1 in server, 2 ..., M}, user collect J={1, and 2 ..., N} and feature set K={1,2 ..., L}, wherein M is the document number, N is user's number, L is Characteristic Number, and carries out following steps:
Receive the signal of user j (j ∈ J) access document i (i ∈ I), described signal comprises the user ID of described user j and the document identification of described document i at least;
According to described document identification, read the file characteristics vector K of described document i
d(i)=(dw
I1, dw
I2..., dw
Ik..., dw
IL), dw wherein
IkThe degree of correlation that represents described document i and feature k (k ∈ K);
According to described user ID, read the user characteristics vector K of described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL), uw wherein
JkThe degree of correlation that represents described user j and feature k (k ∈ K);
Use the user characteristics vector that following algorithm upgrades the described user j of file characteristics vector sum of described document i:
K
d *(i)＝function1[K
d(i)，K
u(j)]
K
u *(j)＝function2[K
d(i)，K
u(j)]
Wherein, described function1[K
d(i), K
uAnd described function2[K (j)]
d(i), K
u(j)] be increasing function, K
d(i) and K
d *(i) represent respectively to upgrade front file characteristics vector with upgrading rear described document i, K
u(j) and K
u *(j) represent respectively to upgrade front user characteristics vector with upgrading rear described user j.
According to the above purpose, the present invention proposes a kind of method of determining on the internet user characteristics, it is characterized in that, the storage user collects J={1 in server, 2 ..., N} and feature set K={1,2, ..., L}, wherein N is user's number, L is Characteristic Number, and carries out following steps in described server:
Receive the signal of user j (j ∈ J) contact user i (i ∈ J), described signal comprises the user ID of described user j and the user ID of described user i at least;
According to the user ID of described user j, read the user characteristics vector K of described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL), uw wherein
JkThe degree of correlation that represents described user j and feature k (k ∈ K);
According to the user ID of described user i, read the user characteristics vector K of described user i
u(i)=(uw
I1, uw
I2..., uw
Ik..., uw
IL), uw wherein
IkThe degree of correlation that represents described user i and feature k (k ∈ K);
Use following algorithm that the user characteristics vector of described user j is upgraded:
K
u *(j)＝function3[K
u(i)，K
u(j)]
Wherein, described function3[K
u(i), K
u(j)] be increasing function, K
u(j) and K
u *(j) be respectively the user characteristics vector that upgrades front and the rear described user j of renewal, K
u(i) be the user characteristics vector of described user i.
Compared with prior art, the inventive method by the user on the internet the signal of access document and user at other users' of social networks contact signal, automatically obtain the individualized feature of user and document, and according to the individualized feature of user and document the network information that the user obtains is effectively filtered and screened, and seek according to the user individual feature and to have the customer group of special characteristic.
Description of drawings
Fig. 1 is the method for expressing of user characteristics vector;
Fig. 2 is the method for expressing of file characteristics vector;
Fig. 3 is the method for expressing of bean vermicelli proper vector;
Fig. 4 is the method for expressing of characteristic of advertisement vector;
Fig. 5 is a kind of method flow diagram of determining user and file characteristics in the internet;
Fig. 6 is a kind of method flow diagram of determining user characteristics in the internet;
Fig. 7 is a kind of system construction drawing of determining user characteristics in the internet.
Embodiment
By reference to the accompanying drawings the inventive method is described in further detail.
Specific embodiments explanation to this patent method comprises following components, illustrate at first that user characteristics vector, file characteristics are vectorial, method for expressing and the initial value method to set up thereof of bean vermicelli proper vector and characteristic of advertisement vector, then update method based on the user characteristics vector sum file characteristics vector of user's access document signal is described, the update method of getting in touch with the user characteristics vector sum bean vermicelli proper vector of other subscriber signals based on the user is described afterwards, provides at last a kind of system that determines on the internet user characteristics.
Fig. 1 is the method for expressing of user characteristics vector.The method for expressing of user characteristics vector is similar to the vectorial expression method of the vector space model that GerardSalton proposes, and namely with the base unit of characteristic item as user characteristics, comes a user's of approximate representation feature with the set of characteristic item.Described user characteristics vector is decided by user characteristics and the feature degree of correlation.User characteristics comprises user's physical feature and user preference feature, wherein user's physical feature comprises age, sex, occupation, educational background, height, body weight and geographic position etc., user's individual preference feature comprises the abstract characteristics such as field that the user pays close attention to, such as science, music, military affairs and physical culture etc.The feature degree of correlation is with the close relation degree between numeric representation user and the feature.If a user is concerned about that music is more, be concerned about that physical culture is a little less, we just say that the degree of correlation of this user and musical features is high, and are low with the degree of correlation of sports feature.
Before the method for expressing of introducing the user characteristics vector, paper Customs Assigned Number method and user characteristics represent method.In the internet, can represent a user by following sign, comprise account number, phone number, IP address, Email address and instant communication number that the user applies in the website etc.For the ease of statement, we carry out Unified number to each user on the internet, and collect J={1 with the user, and 2 ..., N} represents all of user, wherein N represents user's number.Described user collects that each user has at least one feature among the J, and we also carry out Unified number to the feature that the user collects all users among the J, forms user's feature set K={1, and 2 ..., L}, the wherein number of L representation feature.If stored the information of N user take user ID as index in server, we just say that having stored the user in server collects J={1,2 ..., N}.
Collect described user that each user is provided with the user characteristics vector among the J.The method for expressing of the user characteristics vector of user j (j ∈ J) is K
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL), uw wherein
JkThe degree of correlation that represents k the feature of described user j and feature set K.Uw
JkDegree of correlation between numerical value larger expression user j and the feature k is higher, if uw
JkBe negative, then represent user j and feature k negative correlation.
Because described feature set K has comprised all users' feature, therefore its dimension is normally huge, and the feature that each user has is a part very little among the feature set K, so most vector fractional integration series numberical value of quantities all are zero or very little numerical value in the user characteristics vector, this has caused the generation of the Sparse phenomenon of user characteristics vector.Solution is that the form of described user characteristics vector with a kind of simplification represented, soon user characteristics vector representation be [..., (k, uw
Jk) ...].Feature set K={ news for example, science and technology, finance and economics, physical culture, amusement, life, tourism, culture, education ... }, if certain user's user characteristics vectorial=[(finance and economics, 2.4); (education, 6.7)], the degree of correlation that " finance and economics " among this user and the feature set K is described is 2.4, with the degree of correlation of " education " among the feature set K be 6.7.
In actual applications usually the reduced form of user's proper vector [..., (k, uw
Jk) ...], can save like this storage space and reduce computing cost.But for the purpose of the formalization narration was convenient, the expression form of the user characteristics vector of described user j still used (uw in the following description
J1, uw
J2..., uw
Jk..., uw
JL).The reduced form that need to prove user's proper vector does not affect described method essence.
Fig. 2 is the expression method of file characteristics vector.The method for expressing of file characteristics vector is similar to the method for expressing of user characteristics vector, namely with the base unit of characteristic item as file characteristics, comes the feature of a document of approximate representation with the set of characteristic item.The file characteristics vector is to be decided by the feature of document and the feature degree of correlation.The feature of document can be science, music, military affairs and physical culture etc.The feature degree of correlation is to represent close relation degree between document and the individual features with numerical value, if for example the relation of document and social concern is more a little less with relation military issue, we just say that the degree of correlation of the document and social characteristic is high, and are low with the degree of correlation of military feature.
Before the method for expressing of introducing the file characteristics vector, paper document code method and file characteristics method for expressing.Have large volume document in the internet, its content comprises the content of webpage, microblogging and advertisement etc., and its form of expression comprises text, video, music and picture etc.These documents have unique network address URL usually.We carry out Unified number on the internet each document for convenience of explanation, and use document sets I={1, and 2 ..., M} represents.If stored the information of M document take document identification as index in server, we have stored document sets I={1 with regard to saying in server, 2 ..., M}.
The feature of each document can obtain by traditional feature extracting method among the described document sets I, for example document word frequency (DF), information gain (IG), mutual information (MI) and χ
2Statistic law (CHI) etc. also can produce by the mode of artificial setting the feature of document.We screen the feature of all documents among the document sets I, find out representational feature and feature is numbered, and form the feature set K={1 of document, and 2 ..., L}.Of particular note in this article, the feature set of document is identical with user's feature set, and namely feature set K had both represented the user characteristics collection, also represented the file characteristics collection.Therefore, when generating feature collection K, should consider user's feature, also will consider the feature of document.Same feature, feature " computing machine " for example is for user's expression user preference " computing machine ", for document and bright this document of putting into words is relevant with " computing machine ".In addition, has correlativity between some feature, for example physics and relativity, therefore when feature selecting, can reduce by the correlativity between the minimizing feature dimension of feature set K, improve efficiency of algorithm, also can study two correlativitys of setting between the feature by this patent method.Some feature needn't directly be listed in the feature set, because the weighted average calculation of the feature degree of correlation that the degree of correlation of these features can be by one or several feature among the feature set K out.Be conducive to like this reduce the dimension of feature set K.In addition, the feature set of described document also can be different with described user's feature set, but need to set up two mapping relations between the feature set.
