CN109934629A - A kind of information-pushing method and device - Google Patents
A kind of information-pushing method and device Download PDFInfo
- Publication number
- CN109934629A CN109934629A CN201910184486.1A CN201910184486A CN109934629A CN 109934629 A CN109934629 A CN 109934629A CN 201910184486 A CN201910184486 A CN 201910184486A CN 109934629 A CN109934629 A CN 109934629A
- Authority
- CN
- China
- Prior art keywords
- hash
- vector
- function
- historical
- hash vector
- 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
- 238000000034 method Methods 0.000 title claims abstract description 31
- 239000013598 vector Substances 0.000 claims abstract description 180
- 238000004364 calculation method Methods 0.000 claims description 7
- 230000001419 dependent effect Effects 0.000 claims description 6
- 230000006870 function Effects 0.000 description 113
- 238000004422 calculation algorithm Methods 0.000 description 13
- 238000010586 diagram Methods 0.000 description 6
- 238000012549 training Methods 0.000 description 4
- 230000009471 action Effects 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007636 ensemble learning method Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
This application provides a kind of information-pushing method and devices, solve preset objective function, obtain hash function, use hash function, calculate the cryptographic Hash of the first object and the Hash vector of the second object, in the case where the cryptographic Hash of the cryptographic Hash of first object and second object meets preset condition, the information of the second object described in the first Object Push of Xiang Suoshu.Therefore, objective function also can be used for the first new object and obtains its Hash vector, and according to the similarity of Hash vector to the first Object Push information, to improve the specific aim of information push.
Description
Technical Field
The present application relates to the field of electronic information, and in particular, to an information pushing method and apparatus.
Background
Advertisement pushing is a common function of APP, that is, according to attributes of a user, hash vectors are configured for the user, according to attributes of goods, hash vectors are configured for the goods, and advertisements of goods with hash vectors similar to the hash vectors of the user are pushed to the user.
Generally, in the existing APP, on one hand, a commodity which a user is interested in is determined through an operation behavior of the user on the APP, and a hash vector is configured for the user according to the hash vector of the commodity which the user is interested in, and on the other hand, the hash vector is configured for the commodity according to a crowd which the commodity is interested in. And if the hash vector of a commodity is similar to the hash vector of a certain user, pushing the information of the commodity to the user.
However, for a new user, the interested goods cannot be determined, and for a new product, the interested people cannot be determined, so for a new user, usually, only advertisement information can be randomly pushed, and targeted advertisement pushing cannot be realized.
Disclosure of Invention
The application provides an information pushing method and device, and aims to solve the problem that targeted advertisement pushing cannot be achieved for a new user.
In order to achieve the above object, the present application provides the following technical solutions:
an information push method, comprising:
solving a preset target function to obtain a hash function, wherein the target function takes the hash function as a constraint condition;
calculating a hash value of the first object and a hash vector of the second object using the hash function;
and under the condition that the hash value of the first object and the hash value of the second object meet preset conditions, pushing information of the second object to the first object.
Optionally, the arguments of the objective function are a history hash vector of the first object and a history hash vector of the second object;
the objective function includes at least:
a first regularization factor comprised of a historical hash vector of the first object and a historical hash vector of the second object;
the first regularization factor is used to characterize an intra-class constraint for objects of the same type.
Optionally, the objective function further includes:
a second regularization factor comprised of a sum of bits in the historical hash vector of the first object and a sum of bits in the historical hash vector of the second object;
the second regularization factor is used to characterize inter-class constraints for objects of different types.
Optionally, the objective function further includes:
a preliminary objective function consisting of a similarity of the historical hash vector of the first object and the historical hash vector of the second object, and a satisfaction of the first object with the second object.
Optionally, the objective function is:
wherein h isijFor the satisfaction of the ith first object with the jth second object,historical hash vector for ith first objectHistorical hash vector with jth second objectThe degree of similarity of (a) to (b),for the k-th bit in the history hash vector of the ith first object,is the k bit, G, in the history hash vector of the jth second objectuHistorical hash vector for ith first object, GaIs the historical hash vector of the jth second object, xiIs an attribute of the ith first object, xjλ, β are preset coefficients for the attribute of the jth second object.