Each document is provided with the file characteristics vector in described document sets I.The method for expressing of the file characteristics vector of document i (i ∈ I) is K
d(i)=(dw
I1, dw
I2..., dw
Ik..., dw
IL), dw wherein
IkThe degree of correlation that represents k the feature of described document i and feature set K.Described degree of correlation dw
IkNumerical value is larger, and the degree of correlation between expression document i and the feature k is higher, if dw
IkBe negative, then represent document i and feature k negative correlation.Similar to the user characteristics vector, the file characteristics vector also can adopt simplification expression form [..., (k, dw
Ik) ...] solve the Sparse Problem of file characteristics vector.
The method to set up of user characteristics vector initial value is as follows.Take the method to set up of user j as example.
Example 1 is the method that user characteristics vector initial value manually is set.At first determine the principal character of user j, then at least one feature that namely artificial selection user j has in feature set K collects user j and user that other user compares among the J, manually the degree of correlation of each selected feature of definite described user j.For example certain user's user characteristics vector is [(finance and economics, 2.4), (education, 6.7)].The artificial degree of correlation uw that sets
JkThe span of (j ∈ J, k ∈ K) is [a, b], and wherein a and b are setup parameter.
Example 2 is methods that user characteristics vector initial value is set according to the feature that one group of user that the user submits to has.If user j has selected one group of user U
j=..., r ... }, user r (r ∈ U wherein
j) the user characteristics vector be K
u(r)=(uw
R1, uw
R2..., uw
RL), then the initial value of the user characteristics vector of user j is made as:
Uw
Jk=(σ 1/s) ∑
(r ∈ Uj)[uw
Rk/ (∑
(k ∈ UKr)Uw
Rk)], for each k ∈ K
Wherein s is described U
jElement number, σ
1For setting normal number.Described UK
rThe user characteristics vector K by described user r
u(r)=(uw
R1, uw
R2..., uw
RL) in the P of numerical value maximum
rThe set that the corresponding feature of (r ∈ J) individual component forms, P
rBe setup parameter.Regulation is worked as in formula
The time, get uw
Rk=0.
Example 3 is methods that user characteristics vector initial value is set according to the feature that one group of document that the user submits to has.If user j has selected one group of document D
j=..., r ... }, document r (r ∈ D wherein
j) the file characteristics vector be K
d(r)=(dw
R1, dw
R2..., dw
RL), then the initial value of the user characteristics vector of user j is:
Uw
Jk=(σ
2/ s) ∑
(r ∈ Dj)[dw
Rk/ (∑
(k ∈ DKr)Dw
Rk)], for each k ∈ K
Wherein s is described D
jElement number, σ
2For setting normal number.Described DK
rThe file characteristics vector K by described document i
d(r)=(dw
R1, dw
R2..., dw
RL) in the Q of numerical value maximum
rThe set that the corresponding feature of (r ∈ I) individual component forms, Q
rBe setup parameter.Regulation is worked as in formula
The time, get dw
Rk=0.
Because the importance of user characteristics vector initial value can select certain customers to have the right with said method the vectorial initial value of own user characteristics to be set.
A method to set up of file characteristics vector initial value is as follows.
Take the method to set up of the file characteristics vector of document i (i ∈ I) as example.At first determine the principal character of document i, then at least one feature that namely artificial selection document i has in feature set K compares other document among document i and the document sets I, manually determines the degree of correlation of each manually selected feature of described document i.For example the file characteristics of certain document vector is [(science and technology, 8.4), (education, 3.2)].The artificial degree of correlation dw that sets
IkThe span of (i ∈ I) is [a, b], and wherein a and b are for setting constant.
Fig. 3 is the method for expressing of bean vermicelli proper vector.So-called bean vermicelli (Fans) be pay close attention to described user or with described user other users in same interest cohort.The feature that a user's bean vermicelli colony shows on the whole is exactly the feature of bean vermicelli colony.Collect J={1 the user, 2 ..., each user is provided with the bean vermicelli proper vector among the N}.The method for expressing of the bean vermicelli proper vector of user j (j ∈ J) is K
f(j)=(fw
J1, fw
J2..., fw
Jk..., fw
JL), fw wherein
JkThe degree of correlation that represents k the feature of the bean vermicelli colony of described user j and feature set K.Described degree of correlation fw
JkNumerical value is larger, and bean vermicelli colony and the degree of correlation between the feature k of expression user j are higher, if fw
JkThe bean vermicelli colony and the feature k negative correlation that then represent user j for negative.Similar with the user characteristics vector, the bean vermicelli proper vector also can use simplification expression form [..., (k, fw
Jk) ...] solve bean vermicelli characteristic vector data Sparse Problems.The default initial value of bean vermicelli proper vector is null vector.
Fig. 4 is the method for expressing of characteristic of advertisement vector.The method for expressing of characteristic of advertisement vector is similar to the method for expressing of user characteristics vector, namely with the base unit of characteristic item as characteristic of advertisement, comes the feature of an advertisement of approximate representation with the set of characteristic item.We carry out Unified number to the advertisement of storing in the system and obtain set of advertisements A={1, and 2 ..., G}, the characteristic of advertisement vector K of advertisement g (g ∈ A)
a(g)=(aw
G1, aw
G2..., aw
Gk..., aw
GL), aw wherein
GkRepresent the degree of correlation of described advertisement g and feature k (k ∈ K), G is the advertisement number.The reduced representation form of characteristic of advertisement vector be [..., (k, aw
Gk) ...].The initial value of characteristic of advertisement vector can manually be set, and for example the characteristic of advertisement of certain advertisement vector is [(food, 4.6), (student, 3.2)], illustrate that this advertisement is relevant with food, target customers are students, with the degree of correlation of food be 4.6, with student's the degree of correlation be 3.2.The default initial value of characteristic of advertisement vector is null vector.
Fig. 5 is a kind of method flow diagram of determining user and file characteristics in the internet.Signal that this method is based on user's access document upgrades the file characteristics vector of described user's the described document of user characteristics vector sum.Described method comprises following concrete steps:
S10. be document sets I={1,2 ..., a part of document setup file characteristics vector initial value among the M}, the default initial value of file characteristics vector is null vector; For the user collects J={1,2 ..., a part of user arranges user characteristics vector initial value among the N}, and the default initial value of user characteristics vector is null vector;
S11. receive the signal of user j (j ∈ J) access document i (i ∈ I), described signal comprises the user ID of described user j and the document identification of described document i at least;
S12. according to described document identification, read the file characteristics vector K of described document i
d(i)=(dw
I1, dw
I2..., dw
Ik..., dw
IL), dw wherein
IkThe degree of correlation that represents described document i and feature k (k ∈ K);
S13. according to described user ID, read the user characteristics vector K of described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL), uw wherein
JkThe degree of correlation that represents described user j and feature k (k ∈ K);
S14. upgrade the user characteristics vector of the described user j of file characteristics vector sum of described document i, update algorithm is as follows:
K
d *(i)＝function1[K
d(i)，K
u(j)]
K
u *(j)＝function2[K
d(i)，K
u(j)]
Wherein, described function1[K
d(i), K
uAnd function2[K (j)]
d(i), K
u(j)] be increasing function, K
d(i) and K
d *(i) represent respectively to upgrade front file characteristics vector with upgrading rear described document i, K
u(j) and K
u *(j) represent respectively to upgrade front user characteristics vector with upgrading rear described user j.Need K after stating in the use algorithm
d(i) and K
u(j) upgrade, i.e. K
d(i)=K
d *(i), K
u(j)=K
u *(j).
Described access in the described method of Fig. 5 comprises a kind of in the following situation at least: the user browse webpage, a user click advertisement, a user create microblogging, a user transmit microblogging, a user collect a microblogging, user to microblogging comment on, the user is made as document and likes (like) etc.For example the user has clicked a webpage we has just said that the user has accessed this webpage, if user j has transmitted document i (for example one piece of microblogging), we just say that user j has accessed document i.In addition, described signal both can be the signal that the described user j of Real-time Obtaining accesses described document i, also can be the signal that the described user j that afterwards obtains by the analytic system daily record accesses described user i.
Application example 1.