Optionally, the solving a preset objective function to obtain a hash function includes:
and taking a first historical hash vector and a second historical hash vector with the similarity meeting the preset condition as independent variables of the target function, optimizing dependent variables of the target function to obtain W, V, B and C, and determining the hash function as the target function, wherein the first historical hash vector is a hash vector of a first object, and the second historical hash vector is a hash vector of a second object.
Optionally, the calculating the hash value of the first object and the hash vector of the second object by using the hash function includes:
inputting the attribute value of the first object into the hash function to obtain a hash vector of the first object;
and inputting the attribute value of the second object into the hash function to obtain a hash vector of the second object.
An information pushing apparatus comprising:
the hash function acquisition module is used for solving a preset target function to obtain a hash function, and the target function takes the hash function as a constraint condition;
a calculation module for calculating a hash value of a first object and a hash vector of a second object using the hash function;
and the pushing module is used for pushing the information of the second object to the first object under the condition that the hash value of the first object and the hash value of the second object meet preset conditions.
Optionally, the arguments of the objective function are a history hash vector of the first object and a history hash vector of the second object;
the objective function includes at least:
a first regularization factor comprised of a historical hash vector of the first object and a historical hash vector of the second object;
the first regularization factor is used to characterize an intra-class constraint for objects of the same type.
Optionally, the objective function further includes:
a second regularization factor comprised of a sum of bits in the historical hash vector of the first object and a sum of bits in the historical hash vector of the second object;
the second regularization factor is used to characterize inter-class constraints for objects of different types.
Optionally, the objective function further includes:
a preliminary objective function consisting of a similarity of the historical hash vector of the first object and the historical hash vector of the second object, and a satisfaction of the first object with the second object.
Optionally, the objective function is:
wherein h isijFor the satisfaction of the ith first object with the jth second object,historical hash vector for ith first objectHistorical hash vector with jth second objectThe degree of similarity of (a) to (b),for the k-th bit in the history hash vector of the ith first object,is the k bit, G, in the history hash vector of the jth second objectuHistorical hash vector for ith first object, GaIs the historical hash vector of the jth second object, xiIs an attribute of the ith first object, xjλ, β are preset coefficients for the attribute of the ith second object.
Optionally, the hash function obtaining module is configured to solve a preset objective function to obtain a hash function, and includes:
and taking a first historical hash vector and a second historical hash vector with the similarity meeting the preset condition as independent variables of the target function, optimizing dependent variables of the target function to obtain W, V, B and C, and determining the hash function as the target function, wherein the first historical hash vector is a hash vector of a first object, and the second historical hash vector is a hash vector of a second object.
Optionally, the calculating module is configured to calculate the hash value of the first object and the hash vector of the second object by using the hash function, and includes:
the calculation module is specifically configured to input the attribute value of the first object into the hash function to obtain a hash vector of the first object; and inputting the attribute value of the second object into the hash function to obtain a hash vector of the second object.
The information pushing method and the information pushing device solve the preset target function to obtain the hash function, use the hash function to calculate the hash value of the first object and the hash vector of the second object, and push the information of the second object to the first object under the condition that the hash value of the first object and the hash value of the second object meet the preset conditions. Therefore, the hash vector of the new first object can be obtained by using the target function, and the information is pushed to the first object according to the similarity of the hash vector, so that the pertinence of information pushing is improved.
Drawings
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of an information pushing method disclosed in an embodiment of the present application;
FIG. 2(a) is a schematic diagram comparing the technical solution disclosed in the embodiment of the present application with the prior art;
FIG. 2(b) is a schematic diagram comparing the technical solutions disclosed in the examples of the present application with the prior art;
FIG. 2(c) is a schematic diagram comparing the technical solutions disclosed in the examples of the present application with the prior art;
fig. 3 is a schematic structural diagram of an information pushing apparatus disclosed in the embodiment of the present application.