Described algorithm K
d *(i)=function1[K
d(i), K
uAnd K (j)]
u *(j)=function2[K
d(i), K
u(j)] specific algorithm is as follows, namely
Dw
Ik *=dw
Ik+ λ
1(t) ζ
1(jk) f
1(uw
Jk), for each k ∈ UK
j,
Uw
Jk *=uw
Jk+ λ
2(t) ζ
2(i) f
2(dw
Ik), for each k ∈ DK
i,
In the described specific algorithm, t is the type of described access, it is the mode that described user j and described document i set up contact, for example t=11 is illustrated in the j of user described in the microblogging and has transmitted described document i, t=12 is illustrated in the j of user described in the microblogging and has collected described document i, t=21 represents that described user j has clicked " liking (the like) " button on the described document i, and t=31 represents that described user j has browsed described document i (webpage) etc.λ
1(t) and λ
2(t) be the function of t, for example λ
1(11)=1.2, λ
1(12)=1.8, λ
2(21)=1.6.
In the described specific algorithm, ζ
1(jk) the described user j of expression collects with described user that other users compare the significance level of its k feature among the J.ζ
1(jk) can be made as k component fw of bean vermicelli proper vector of user j
JkIncreasing function, for example establish ζ
1(jk)=c
1Fw
Jk/ (∑
(k ∈ K)Fw
Jk), c wherein
1For setting constant.ζ
1(jk) also can be an artificial parameter that arranges, for example any j ∈ J and k ∈ K be arranged ζ
1(jk)=1.ζ
2(i) be the significance level parameter that described document i compares with other document among the described document sets I.ζ for example
2(i) can be made as the linear combination of one group of parameter, described parameter comprise at least described document i the PageRank value, like (like) number of times, transmit this number, comment number of times and collection number of times.ζ
2(i) also can be an artificial parameter that arranges, for example any i ∈ I be arranged ζ
2(i)=1.
In the described specific algorithm, described DK
iThe file characteristics vector K by described document i
d(i)=(dw
I1, dw
I2..., dw
Ik..., dw
IL) in the Q of numerical value maximum
iThe set that the corresponding feature of (i ∈ I) individual component forms, described UK
jThe user characteristics vector K by described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL) in the P of numerical value maximum
jThe set that the corresponding feature of (j ∈ J) individual component forms, Q
iAnd P
jBe setup parameter.For example, i=30, Q
30=3, DK
30={ science, computing machine, DNA}; J=265, P
265=2, UK
265={ science, biology }.
In the described specific algorithm, f
1(uw
Jk) and f
2(dw
Ik) being increasing function, their typical case uses as follows.Example 1 is for establishing f
1(uw
Jk)=σ
3Uw
Jk/ (∑
(k ∈ UKj)Uw
Jk), f
2(dw
Ik)=σ
4Dw
Ik/ (∑
(k ∈ DKi)Dw
Ik).Example 2 is for establishing f
1(uw
Jk)=σ
3Uw
Jk, f
2(dw
Ik)=σ
4Dw
Ikσ wherein
3And σ
4For setting normal number.
In the described specific algorithm, dw
IkAnd dw
Ik *Represent respectively to upgrade the file characteristics of front and the rear described document i of renewal to a flow control k component, uw
JkAnd uw
Jk *Represent respectively to upgrade the user characteristics of front and the rear described user j of renewal to a flow control k component.After described specific algorithm is complete, carry out following assignment, namely for each k ∈ UK
jDw is arranged
Ik=dw
Ik *, for each k ∈ DK
iUw is arranged
Jk=uw
Jk *
After application drawing 5 described algorithms repeatedly, user characteristics component of a vector and file characteristics vector fractional integration series numberical value of quantity can increase gradually, surpass at last the memory capacity of storer.Solution is to adopt the correction algorithm of following user characteristics vector sum file characteristics vector:
For user characteristics vector, select at least one k ∈ K and each k that selects be handled as follows: for each j (j ∈ J) with uw
JkBe mapped as g
1(uw
Jk), then for each j (j ∈ J) uw is set
Jk=g
1(uw
Jk), g wherein
1(uw
Jk) be increasing function, g
1(uw
Jk) ∈ [a, b], a and b are the setting constant.
For file characteristics vector, select at least one k ∈ K and each k that selects be handled as follows: for each i (i ∈ I) with dw
IkBe mapped as g
2(dw
Ik), then each i (i ∈ I) is arranged dw
Ik=g
2(dw
Ik), g wherein
2(dw
Ik) be increasing function, g
2(dw
Ik) ∈ [a, b], a and b are the setting constant.
The below illustrates the correction algorithm of user characteristics vector with instantiation.
1. couples of data uw of example
1k, uw
2k..., uw
NkSort, obtain each uw
JkRank Rank (uw
Jk), and regulation Rank (max
jUw
Jk)=1, therefore described correction algorithm is for each j ∈ J:
g
1(uw
Jk)=a+ (b-a) [N-Rank (uw
Jk)+1]/N and uw
Jk=g
1(uw
Jk).
Wherein, N is user's number.Determine at random rank if rank is identical, for example two user's ranks are 87 all, and first user's rank of randomly drawing so is 87, and another rank is 88.
Example 2. is established MAX=max
jUw
JkBe data uw
1k, uw
2k..., uw
NkThe mean value of some (such as front 10) of middle numerical value maximum, therefore described correction algorithm is g
1(uw
Jk)=a+ (b-a) uw
Jk/ MAX.If uw
Jk＞MAX then gets g
1(uw
Jk)=b arranges uw for each j ∈ J at last
Jk=g
1(uw
Jk).
Example 3. is at data uw
1k, uw
2k..., uw
NkIn randomly draw R data { s
1, s
2..., s
R, s wherein is set
1=min
jUw
Jk, s
R=max
jUw
JkIf, for each j ∈ J s
m≤ uw
Jk≤ s
M+1, g then is set
1(uw
Jk)=s
mAnd uw
Jk=g
1(uw
Jk).
The example of file characteristics vector corrected algorithm is identical with the example principle of described user characteristics vector corrected algorithm, repeats no more.
In the correction algorithm of described user characteristics vector sum file characteristics vector, the k that selects will travel through each feature among the described feature set K, consider the restriction of system-computed processing power, can use above-mentioned correction algorithm to the different characteristic k among the feature set K at times.
Application example 2.
Application example 2 is application of Fig. 5 method, i.e. the application of one group of particular document of inquiry in the internet.Its step is as follows:
The described user of file characteristics vector sum who repeatedly uses described method to obtain a plurality of documents among the described document sets I collects the user characteristics vector of a plurality of users among the J, for example by receiving a plurality of users to the request of access of a plurality of different document, upgrade corresponding user characteristics vector sum file characteristics vector;
Receive the querying condition that user m (m ∈ J) submits to, described querying condition comprises at least one feature that is queried document;
According to described querying condition, generate the query feature vector of described user m;
Calculate the file characteristics vector of each document among the described document sets I and the mathematical distance between the described query feature vector and according to described mathematical distance described document sets I is sorted;
According to described ranking results the sign of partial document among the described document sets I is sent to described user m.
In described application example 2, the query feature vector that described user m is set usually is K
s(m)=(sw
M1, sw
M2..., sw
Mk..., sw
ML), sw wherein
MkRepresent the described degree of correlation that is queried k the feature of document and feature set K, sw
Jk∈ [a, b], a and b are for setting constant.
Described query feature vector has following four kinds of generation methods.The firstth, by the numerical value that described user m oneself arranges each component of query feature vector, for example establish sw
J2=2.3, sw
J6=6.1, other each component is 0.The secondth, give described query feature vector the user characteristics of described user m vector assignment.The 3rd is that described user m submits one group of document identification DS to
m=..., r ... }, document r (r ∈ DS wherein
m) the file characteristics vector be (dw
R1, dw
R2..., dw
RL), therefore the query feature vector of described user m is made as:
Sw
Mk=(σ
5/ s) ∑
(r ∈ DSm)[dw
Rk/ (∑
(k ∈ DKr)Dw
Rk)], for each k ∈ K
Wherein s is described set DS
mElement number, σ
5For setting normal number.Described DK
rThe file characteristics vector K by described document i
d(r)=(dw
R1, dw
R2..., dw
RL) in the Q of numerical value maximum
rThe set that the corresponding feature of (r ∈ I) individual component forms, Q
rBe setup parameter.Regulation is worked as in formula
The time, get dw
Rk=0.The 4th is that described user m submits one group of user ID US to
m=..., r ... }, user r (r ∈ US wherein
m) the user characteristics vector be (uw
R1, uw
R2..., uw
RL), therefore the query feature vector of described user m is made as:
Sw
Mk=(σ
6/ s) ∑
(r ∈ USm)[uw
Rk/ (∑
(k ∈ UKr)Uw
Rk)], for each k ∈ K
Wherein s is described set US
mElement number, σ
6For setting normal number.Described UK
rThe user characteristics vector K by described user r
u(r)=uw
R1, uw
R2..., uw
RL) in the P of numerical value maximum
rThe set that the corresponding feature of (r ∈ J) individual component forms, Pr is setup parameter.Regulation is worked as in formula
The time, get uw
Rk=0.