Detailed Description
The information pushing method and the information pushing device disclosed by the embodiment of the application can be applied to any information pushing scene, the APP is taken as an example, and advertisement information needs to be pushed for a user while a business is provided for the user by one APP. Furthermore, constraint factors among different categories of commodities can be added in the pushing process, and the pertinence of advertisement pushing is further improved.
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
Fig. 1 is an information pushing method disclosed in an embodiment of the present application, including the following steps:
s101: and constructing an objective function.
In this embodiment, the linear hash function is defined as:
wherein,for the k-th bit in the history hash vector of the ith first object,is the k bit, x, in the history hash vector of the jth second objectiIs an attribute of the ith first object, xjIn the embodiment of the present application, the description will be given by taking the first object as a user and the second object as a product as an example, where the first object is an attribute of the ith second object.
W＝[w1,w2,...,wK]Is a parameter of each bit in the hash vector of the user, i.e. w1Parameter 1, … … and so on, wKIs the parameter of the k-th bit.
V＝[v1,v2,...,vK]Is a parameter for each bit in the hash vector of the advertisement, i.e., v1Parameter 1, … … and so on, vKIs the parameter of the k-th bit.
B＝[b1,b2,...,bK]Is the threshold value of each bit in the hash vector of the user, i.e. b1Threshold for bit 1, … … and so on, bKIs the threshold for the k-th bit.
C＝[c1,c2,...,cK]Is the threshold value of each bit in the hash vector of the advertisement, i.e., c1Threshold for bit 1, … … and so on, cKIs the threshold for the k-th bit.
sgn (x) is a sign function, and the expression is:
the similarity between the hash vector of the user and the hash vector of the commodity is as follows:
wherein equ (x, y) is a defined judgment function, i.e.
The above equation indicates that if the hash value of each bit in the hash vector of the user i and the hash vector of the commodity j are the same, then the hash values of the bits in the hash vectors of the user i and the commodity j are the sameIf the hash value of each bit in the hash vector of the user i is different from that of the hash vector of the commodity j, the hash value of each bit in the hash vector of the user i is different from that of the commodity j
In the advertisement pushing process, whether the advertisement is pushed for the user is judged according to the similarity of the hash vectors between the user and the commodities, and then the similarity of the hash vectors between the user and the commodities is consistent with the satisfaction degree of the user to the commodities.
Therefore, in this embodiment, the satisfaction of the user on the commodity is determined by the evaluation of the user on the commodity, and is divided into two levels, namely, the satisfaction hij1 or dissatisfaction hijIs-1. If the user rates the product as satisfactory, it is desirable that the hash vectors of the user and the advertisement are the same, and if the user rates the product as unsatisfactory, it is desirable that the hash vectors of the user and the advertisement are different. Namely, there are four cases:
1、hij＝1，
2、hij＝1，
3、hij＝-1，
4、hij＝-1，
the hash vector considering the user satisfaction is defined as
From the above four cases, it can be seen that if the similarity of the hash vectors of the user and the commodity is consistent with the satisfaction of the user on the commodity, then the similarity is consistent with the satisfaction of the user on the commodityEqual to 1, otherwise equal to-1. The objective function is to make the hash value similarity based on the user satisfaction as high as possible, i.e. make all the hash values of all the users and the advertisements as equal as possible, and construct a preliminary objective function as:
by optimizing the preliminary objective function, the hash vectors between the users and the commodities obtained through the final learning can correctly reflect the satisfaction degrees of the users to the commodities, namely if the users are satisfied with the commodities, the similarity degrees of the hash vectors of the users and the commodities are higher, otherwise, if the users are not satisfied with the commodities, the similarity degrees of the hash vectors of the users and the commodities are lower.
If the inter-class similarity problem and the intra-class similarity problem are considered when processing heterogeneous data, the accuracy of advertisement pushing is improved, and therefore the preliminary objective function is improved and two types of constraints are added, one type is inter-hash value class constraint and the other type is intra-hash value class constraint.
The inter-class constraint of the hash value is the constraint between the hash values of the same bit, and the inter-class constraint is added to avoid overfitting. According to the principle of information theory, the information provided by the maximized hash value can prevent the over-fitting phenomenon from occurring, and according to the principle of maximum entropy, the maximum entropy of one hash bit can ensure that the contained information amount is maximum. So that inter-class constraints can be converted into solving the maximum entropy problem, i.e.
WhereinThe k-th bit in the hash vector for all users. If and only ifThe time equation is maximum, namely half of the hash value of all users at the k bit is 1, and half of the hash value is-1, namely the hash value satisfiesSimilarly, the hash value of the advertisement is also satisfiedIt is not required in this embodiment that the sum of certain one-bit hash values of all users or all advertisements must be equal to 0, but the above condition is desirably as small as possible.
The intra-class constraint of hash values is to prevent extreme cases, such as: when the values of all hash bits of each user are the same, only one hash value is needed to represent one user, so the intra-class constraint should make different hash bits uncorrelated as much as possible, and the constraint condition is constructed
||Gu(Gu)T-I||2and||Ga(Ga)T-I||2
Where I is the unit array. In order to make the in-class constraint satisfied, the smaller the value of the above formula, the better.
Combining the constructed initial objective function, adding the two constraint conditions into the objective function as regularization factors, and obtaining a final objective function as follows:
wherein h isijFor the satisfaction of the ith first object with the jth second object,historical hash vector for ith first objectHistorical hash vector with jth second objectThe degree of similarity of (a) to (b),for the k-th bit in the history hash vector of the ith first object,is the k bit, G, in the history hash vector of the jth second objectuEntropy of the historical hash vector for the ith first object, GaEntropy, x, of the historical hash vector for the jth second objectiIs an attribute of the ith first object, xjλ, β are preset coefficients for the attribute of the ith second object.
In the above-mentioned objective function, the target function,as a preliminary objective function, β (| | G)u(Gu)T-I||2+||Ga(Ga)T-I||2) Is a first one of the regularization factors,is the second regularization factor.
It should be noted that, according to the above principle, constructing the above objective function can be regarded as a preprocessing step, and the objective function can be reused without repeatedly performing S101.
S102: and solving a preset target function to obtain a hash function.
Specifically, the hash vector of the first historical user and the hash vector of the second historical commodity, of which the similarity meets a preset condition (for example, is greater than a preset threshold), are used as independent variables of the objective function, so that the dependent variable of the objective function is optimal, and W, V, B and C are obtained. And W, V, B and C are substituted into the hash function which is the constraint condition of the target function to obtain the determined hash function.
S103: and calculating the hash vector of the new user and/or the hash vector of the new commodity by using the determined hash function.
Specifically, the attribute value of the new user may be input to a hash function to obtain a hash vector of the new user. And inputting the attribute value of the new commodity into a hash function to obtain a hash vector of the new commodity.
S104: and searching the commodities of which the similarity between the hash vector and the hash vector of the new user is greater than a preset threshold value as target commodities.
In this embodiment, the range of searching for commodities is not limited to new commodities, and may include historical commodities.
S105: and pushing the information of the target commodity to the new user.
As can be seen from the flow shown in fig. 1, according to the hash vectors of the history users and the history commodities which are matched (have similar hash vectors), a preset objective function is solved to obtain a hash function, the obtained hash function is used to calculate the hash vector of the new user, and information of the commodities having similar hash vectors is pushed to the new user. Since the hash function is obtained from the hash vectors of the matched historical users and historical commodities, the hash function can be considered to be obtained by learning from the matching situation of the historical users and the historical commodities, and the hash vector of the new user is obtained by using the hash function, the hash vector can be obtained for the new user, so that compared with the existing mode of randomly pushing advertisements to the new user, the information of the commodities pushed to the new user has higher pertinence (namely, matching performance) to the new user.
Furthermore, the target function integrates the inter-class constraint term and the intra-class constraint term, so that the pertinence of information push can be further improved.