In described application example 2, a kind of algorithm of described mathematical distance is as follows: the query feature vector of establishing described user m is K
s(m)=(sw
M1, sw
M2..., sw
Mk..., sw
ML), the file characteristics of document i vector is K among the described document sets I
d(i)=(dw
I1, dw
I2..., dw
Ik..., dw
IL), then the mathematical distance between the file characteristics vector of document i is defined as among the query feature vector of described user m and the described document sets I:
||K
s(m)，K
d(i)||＝[∑
k(sw
mk·dw
ik)]/{[∑
k(sw
mk)
2]
1/2·[∑
k(dw
ik)
2]
1/2}，k∈K
Application example 3.
Application example 3 is application of the described method of Fig. 5, i.e. one group of specific user's of inquiry application in the internet.Its step is as follows:
Repeatedly use described method to obtain the user characteristics vector that described user collects a plurality of users among the J;
Receive the querying condition that user m (m ∈ J) submits to, described querying condition comprises at least one feature that is queried the user;
According to described querying condition, generate the query feature vector of described user m;
Calculating described user collects each user's among the J (not comprising described user m) user characteristics vector and the mathematical distance between the described query feature vector and according to described mathematical distance described user is collected J and sort;
According to described ranking results described user is collected that the sign of certain customers sends to described user m among the J.
In described application example 3, the query feature vector K of described user m
s(m)=(sw
M1, sw
M2..., sw
Mk..., sw
ML) method to set up identical with the method to set up of query feature vector in the application example 2.
In described application example 3, a kind of algorithm of described mathematical distance is as follows: the query feature vector of establishing described user m is K
s(m)=(sw
M1, sw
M2..., sw
Mk..., sw
ML), described user integrates the user characteristics vector of user i among the J as K
u(i)=(uw
I1, uw
I2..., uw
Ik..., uw
IL), then the mathematical distance that collects between the user characteristics vector of user i among the I of the query feature vector of described user m and described user is defined as:
||K
s(m)，K
u(i)||＝[∑
k(sw
mk·uw
ik)]/{[∑
k(sw
mk)
2]
1/2·[∑
k(uw
ik)
2]
1/2}，k∈K
Application example 4.
Application example 4 be Fig. 5 method in an application in ad distribution field, comprise the steps:
The described user of file characteristics vector sum who repeatedly uses described method to obtain a plurality of documents among the described document sets I collects the user characteristics vector of a plurality of users among the J, and wherein said user integrates the user characteristics vector of user m among the J as K
u(m)=(uw
M1, uw
M2..., uw
Mk..., uw
ML), the file characteristics vector of the document n among the described document sets I is K
d(n)=(dw
N1, dw
N2..., dw
Nk..., dw
NL).If set of advertisements A={1,2 ..., G}, the characteristic of advertisement vector K of advertisement g (g ∈ A)
a(g)=(aw
G1, aw
G2..., aw
Gk..., aw
GL), aw wherein
GkRepresent the degree of correlation of described advertisement g and feature k (k ∈ K), G is the advertisement number, then carries out following steps:
Receive the signal of user m (m ∈ J) access document n (n ∈ I), described signal comprises the user ID of described user m and the document identification of described document n at least;
According to the user ID of described user m, obtain the user characteristics vector K of described user m
u(m);
According to the document identification of described document n, obtain the file characteristics vector K of described document n
d(n);
Calculate the characteristic of advertisement vector K of each advertisement among the described set of advertisements A
a(g) with the vectorial K of the user characteristics of described user m
u(m) mathematical distance 1 between; Calculate the characteristic of advertisement vector K of each advertisement among the described set of advertisements A
a(g) with the vectorial K of the file characteristics of described document n
d(n) mathematical distance 2 between;
Described mathematical distance 1 and described mathematical distance 2 according to each advertisement among the described set of advertisements A generate mathematical distance 3, and sort according to each advertisement among 3 couples of described set of advertisements A of described mathematical distance;
According to the result of described ordering, the described document n that at least one advertisement among the described set of advertisements A is put into described document n and will be put into advertisement sends to described user m.
A kind of algorithm of mathematical distance described in the application example 4 is as follows: establish described mathematical distance 1 and be ug (g, m), described mathematical distance 2 is dg (g, n), and described mathematical distance 3 is distance (g, m, n), then
ug(g，m)＝[∑
k(uw
mk·aw
gk)]/{[∑
k(uw
mk)
2]
1/2·[∑
k(aw
gk)
2]
1/2}
dg(g，n)＝[∑
k(dw
nk·aw
gk)]/{[∑
k(dw
nk)
2]
1/2·[∑
k(aw
gk)
2]
1/2}
distance(g，m，n)＝σ
7·ug(g，m)+σ
8·dg(g，n)
σ wherein
7And σ
8Be the setting nonnegative number, and σ
7+ σ
8=1, n ∈ I, m ∈ J, g ∈ A, k ∈ K.
In described application example 4, the characteristic of advertisement of described advertisement g (g ∈ A) vector K
a(g)=(aw
G1, aw
G2..., aw
Gk..., aw
GL) initial value two kinds of methods to set up are arranged.The first is the static assignment method, namely according to the affiliated field of advertisement and the audient group of advertisement, the initial value of each component of characteristic of advertisement vector is set manually.Aw for example is set
G2=3.5, aw
G4=3.7, other component values is 0, aw
Gk∈ [a, b], wherein a and b are for setting constant.The second is dynamic assignment method, is about to advertisement g (g ∈ A) and regards a document h (h ∈ I) as, (for example clicks advertisement) after each user has accessed advertisement document h, and application drawing 5 described methods are upgraded the file characteristics vector of advertisement document h.When needs use the characteristic of advertisement vector of described advertisement g (g ∈ A), K is set
a(g)=K
d(h).Described advertisement g is two kinds of different numbering forms of same advertisement with described advertisement document h, and described advertisement g is the numbering in described set of advertisements A, and described advertisement document h is the numbering in described document sets I.
Fig. 6 is a kind of method flow diagram of determining user characteristics in the internet.This method is based on signal that the user gets in touch with other users and upgrades described user characteristics vector.Described method comprises the steps:
S20. for the user collects J={1,2 ..., a part of user arranges user characteristics vector initial value among the N}, and the default initial value of user characteristics vector is null vector;
S21. receive the signal of user j (j ∈ J) contact user i (i ∈ J), described signal comprises the user ID of described user j and the user ID of described user i at least;
S22. according to the user ID of described user j, read the user characteristics vector K of described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL), uw wherein
JkThe degree of correlation that represents described user j and feature k (k ∈ K);
S23. according to the user ID of described user i, read the user characteristics vector K of described user i
u(i)=(uw
I1, uw
I2..., uw
Ik..., uw
IL), uw wherein
IkThe degree of correlation that represents described user i and feature k (k ∈ K);
S24. use following algorithm, upgrade the user characteristics vector of described user j:
K
u *(j)＝function3[K
u(i)，K
u(j)]
Wherein, described function3[K
u(i), K
u(j)] be increasing function, K
u(j) and K
u *(j) represent respectively to upgrade front user characteristics vector with upgrading rear described user j, K
u(i) the user characteristics vector of the described user i of expression.After stating algorithm in the use, need to be to K
u(j) upgrade, i.e. K
u(j)=K
u *(j).
In the described method of Fig. 6, also comprise the algorithm that the bean vermicelli proper vector of described user i is upgraded.Namely according to the user ID of described user i, read the bean vermicelli proper vector K of described user i
f(i)=(fw
I1, fw
I2..., fw
Ik..., fw
IL), fw wherein
IkBe the bean vermicelli colony of described user i and the degree of correlation of feature k (k ∈ K), then use the bean vermicelli proper vector that following algorithm upgrades described user i:
K
f *(i)＝function4[K
f(i)，K
u(j)]
Wherein, described function4[K
f(i), K
u(j)] be increasing function, K
f(i) and K
f *(i) represent respectively to upgrade bean vermicelli proper vector front and the rear described user i of renewal, K
u(j) be the user characteristics vector of described user j.After stating algorithm in the use, need to be to K
f(j) upgrade, i.e. K
f(j)=K
f *(j).
In the described method of Fig. 6, also comprise the algorithm that the user characteristics vector to described user i upgrades.Namely according to the user characteristics vector K that obtains
u(i) and K
u(j), the user characteristics vector of described user i upgraded:
K
u *(i)＝function5[K
u(i)，K
u(j)]
Wherein, described function5[K
u(i), K
u(j)] be increasing function, K
u(i) and K
u *(i) represent respectively to upgrade front user characteristics vector with upgrading rear described user i, K
u(j) the user characteristics vector of the described user j of expression.
Need to prove and upgrade K
u *(j) algorithm, renewal K
f *(i) algorithm and renewal K
u *(i) algorithm can independently use, and also can will upgrade K
u *(j) algorithm and renewal K
f *(i) algorithm uses together and will upgrade K
u *(j) algorithm and renewal K
u *(i) algorithm uses together.