The flow shown in fig. 1 will be exemplified as follows:
assuming that the movie is a commodity sample (i.e., a historical commodity), according to the attributes of the movie: gaThe attribute is associated with the hash value, wherein the attribute is associated with the hash value, and the mapping time is 80 years, type is action, director is wuyusen, country is china, label is meaning qi, and the attribute is associated with the hash value:
in the 80 s ═ 111 ], action ═ 1-1-11 ], wuyusen [ -11-1 … -1], china ═ 1-1-1 … 1], qi-meaning ═ 1-1-1 … 1,
the hash vector configured for the commodity sample is: ga＝[-1 1 1 1 -1 -1 1 -1 1 -1 … -1 1 -1-1 … 1 1 -1 -1 … 1]
Let G be the attribute information of the user sample (i.e., the historical user)uThe hash value is mapped as follows:
-11, age ═ 11-1-11-1, chinese ═ 1-1-1 … 1, like movie ═ 11, like action movie [ -111-1-1 … 1]
The hash vector configured for the user sample is: gu＝[-1 1 1 1 -1 -1 1 -1 1 -1 -1 … 1 11 -1 1 1 -1 -1 … 1]。
And (3) assuming that the similarity of the two hash vectors is greater than a preset threshold value, the user sample is matched with the commodity sample, namely the advertisement of the commodity recommended by the user is completely reasonable, the hash vectors of the two hash vectors are used as the independent variables of the objective function, and the parameters (W, V, B and C) in the objective function are obtained by Boosting learning by using the existing ensemble learning method to obtain the hash function.
And for the new user, solving through the hash function obtained by learning according to the attribute information of the new user to obtain a corresponding hash vector, and then carrying out corresponding similarity calculation. And setting a threshold value, and if the similarity of the hash vector between the new user and the commodity obtained by mapping according to the hash function is greater than the threshold value, pushing the advertisement of the commodity for the user.
That is to say, on one hand, the advertisement push problem based on the hash algorithm is modeled to obtain the target function, and the similarity between classes and the similarity in the classes are considered when the constraint condition is constructed, so that the hash function is more reasonable and effective in processing heterogeneous data. On the other hand, the constructed target function and the constraint condition are solved by adopting an ensemble learning algorithm boosting to obtain a hash function, then hash vector mapping is carried out on the user and the commodity, the similarity of the hash vectors between the user and the commodity is solved, and accurate commodity advertisement pushing is carried out on the user.
Although the history user is already configured with the hash vector, the history user may also use the obtained hash function to recalculate the hash vector, and similarly, the obtained hash function may also be used to calculate the hash vector of the new commodity or the history commodity, that is, the embodiment of the present application does not limit that only the hash vector of the new user is calculated by using the hash function.
Fig. 2(a), fig. 2(b) and fig. 2(c) are schematic diagrams of the effect of pushing advertisements to users in different scenarios (information of recommending historical commodities for new users, information of recommending new commodities for old users) and different hash vector lengths (i.e. 8, 16, 24, 32, 40, 48) by using the existing method (iterative update method and direct substitution method) and the method described in the present application (i.e. the Boost hash method), respectively, and it can be seen from these three diagrams that the Boost hash algorithm is significantly superior to the iterative update method and the direct substitution method because the two methods inevitably cause overfitting when the hash value increases, and the effect is very significantly reduced, and the Boost hash algorithm, due to the characteristic of strong ability of boosting generalization, can capture more characteristics of users and articles when the hash value increases, thus still maintaining a strong generalization ability.
TABLE 1 training time comparison
TABLE 2 comparison of test times
Table 1 compares the training times of the Boost hash algorithm, PAM algorithm and Pairwise algorithm:
the Pairwise algorithm has the longest time, and is very complex because the Pairwise algorithm takes each user advertisement score as a sample, and features are Pairwise combinations of user and advertisement features. The PAM algorithm has the lowest training time, and the training time of the Boost hash method is 126s, so that the time consumption is not very long.
Table 2 compares the test times of the three: the Boost hash algorithm has obvious advantages, the time consumption of other algorithms is basically more than 20 times of that of the Boost hash algorithm, and the conclusion fully proves that the pushing efficiency can be greatly improved by utilizing the hash method to recommend in cold starting.
Fig. 3 is an information pushing apparatus disclosed in an embodiment of the present application, including: the device comprises a hash function acquisition module, a calculation module and a push module.
The hash function obtaining module is used for solving a preset target function to obtain a hash function, and the target function takes the hash function as a constraint condition. The calculation module is configured to calculate a hash value of the first object and a hash vector of the second object using a hash function. The pushing module is used for pushing the information of the second object to the first object under the condition that the hash value of the first object and the hash value of the second object meet preset conditions.
Specifically, the construction principle of the objective function is as described above, and is not described herein again.
Optionally, the apparatus may further include an objective function constructing module, configured to construct an objective function according to the above principle.
Specifically, the hash function obtaining module solves a preset target function, and the specific implementation manner of obtaining the hash function is as follows: and taking the first historical hash vector and the second historical hash vector with the similarity meeting the preset condition as independent variables of the target function, optimizing dependent variables of the target function to obtain W, V, B and C, and determining the hash function as the target function, wherein the first historical hash vector is the hash vector of the first object, and the second historical hash vector is the hash vector of the second object.
Specifically, the specific implementation manner of the calculating module using the hash function to calculate the hash value of the first object and the hash vector of the second object is as follows: inputting the attribute value of the first object into the hash function to obtain a hash vector of the first object; and inputting the attribute value of the second object into the hash function to obtain a hash vector of the second object.
The information pushing device can recommend information for the new first object in a targeted mode, the similarity between the classes of the second object is considered, and the pertinence of the information recommended for the first object can be further improved.
The functions described in the method of the embodiment of the present application, if implemented in the form of software functional units and sold or used as independent products, may be stored in a storage medium readable by a computing device. Based on such understanding, part of the contribution to the prior art of the embodiments of the present application or part of the technical solution may be embodied in the form of a software product stored in a storage medium and including several instructions for causing a computing device (which may be a personal computer, a server, a mobile computing device or a network device) to execute all or part of the steps of the method described in the embodiments of the present application. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The embodiments are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same or similar parts among the embodiments are referred to each other.
The previous description of the disclosed embodiments is provided to enable any person skilled in the art to make or use the present application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other embodiments without departing from the spirit or scope of the application. Thus, the present application is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims (14)
1. An information pushing method, comprising:
solving a preset target function to obtain a hash function, wherein the target function takes the hash function as a constraint condition;
calculating a hash value of the first object and a hash vector of the second object using the hash function;
and under the condition that the hash value of the first object and the hash value of the second object meet preset conditions, pushing information of the second object to the first object.
2. The method of claim 1, wherein the arguments of the objective function are a historical hash vector of the first object and a historical hash vector of the second object;
the objective function includes at least:
a first regularization factor comprised of a historical hash vector of the first object and a historical hash vector of the second object;
the first regularization factor is used to characterize an intra-class constraint for objects of the same type.
3. The method of claim 2, wherein the objective function further comprises:
a second regularization factor comprised of a sum of bits in the historical hash vector of the first object and a sum of bits in the historical hash vector of the second object;
the second regularization factor is used to characterize inter-class constraints for objects of different types.
4. The method of claim 2, wherein the objective function further comprises:
a preliminary objective function consisting of a similarity of the historical hash vector of the first object and the historical hash vector of the second object, and a satisfaction of the first object with the second object.
5. The method according to any of claims 2-4, wherein the objective function is:
wherein h isijFor the satisfaction of the ith first object with the jth second object,historical hash vector for ith first objectHistorical hash vector with jth second objectThe degree of similarity of (a) to (b),for the k-th bit in the history hash vector of the ith first object,is the k bit, G, in the history hash vector of the jth second objectuHistorical hash vector for ith first object, GaIs the historical hash vector of the jth second object, xiIs an attribute of the ith first object, xjλ, β are preset coefficients for the attribute of the jth second object.
6. The method of claim 5, wherein solving the predetermined objective function to obtain the hash function comprises:
and taking a first historical hash vector and a second historical hash vector with the similarity meeting the preset condition as independent variables of the target function, optimizing dependent variables of the target function to obtain W, V, B and C, and determining the hash function as the target function, wherein the first historical hash vector is a hash vector of a first object, and the second historical hash vector is a hash vector of a second object.