In the described method of Fig. 6, described contact comprises a kind of in the following situation at least: add concern (follow) in microblogging, a user transmits, comments on or collect another user's a microblogging, adds as a friend in the social networks or point-to-point posting a letter.If for example user j has paid close attention to user i in microblogging, we just say that user j has got in touch with user i.
Application example 5.
Application example 5 provides K
u *(j)=function3[K
u(i), K
u(j)] specific algorithm, namely
Uw
Jk *=uw
Jk+ λ
3(t) ζ
3(ik) f
3(uw
Ik), for each k ∈ UK
i,
In described specific algorithm, t is the type of described contact, it is the mode that described user j gets in touch with described user i, for example t=41 represents that user j has paid close attention to microblogging content, the t=43 that described user i, t=42 represent that user j has transmitted described user i and represented that user j has commented on the microblogging content of described user i, t=44 represents that user j has collected the microblogging content of described user i, and t=51 represents to add as a friend etc.λ
3(t) be the function of t, for example λ
3(41)=1.9, λ
3(42)=1.3.
In the described specific algorithm, ζ
3(ik) other user compares among the J in order described user i and described user are collected, the significance level parameter of its k feature.ζ
3(ik) can be made as k component fw of the bean vermicelli proper vector of described user i
IkIncreasing function.For example establish ζ
3(ik)=c
2Fw
Ik/ (∑
(k ∈ K)Fw
Ik), c wherein
2For setting constant.ζ
3(ik) also can artificially arrange, for example any i ∈ J and k ∈ K be arranged ζ
3(ik)=1.
In described specific algorithm, described UK
iThe user characteristics vector K by described user i
u(i)=(uw
I1, uw
I2..., uw
Ik..., uw
IL) in the P of numerical value maximum
iThe set that the corresponding feature of (i ∈ J) individual component forms, described UK
jThe user characteristics vector K by described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL) in the P of numerical value maximum
jThe set that the corresponding feature of (j ∈ J) individual component forms, P
iAnd P
jFor setting constant.
In the described specific algorithm, f
3(uw
Ik) be increasing function.For example establish f
3(uw
Ik)=σ
9Uw
Ik/ (∑
(k ∈ UKi)Uw
Ik) or establish f
3(uw
Ik)=σ
9Uw
Ik, σ wherein
9For setting normal number.
In described specific algorithm, uw
IkAnd uw
Ik *Represent respectively to upgrade the user characteristics of front and the rear described user i of renewal to a flow control k component, uw
JkAnd uw
Jk *Represent respectively to upgrade the user characteristics of front and the rear described user j of renewal to a flow control k component.After described specific algorithm is complete, carry out following assignment, namely for each k ∈ UK
i, uw is set
Jk=uw
Jk *
After repeatedly using the described algorithm of Fig. 6, the numerical value of the component of user characteristics vector can increase the last memory capacity that surpasses storer gradually.Solution is identical with the method for process user proper vector among Fig. 5.
The below provides described algorithm K
f *(i)=function4[K
f(i), K
u(j)] concrete methods of realizing, namely
Fw
Ik *=fw
Ik+ λ
4(t) ζ
4(jk) f
4(uw
Jk), for each k ∈ UK
j,
Wherein, t is the type of described contact, λ
4(t) be the function of t, ζ
4(jk) be the significance level parameter of k the feature of described user j.ζ
4(jk) can be made as k component fw of bean vermicelli proper vector of described user j
JkIncreasing function, for example establish ζ
4(jk)=c
3Fw
Jk/ (∑
(k ∈ K)Fw
Jk), c wherein
3For setting constant.ζ
4(jk) also can artificially arrange, for example any j ∈ J and k ∈ K be arranged ζ
4(jk)=1.Described UK
jThe user characteristics vector K by described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL) in the P of numerical value maximum
jThe set that the corresponding feature of individual component forms, P
jBe setup parameter, fw
IkAnd fw
Ik *Represent respectively to upgrade k component of bean vermicelli proper vector front and the rear described user i of renewal, uw
JkRepresent that the user characteristics of described user j is to a flow control k component.After described specific algorithm is complete, for each k ∈ UK
j, fw is set
Ik=fw
Ik *
In the described specific algorithm, f
4(uw
Jk) be increasing function.For example establish f
4(uwjk)=σ
10Uw
Jk/ (∑
(k ∈ UKj)Uw
Jk) or establish f
4(uw
Jk)=σ
10Uw
Jk, σ wherein
10For setting normal number.
Upgrading described K
f *(i) also comprise in the algorithm described user collected the algorithm that the bean vermicelli proper vector of each user among the J is revised, namely select at least one k ∈ K and each k that selects is handled as follows: to each j (j ∈ J) with fw
JkBe mapped as g
3(fw
Jk), then each j (j ∈ J) is arranged fw
Jk=g
3(fw
Jk), g wherein
3(fw
Jk) be increasing function, g
3(fw
Jk) ∈ [a, b], a and b are the setting constant.In this correction algorithm, the k that selects will travel through each feature among the described feature set K, considers the restriction of system-computed processing power, can select at times the different characteristic k among the feature set K to be used described correction algorithm.
Described algorithm K
u *(i)=function5[K
u(i), K
u(j)] specific algorithm and described algorithm K
u *(j)=function3[K
u(i), K
u(j)] specific algorithm is similar, as long as i and j in latter's specific algorithm are exchanged.
In addition, the described method of Fig. 6 comprises one group of specific user's of inquiry in the internet application.Its specific implementation step is identical with application example 3.
Application example 6.
Application example 6 is that Fig. 6 method is in an application in ad distribution field.Comprise the steps:
Repeatedly use described method to obtain the user characteristics vector that described user collects a plurality of users among the J, wherein said user integrates the user characteristics vector of user m among the J as K
u(m)=(uw
M1, uw
M2..., uw
Mk..., uw
ML), the user characteristics vector of user n is K
u(n)=(uw
N1, uw
N2..., uw
Nk..., uw
NL).If set of advertisements is A={1,2 ..., G}, the characteristic of advertisement vector K of advertisement g (g ∈ A)
a(g)=(aw
G1, aw
G2..., aw
Gk..., aw
GL), aw wherein
GkRepresent the degree of correlation of described advertisement g and feature k (k ∈ K), G is the advertisement number, then carries out following steps:
Receive the signal of user m (m ∈ J) contact user n (n ∈ J), described signal comprises the user ID of described user m and the user ID of described user n at least;
According to the user ID of described user m, read the user characteristics vector K of described user m
u(m);
According to the user ID of described user n, read the user characteristics vector K of described user n
u(n);
Calculate the characteristic of advertisement vector K of each advertisement among the described set of advertisements A
a(g) with the vectorial K of the user characteristics of described user m
u(m) mathematical distance 4 between; Calculate the characteristic of advertisement vector K of each advertisement among the described set of advertisements A
a(g) with the vectorial K of the user characteristics of described user n
u(n) mathematical distance 5 between;
Described mathematical distance 4 and described mathematical distance 5 according to each advertisement among the described set of advertisements A generate mathematical distance 6, and sort according to each advertisement among 6 couples of described set of advertisements A of described mathematical distance;
Give described user m according to described ranking results with at least one advertisement pushing among the described set of advertisements A.
A kind of algorithm in mathematical distance described in the application example 6 is as follows: establish described mathematical distance 4 and be ug (g, m), described mathematical distance 5 is ug (g, n), and described mathematical distance 6 is distance (g, m, n), then has
ug(g，m)＝[∑
k(uw
mk·aw
gk)]/{[∑
k(uw
mk)
2]
1/2·[∑
k(aw
gk)
2]
1/2}
ug(g，n)＝[∑
k(uw
nk·aw
gk)]/{[∑
k(uw
nk)
2]
1/2·[∑
k(aw
gk)
2]
1/2}
distance(g，m，n)＝σ
11·ug(g，m)+σ
12·ug(g，n)
σ wherein
11And σ
12Be the setting nonnegative number, and σ
11+ σ
12=1, n ∈ I, m ∈ J, g ∈ A, k ∈ K.
In described application example 6, the characteristic of advertisement of described advertisement g (g ∈ A) vector K
a(g)=(aw
G1, aw
G2..., aw
Gk..., aw
GL) the initial value method to set up identical with method to set up in the application example 4.
The described method of Fig. 5 and Fig. 6 can be united use.These two kinds of methods not only can both be upgraded the user characteristics vector, and two kinds of methods can replenish mutually, so that the user characteristics vector reflects the feature of relative users better.
Fig. 7 is a kind of system that determines user characteristics in the internet.