7. The method of claim 1, wherein computing the hash value of the first object and the hash vector of the second object using the hash function comprises:
inputting the attribute value of the first object into the hash function to obtain a hash vector of the first object;
and inputting the attribute value of the second object into the hash function to obtain a hash vector of the second object.
8. An information pushing apparatus, comprising:
the hash function acquisition module is used for solving a preset target function to obtain a hash function, and the target function takes the hash function as a constraint condition;
a calculation module for calculating a hash value of a first object and a hash vector of a second object using the hash function;
and the pushing module is used for pushing the information of the second object to the first object under the condition that the hash value of the first object and the hash value of the second object meet preset conditions.
9. The apparatus of claim 8, wherein the arguments of the objective function are a historical hash vector of the first object and a historical hash vector of the second object;
the objective function includes at least:
a first regularization factor comprised of a historical hash vector of the first object and a historical hash vector of the second object;
the first regularization factor is used to characterize an intra-class constraint for objects of the same type.
10. The apparatus of claim 9, wherein the objective function further comprises:
a second regularization factor comprised of a sum of bits in the historical hash vector of the first object and a sum of bits in the historical hash vector of the second object;
the second regularization factor is used to characterize inter-class constraints for objects of different types.
11. The apparatus of claim 9, wherein the objective function further comprises:
a preliminary objective function consisting of a similarity of the historical hash vector of the first object and the historical hash vector of the second object, and a satisfaction of the first object with the second object.
12. The apparatus according to any of claims 9-10, wherein the objective function is:
wherein h isijFor the satisfaction of the ith first object with the jth second object,historical hash vector for ith first objectHistorical hash vector with jth second objectThe degree of similarity of (a) to (b),for the k-th bit in the history hash vector of the ith first object,is the k bit, G, in the history hash vector of the jth second objectuHistorical hash vector for ith first object, GaIs the historical hash vector of the jth second object, xiIs an attribute of the ith first object, xjλ, β are preset coefficients for the attribute of the ith second object.
13. The apparatus of claim 12, wherein the hash function obtaining module is configured to solve a preset objective function to obtain a hash function, and includes:
and taking a first historical hash vector and a second historical hash vector with the similarity meeting the preset condition as independent variables of the target function, optimizing dependent variables of the target function to obtain W, V, B and C, and determining the hash function as the target function, wherein the first historical hash vector is a hash vector of a first object, and the second historical hash vector is a hash vector of a second object.
14. The apparatus of claim 8, wherein the computing module, configured to compute the hash value of the first object and the hash vector of the second object using the hash function, comprises:
the calculation module is specifically configured to input the attribute value of the first object into the hash function to obtain a hash vector of the first object; and inputting the attribute value of the second object into the hash function to obtain a hash vector of the second object.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910184486.1A CN109934629A (en) | 2019-03-12 | 2019-03-12 | A kind of information-pushing method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910184486.1A CN109934629A (en) | 2019-03-12 | 2019-03-12 | A kind of information-pushing method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109934629A true CN109934629A (en) | 2019-06-25 |
Family
ID=66987020
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910184486.1A Pending CN109934629A (en) | 2019-03-12 | 2019-03-12 | A kind of information-pushing method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109934629A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113961805A (en) * | 2021-10-18 | 2022-01-21 | 珠海格力电器股份有限公司 | Method and device for pushing presentation object, electronic equipment and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102890713A (en) * | 2012-09-20 | 2013-01-23 | 浙江大学 | Music recommending method based on current geographical position and physical environment of user |