Described system upgrades the user characteristics vector by dual mode, and a kind of is after the user has accessed a document, upgrades the file characteristics vector of described user's the described document of user characteristics vector sum; Another kind is after the user has got in touch with another user, upgrades wherein at least one user's user characteristics vector.Described system comprises following functional module:
User characteristics vector initial value arranges module 211: according to user characteristics the user is set and collects J={1, and 2 ..., the user characteristics of certain customers vector initial value among the N}, and it is stored in the customer data base 220;
File characteristics vector initial value arranges module 212: according to file characteristics document sets I={1 is set, and 2 ..., the file characteristics of partial document vector initial value among the M}, and it is stored in the document database 230;
Characteristic of advertisement vector initial value arranges module 213: according to characteristic of advertisement set of advertisements A={1 is set, and 2 ..., the characteristic of advertisement of part advertisement vector initial value among the G}, and it is stored in the advertising database 240;
User's access document signal acquisition module 214: the signal that obtains user j (j ∈ J) access document i (i ∈ I), wherein comprise at least the document identification of described document i and the user ID of described user j, described signal storage is in customer data base 220; Described user j be user 1 (101), user 2 (102) ..., any one user among the user N (103); Described document i be on the internet website 1 (301), website 2 (302) ..., storage has a uniquely identified file among the website N (303);
User's contact user signal acquisition module 215: the signal that obtains user j (j ∈ J) contact user i (i ∈ J), wherein comprise at least the user ID of described user j and the user ID of described user i, and with described signal storage in customer data base 220; Described user j and described user i be user 1 (101), user 2 (102) ..., any two users among the user N (103);
Proper vector update module 1 (216): according to the document identification of the described document i that in described user's access document signal acquisition module 214, obtains and the user ID of described user j, read the user characteristics vector of the described user j of file characteristics vector sum of described document i, then upgrade the user characteristics vector of the described user j of file characteristics vector sum of described document i, the file characteristics vector of described document i is the function of the user characteristics vector of the described user j of file characteristics vector sum of described document i before upgrading after upgrading, the user characteristics vector of described user j is the function of the user characteristics vector of the described user j of file characteristics vector sum of described document i before upgrading after upgrading, and concrete methods of realizing is identical with the described method of Fig. 5;
Proper vector update module 2 (216): according to the described user j that in described user's contact user signal acquisition module 215, obtains and the user ID of described user i, read the user characteristics vector of described user j and described user i and the bean vermicelli proper vector of described user i, then upgrade the bean vermicelli proper vector of the user characteristics vector sum user i of described user j, the user characteristics vector of described user j is the function of the user characteristics vector of the described user i of user characteristics vector sum of described user j before upgrading after upgrading, the bean vermicelli proper vector of described user i is the function of the user characteristics vector of the bean vermicelli proper vector of described user i before upgrading and described user j after upgrading, and concrete methods of realizing is identical with the described method of Fig. 6;
Advertisement selection and present module 217: according to the document identification of the described document i that in described user's access document signal acquisition module 214, obtains and the user ID of described user j, read the user characteristics vector of the described user j of file characteristics vector sum of described document i, calculate the mathematical distance 3 of each advertisement among the described set of advertisements A, and sort according to 3 couples of described set of advertisements A of described mathematical distance, and according to described ranking results described user j is presented at least one advertisement; According to the described user j that in described user's contact user signal acquisition module 215, obtains and the user ID of described user i, calculate the mathematical distance 6 of each advertisement among the described set of advertisements A, and sort according to 6 couples of described set of advertisements A of described mathematical distance, and according to described ranking results described user j is presented at least one advertisement; The computing method of the mathematical distance 3 in described mathematical distance 3 and the application example 4 are identical, and the computing method of the mathematical distance 6 in described mathematical distance 6 and the application example 6 are identical;
Document query module 218: receive the querying condition that user m (m ∈ J) submits to, and according to described querying condition generated query proper vector 1, then calculate the file characteristics vector of each document among the described document I and the mathematical distance 7 between the described query feature vector 1, and accordingly described document I is sorted, and according to described ranking results the sign of partial document among the described document sets I is sent to described user m, described query feature vector 1 is identical with the generation method of query feature vector in the described application example 2, and the computing method of the described mathematical distance in described mathematical distance 7 and the described application example 2 are identical;
User's enquiry module 219: receive the querying condition that user m (m ∈ J) submits to, according to described querying condition generated query proper vector 2, then calculate described user and collect the user characteristics vector of each user among the J and the mathematical distance 8 between the described query feature vector 2, and accordingly described user is collected J and sort, and according to described ranking results described user is collected that a part of user's sign sends to described user m among the J, described query feature vector 2 is identical with the generation method of query feature vector in the described application example 2, and the computing method of the described mathematical distance in described mathematical distance 8 and the described application example 3 are identical.
Described user i, user j and user m in above-mentioned each module represent respectively described user and collect any one user among the J.Described document i and document n represent respectively any one document among the described document sets I.In order to express easily, in modules, only listed the application example of described user i, user j and user m and document i and document n.In addition, described proper vector update module 216 is comprised of two parts, comprises described proper vector update module 1 and described proper vector update module 2.
The above application example only is better application example of the present invention, is not to limit protection scope of the present invention.
Claims (17)
1. a method of determining on the internet user characteristics is characterized in that, storage document sets I={1 in server, 2 ..., M}, user collect J={1,2 ..., N} and feature set K={1,2 ..., L}, wherein M is the document number, N is user's number, and L is Characteristic Number, and carries out following steps:
Receive the signal of user j (j ∈ J) access document i (i ∈ I), described signal comprises the user ID of described user j and the document identification of described document i at least;
According to described document identification, read the file characteristics vector K of described document i
d(i)=(dw
I1, dw
I2..., dw
Ik..., dw
IL), dw wherein
IkThe degree of correlation that represents described document i and feature k (k ∈ K);
According to described user ID, read the user characteristics vector K of described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL), uw wherein
JkThe degree of correlation that represents described user j and feature k (k ∈ K);
Upgrade the user characteristics vector of the described user j of file characteristics vector sum of described document i with following algorithm:
K
d *(i)＝function1[K
d(i)，K
u(j)]
K
u *(j)＝function2[K
d(i)，K
u(j)]
Wherein, described function1[K
d(i), K
uAnd described function2[K (j)]
d(i), K
u(j)] be increasing function, K
d(i) and K
d *(i) represent respectively to upgrade front file characteristics vector with upgrading rear described document i, K
u(j) and K
u *(j) represent respectively to upgrade front user characteristics vector with upgrading rear described user j.
2. method according to claim 1 is characterized in that, in an application example of described algorithm, uses following specific algorithm to upgrade the user characteristics vector of the described user j of file characteristics vector sum of described document i:
Dw
Ik *=dw
Ik+ λ
1(t) ζ
1(jk) f
1(uw
Jk), for each k ∈ UK
j,
Uw
Jk *=uw
Jk+ λ
2(t) ζ
2(i) f
2(dw
Ik), for each k ∈ DK
i,
Wherein, described f
1(uw
Jk) and f
2(dw
Ik) being increasing function, t is the type of described access, λ
1(t) and λ
2(t) be the function of t, ζ
1(jk) be the significance level parameter of k the feature of described user j, ζ
2(i) be the significance level parameter of described document i, described DK
iThe file characteristics vector K by described document i
d(i)=(dw
I1, dw
I2..., dw
Ik..., dw
IL) in the Q of numerical value maximum
iThe set that the corresponding feature of individual component forms, described UK
jThe user characteristics vector K by described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL) in the P of numerical value maximum
jThe set that the corresponding feature of individual component forms, Q
iAnd P
jBe setup parameter, dw
IkAnd dw
Ik *Represent respectively to upgrade the file characteristics of front and the rear described document i of renewal to a flow control k component, uw
JkAnd uw
Jk *Represent respectively to upgrade the user characteristics of front and the rear described user j of renewal to a flow control k component.
3. method according to claim 1, it is characterized in that, described method also comprises described user collected the algorithm that the user characteristics vector of each user among the J is revised, and namely selects at least one k ∈ K and each k that selects is handled as follows: to each j (j ∈ J) with uw
JkBe mapped as g
1(uw
Jk), then each j (j ∈ J) is arranged uw
Jk=g
1(uw
Jk), g wherein
1(uw
Jk) be increasing function, g
1(uw
Jk) ∈ [a, b], a and b are the setting constant.
4. method according to claim 1, it is characterized in that, described method also comprises the algorithm that the file characteristics vector of each document among the described document sets I is revised, and namely selects at least one k ∈ K and each k that selects is handled as follows: to each i (i ∈ I) with dw
IkBe mapped as g
2(dw
Ik), then each i (i ∈ I) is arranged dw
Ik=g
2(dw
Ik), g wherein
2(dw
Ik) be increasing function, g
2(dw
Ik) ∈ [a, b], a and b are the setting constant.
5. method according to claim 1 is characterized in that, described method comprises the application example of inquiring about on the internet one group of particular document, and its step is as follows:
The described user of file characteristics vector sum who repeatedly uses described method to obtain a plurality of documents among the described document sets I collects the user characteristics vector of a plurality of users among the J;
Receive the querying condition that user m (m ∈ J) submits to, comprising at least one feature that is queried document;
According to described querying condition, generate the query feature vector of described user m;
Calculate the file characteristics vector of each document among the described document sets I and the mathematical distance between the described query feature vector, and according to described mathematical distance described document sets I is sorted;
According to described ranking results the sign of partial document among the described document sets I is sent to described user m.
6. method according to claim 1 is characterized in that, described method comprises the application example of inquiring about on the internet one group of specific user, and its step is as follows:
Repeatedly use described method to obtain the user characteristics vector that described user collects a plurality of users among the J;
Receive the querying condition that user m (m ∈ J) submits to, comprising at least one feature that is queried the user;
According to described querying condition, generate the query feature vector of described user m;
Calculate described user and collect the user characteristics vector of each user among the J and the mathematical distance between the described query feature vector, and according to described mathematical distance described user is collected J and sort;
According to described ranking results described user is collected that the sign of certain customers sends to described user m among the J.
7. method according to claim 1, it is characterized in that, in an application example of described method, the described user of file characteristics vector sum who repeatedly uses described method to obtain a plurality of documents among the described document sets I collects the user characteristics vector of a plurality of users among the J, storage set of advertisements A={1,2 ..., G}, the characteristic of advertisement vector K of advertisement g (g ∈ A)
a(g)=(aw
G1, aw
G2..., aw
Gk..., aw
GL), aw wherein
GkRepresent the degree of correlation of described advertisement g and feature k (k ∈ K), G is the advertisement number, then carries out following steps:
Receive the signal of user m (m ∈ J) access document n (n ∈ I), described signal comprises the user ID of described user m and the document identification of described document n at least;
According to the user ID of described user m, obtain the user characteristics vector of described user m;
According to the document identification of described document n, obtain the file characteristics vector of described document n;
Mathematical distance 1 between the characteristic of advertisement vector that calculates each advertisement among the described set of advertisements A and the user characteristics vector of described user m is calculated the mathematical distance 2 between the file characteristics vector of the vectorial and described document n of the characteristic of advertisement of each advertisement among the described set of advertisements A;
Described mathematical distance 1 and described mathematical distance 2 according to each advertisement among the described set of advertisements A generate mathematical distance 3, and sort according to each advertisement among 3 couples of described set of advertisements A of described mathematical distance;
According to the result of described ordering, the described document n that at least one advertisement among the described set of advertisements A is put into described document n and will be put into advertisement sends to described user m.
8. a method of determining on the internet user characteristics is characterized in that, the storage user collects J={1 in server, 2 ..., N} and feature set K={1,2 ..., L}, wherein N is user's number, L is Characteristic Number; And in described server, carry out following steps:
Receive the signal of user j (j ∈ J) contact user i (i ∈ J), described signal comprises the user ID of described user j and the user ID of described user i at least;
According to the user ID of described user j, read the user characteristics vector K of described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL), uw wherein
JkThe degree of correlation that represents described user j and feature k (k ∈ K);
According to the user ID of described user i, read the user characteristics vector K of described user i
u(i)=(uw
I1, uw
I2..., uw
Ik..., uw
IL), uw wherein
IkThe degree of correlation that represents described user i and feature k (k ∈ K);
Use following algorithm that the user characteristics vector of described user j is upgraded:
K
u *(j)＝function3[K
u(i)，K
u(j)]
Wherein, described function3[K
u(i), K
u(j)] be increasing function, K
u(j) and K
u *(j) represent respectively to upgrade front user characteristics vector with upgrading rear described user j, K
u(i) the user characteristics vector of the described user i of expression.
9. method according to claim 8 is characterized in that, in an application example of described algorithm, the concrete update algorithm of the user characteristics vector of described user j is as follows:
Uw
Jk *=uw
Jk+ λ
3(t) ζ
3(ik) f
3(uw
Ik), for each k ∈ UK
i,
F wherein
3(uw
Ik) be increasing function, t is the type of described contact, λ
3(t) be the function of t, ζ
3(ik) be the significance level parameter of k the feature of described user i, described UK
iThe user characteristics vector K by described user i
u(i)=(uw
I1, uw
I2..., uw
Ik..., uw
IL) in the P of numerical value maximum
iThe set that the corresponding feature of individual component forms, P
iFor setting constant, uw
JkAnd uw
Jk *Represent respectively to upgrade the user characteristics of front and the rear described user j of renewal to a flow control k component, uw
IkRepresent that the user characteristics of described user i is to a flow control k component.
10. method according to claim 8, it is characterized in that, described method also comprises described user collected the algorithm that the user characteristics vector of each user among the J is revised, and namely selects at least one k ∈ K and each k that selects is handled as follows: to each j (j ∈ J) with uw
JkBe mapped as g
1(uw
Jk), then each j (j ∈ J) is arranged uw
Jk=g
1(uw
Jk), g wherein
1(uw
Jk) be increasing function, g
1(uw
Jk) ∈ [a, b], a and b are the setting constant.
11. method according to claim 8 is characterized in that, described method is further comprising the steps of:
According to the user ID of described user i, read the bean vermicelli proper vector K of described user i
f(i)=(fw
I1, fw
I2..., fw
Ik..., fw
IL), fw wherein
IkBe the bean vermicelli colony of described user i and the degree of correlation of feature k (k ∈ K), then use the bean vermicelli proper vector that following algorithm upgrades described user i:
K
f *(i)＝function4[K
f(i)，K
u(j)]
Wherein, described function4[K
f(i), K
u(j)] be increasing function, K
f(i) and K
f *(i) represent respectively to upgrade bean vermicelli proper vector front and the rear described user i of renewal, K
u(j) the user characteristics vector of the described user j of expression.
12. method according to claim 11 is characterized in that, in an application example of described algorithm, the concrete update algorithm of the bean vermicelli proper vector of described user i is as follows:
Fw
Ik *=fw
Ik+ λ
4(t) ζ
4(jk) f
4(uw
Jk), for each k ∈ UK
j,
F wherein
4(uw
Jk) be increasing function, t is the type of described contact, λ
4(t) be the function of t, ζ
4(jk) be the significance level parameter of k the feature of described user j, described UK
jThe user characteristics vector K by described user j
u(j)=(uw
J1, uw
J2..., uw
Jk..., uw
JL) in the P of numerical value maximum
jThe set that the corresponding feature of individual component forms, P
jBe setup parameter, fw
IkAnd fw
Ik *Represent respectively to upgrade k component of bean vermicelli proper vector front and the rear described user i of renewal, uw
JkRepresent that the user characteristics of described user j is to a flow control k component.
13. method according to claim 11, it is characterized in that, described method also comprises described user collected the algorithm that the bean vermicelli proper vector of each user among the J is revised, and namely selects at least one k ∈ K and each k that selects is handled as follows: to each j (j ∈ J) with fw
JkBe mapped as g
3(fw
Jk), then each j (j ∈ J) is arranged fw
Jk=g
3(fw
Jk), g wherein
3(fw
Jk) be increasing function, g
3(fw
Jk) ∈ [a, b], a and b are the setting constant.
14. method according to claim 8 is characterized in that, described algorithm also comprises the step that the user characteristics vector to described user i upgrades:
K
u *(i)＝function5[K
u(i)，K
u(j)]
Wherein, described function5[K
u(i), K
u(j)] be increasing function, K
u(i) and K
u *(i) represent respectively to upgrade front user characteristics vector with upgrading rear described user i, K
u(j) the user characteristics vector of the described user j of expression.
15. method according to claim 8 is characterized in that, described method comprises the application example of inquiring about on the internet one group of specific user, and its step is as follows:
Repeatedly use described method to obtain the user characteristics vector that described user collects a plurality of users among the J;
Receive the querying condition that user m (m ∈ J) submits to, comprising at least one feature that is queried the user;
According to described querying condition, generate the query feature vector of described user m;
Calculate described user and collect the user characteristics vector of each user among the J and the mathematical distance between the described query feature vector, and according to described mathematical distance described user is collected J and sort;
According to described ranking results described user is collected that the sign of certain customers sends to described user m among the J.