| CN103605723A (en) * | 2013-11-15 | 2014-02-26 | 南京云川信息技术有限公司 | Video recommending method based on particle swarm algorithm |
| CN104424254A (en) * | 2013-08-28 | 2015-03-18 | 阿里巴巴集团控股有限公司 | Method and device for obtaining similar object set and providing similar object set |
| CN105809462A (en) * | 2014-12-30 | 2016-07-27 | Tcl集团股份有限公司 | Estimation method and device for estimating advertisement click rate |
| CN105894310A (en) * | 2014-10-15 | 2016-08-24 | 祁勇 | Personalized recommendation method |
| CN106060762A (en) * | 2016-05-24 | 2016-10-26 | 中国联合网络通信集团有限公司 | Information pushing method, information pushing device and information pushing system |
| CN106599022A (en) * | 2016-11-01 | 2017-04-26 | 中山大学 | User portrait forming method based on user access data |
| CN107492008A (en) * | 2017-08-09 | 2017-12-19 | 阿里巴巴集团控股有限公司 | Information recommendation method, device, server and computer-readable storage medium |
-
2019
- 2019-03-12 CN CN201910184486.1A patent/CN109934629A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102890713A (en) * | 2012-09-20 | 2013-01-23 | 浙江大学 | Music recommending method based on current geographical position and physical environment of user |
| CN104424254A (en) * | 2013-08-28 | 2015-03-18 | 阿里巴巴集团控股有限公司 | Method and device for obtaining similar object set and providing similar object set |
| CN103605723A (en) * | 2013-11-15 | 2014-02-26 | 南京云川信息技术有限公司 | Video recommending method based on particle swarm algorithm |
| CN105894310A (en) * | 2014-10-15 | 2016-08-24 | 祁勇 | Personalized recommendation method |
| CN105809462A (en) * | 2014-12-30 | 2016-07-27 | Tcl集团股份有限公司 | Estimation method and device for estimating advertisement click rate |
| CN106060762A (en) * | 2016-05-24 | 2016-10-26 | 中国联合网络通信集团有限公司 | Information pushing method, information pushing device and information pushing system |
| CN106599022A (en) * | 2016-11-01 | 2017-04-26 | 中山大学 | User portrait forming method based on user access data |
| CN107492008A (en) * | 2017-08-09 | 2017-12-19 | 阿里巴巴集团控股有限公司 | Information recommendation method, device, server and computer-readable storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 陈敏: "计算广告中冷启动问题的研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113961805A (en) * | 2021-10-18 | 2022-01-21 | 珠海格力电器股份有限公司 | Method and device for pushing presentation object, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Sun et al. | Learning multiple-question decision trees for cold-start recommendation | |
| CN105701191B (en) | Pushed information click rate estimation method and device | |
| Panniello et al. | Experimental comparison of pre-vs. post-filtering approaches in context-aware recommender systems | |
| CN109242612B (en) | Product recommendation method and device | |
| Christakopoulou et al. | Hoslim: Higher-order sparse linear method for top-n recommender systems | |
| CN111815415A (en) | Commodity recommendation method, system and equipment | |
| CN107644036B (en) | Method, device and system for pushing data object | |
| Bhagat et al. | Recommending with an agenda: Active learning of private attributes using matrix factorization | |
| EP2377080A1 (en) | Machine optimization devices, methods, and systems | |
| US9269049B2 (en) | Methods, apparatus, and systems for using a reduced attribute vector of panel data to determine an attribute of a user | |
| CN113268656A (en) | User recommendation method and device, electronic equipment and computer storage medium | |
| Xu et al. | Personalized product recommendation method for analyzing user behavior using DeepFM | |
| Qian et al. | Community-based user domain model collaborative recommendation algorithm | |
| KR101639656B1 (en) | Method and server apparatus for advertising | |
| Wang et al. | Customized products recommendation based on probabilistic relevance model | |
| Wang et al. | Time series clustering based on sparse subspace clustering algorithm and its application to daily box-office data analysis | |
| Maldonado et al. | Embedded heterogeneous feature selection for conjoint analysis: A SVM approach using L1 penalty | |
| CN111339435A (en) | Matrix decomposition completion hybrid recommendation method based on potential factors | |
| Zhang et al. | Local low-rank matrix approximation with preference selection of anchor points | |
| CA3111094A1 (en) | Noise contrastive estimation for collaborative filtering | |
| Sengupta et al. | Simple surveys: Response retrieval inspired by recommendation systems | |
| CN109934629A (en) | A kind of information-pushing method and device | |
| Chen et al. | Precision marketing for financial industry using a PU-learning recommendation method | |
| Choi et al. | Extended collaborative filtering technique for mitigating the sparsity problem | |
| Compiani et al. | Demand estimation with text and image data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190625 |
|
| RJ01 | Rejection of invention patent application after publication |