16. method according to claim 8 is characterized in that, in an application example of described method, repeatedly use described method to obtain the user characteristics vector that described user collects a plurality of users among the J, storage set of advertisements A={1,2, ..., G}, the characteristic of advertisement vector of advertisement g (g ∈ A) is K
a(g)=(aw
G1, aw
G2..., aw
Gk..., aw
GL), aw wherein
GkRepresent the degree of correlation of described advertisement g and feature k (k ∈ K), G is the advertisement number; Then carry out following steps:
Receive the signal of user m (m ∈ J) contact user n (n ∈ J), described signal comprises the user ID of described user m and the user ID of described user n at least;
According to the user ID of described user m, read the user characteristics vector of described user m;
According to the user ID of described user n, read the user characteristics vector of described user n;
Calculate the mathematical distance 4 between the user characteristics vector of the characteristic of advertisement vector of each advertisement among the described set of advertisements A and described user m; Calculate the mathematical distance 5 between the user characteristics vector of the characteristic of advertisement vector of each advertisement among the described set of advertisements A and described user n;
Described mathematical distance 4 and described mathematical distance 5 according to each advertisement among the described set of advertisements A generate mathematical distance 6, and sort according to each advertisement among 6 couples of described set of advertisements A of described mathematical distance;
Give described user m according to described ranking results with at least one advertisement pushing among the described set of advertisements A.
17. a system that determines on the internet user characteristics is characterized in that, comprises with lower module:
User characteristics vector initial value arranges module: according to user's feature the user is set and collects J={1, and 2 ..., the user characteristics of certain customers vector initial value among the N}, and it is stored in the customer data base;
File characteristics vector initial value arranges module: the feature according to document arranges document sets I={1, and 2 ..., the file characteristics of partial document vector initial value among the M}, and it is stored in the document database;
Characteristic of advertisement vector initial value arranges module: according to the characteristic storage set of advertisements A={1 of advertisement, and 2 ..., the characteristic of advertisement of part advertisement vector initial value among the G}, and it is stored in the advertising database;
User's access document signal acquisition module: obtain the signal of user j (j ∈ J) access document i (i ∈ I), described signal comprises the document identification of described document i and the user ID of described user j at least, and described signal storage is in customer data base;
User's contact user signal acquisition module: the signal that obtains user j (j ∈ J) contact user i (i ∈ J), described signal comprises the user ID of described user j and the user ID of described user i at least, and with described signal storage in customer data base;
Proper vector update module 1: according to the document identification of the described document i that in described user's access document signal acquisition module, obtains and the user ID of described user j, read the user characteristics vector of the described user j of file characteristics vector sum of described document i, then upgrade the user characteristics vector of the described user j of file characteristics vector sum of described document i, the file characteristics vector of described document i is the function of the user characteristics vector of the described user j of file characteristics vector sum of described document i before upgrading after upgrading, and the user characteristics vector of described user j is the function of the user characteristics vector of the described user j of file characteristics vector sum of described document i before upgrading after upgrading;
Proper vector update module 2: according to the described user j that in described user's contact user signal acquisition module, obtains and the user ID of described user i, read the user characteristics vector of described user j and described user i and the bean vermicelli proper vector of described user i, then upgrade the bean vermicelli proper vector of the described user i of user characteristics vector sum of described user j, the user characteristics vector of described user j is the function of the user characteristics vector of the described user j of user characteristics vector sum of described user i before upgrading after upgrading, and the bean vermicelli proper vector of described user i is the function of the user characteristics vector of the bean vermicelli proper vector of described user i before upgrading and described user j after upgrading;
Advertisement selection and present module: according to the document identification of the described document i that in described user's access document signal acquisition module, obtains and the user ID of described user j, calculate the mathematical distance 3 of each advertisement among the described set of advertisements A, and sort according to 3 couples of described set of advertisements A of described mathematical distance, and according to described ranking results described user j is presented at least one advertisement; According to the described user j that in described user's contact user signal acquisition module, obtains and the user ID of described user i, calculate the mathematical distance 6 of each advertisement among the described set of advertisements A, and sort according to 6 couples of described set of advertisements A of described mathematical distance, and according to described ranking results described user j is presented at least one advertisement;
Document query module: receive the querying condition that user m (m ∈ J) submits to, described querying condition comprises at least one feature that is queried document, according to described querying condition generated query proper vector 1, then calculate the mathematical distance 7 between the query feature vector 1 of the file characteristics vector of each document among the described document sets I and described user m, and sort according to 7 couples of described document sets I of described mathematical distance, and according to described ranking results, the sign of a part of document among the described document sets I is sent to described user m;
User's enquiry module: receive the querying condition that user m (m ∈ J) submits to, described querying condition comprises at least one feature that is queried the user, according to described querying condition generated query proper vector 2, then calculate described user and collect the user characteristics vector of each user among the J and the mathematical distance 8 between the described query feature vector 2, and collect J according to 8 couples of described users of described mathematical distance and sort, and according to described ranking results described user is collected that a part of user's sign sends to described user m among the J.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102809074A CN102999540A (en) | 2011-09-10 | 2011-09-10 | Method and system for determining user features on Internet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2011102809074A CN102999540A (en) | 2011-09-10 | 2011-09-10 | Method and system for determining user features on Internet |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102999540A true CN102999540A (en) | 2013-03-27 |
Family
ID=47928113
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2011102809074A Pending CN102999540A (en) | 2011-09-10 | 2011-09-10 | Method and system for determining user features on Internet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102999540A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324648A (en) * | 2012-03-20 | 2013-09-25 | 祁勇 | Method and system for acquiring individuation characteristics of users and documents |
| CN103390008A (en) * | 2012-05-08 | 2013-11-13 | 祁勇 | Method and system for acquiring personalized features of user |
| CN107038169A (en) * | 2016-02-04 | 2017-08-11 | 佳能株式会社 | Object recommendation method and object recommendation equipment |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102651011A (en) * | 2011-02-27 | 2012-08-29 | 祁勇 | Method and system for determining document characteristic and user characteristic |
| CN102737055A (en) * | 2011-04-06 | 2012-10-17 | 祁勇 | Method and system for determining user features in social network |
| CN102880622A (en) * | 2011-07-15 | 2013-01-16 | 祁勇 | Method and system for determining user characteristics on internet |
-
2011
- 2011-09-10 CN CN2011102809074A patent/CN102999540A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102651011A (en) * | 2011-02-27 | 2012-08-29 | 祁勇 | Method and system for determining document characteristic and user characteristic |
| CN102737055A (en) * | 2011-04-06 | 2012-10-17 | 祁勇 | Method and system for determining user features in social network |
| CN102880622A (en) * | 2011-07-15 | 2013-01-16 | 祁勇 | Method and system for determining user characteristics on internet |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103324648A (en) * | 2012-03-20 | 2013-09-25 | 祁勇 | Method and system for acquiring individuation characteristics of users and documents |
| CN103390008A (en) * | 2012-05-08 | 2013-11-13 | 祁勇 | Method and system for acquiring personalized features of user |
| CN103390008B (en) * | 2012-05-08 | 2018-09-28 | 六六鱼信息科技（上海）有限公司 | A kind of method and system obtaining user individual feature |
| CN107038169A (en) * | 2016-02-04 | 2017-08-11 | 佳能株式会社 | Object recommendation method and object recommendation equipment |
| CN107038169B (en) * | 2016-02-04 | 2021-03-12 | 佳能株式会社 | Object recommendation method and object recommendation device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Nitu et al. | Improvising personalized travel recommendation system with recency effects | |
| CN101551806B (en) | Personalized website navigation method and system | |
| CN103544216B (en) | The information recommendation method and system of a kind of combination picture material and keyword | |
| CN103106285B (en) | Recommendation algorithm based on information security professional social network platform | |
| WO2018040069A1 (en) | Information recommendation system and method | |
| CN105488233A (en) | Reading information recommendation method and system | |
| JP2013168186A (en) | Review processing method and system | |
| TW201447797A (en) | Method and system for multi-phase ranking for content personalization | |
| KR20130079352A (en) | Product synthesis from multiple sources | |
| CN101329674A (en) | System and method for providing personalized searching | |
| KR20020007742A (en) | Methods of Information Recommendation based on Analysis of User's Information Preferences and Measurement of Information Popularity on the Internet and the System thereof | |
| CN102982042A (en) | Personalization content recommendation method and platform and system | |
| CN102651011B (en) | Method and system for determining document characteristic and user characteristic | |
| KR20120047079A (en) | Recommendation system using collective intelligence and method thereof | |
| CN103617230A (en) | Method and system for advertisement recommendation based microblog | |
| KR101090406B1 (en) | A profit distribution system for content provider and method thereof | |
| CN102929939A (en) | Personalized information supply method and device | |
| CN102880622A (en) | Method and system for determining user characteristics on internet | |
| Kotzias et al. | Home is where your friends are: Utilizing the social graph to locate twitter users in a city | |
| CN104050243A (en) | Network searching method and system combined with searching and social contact | |
| CN104199938A (en) | RSS-based agricultural land information sending method and system | |
| CN112868003A (en) | Entity-based search system using user interactivity | |
| CN104536957B (en) | Agricultural land circulation information retrieval method and system | |
| CN104123321B (en) | A kind of determining method and device for recommending picture | |
| CN108536763B (en) | Pull-down prompting method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20130327 |