CN109409928A - A kind of material recommended method, device, storage medium, terminal - Google Patents
A kind of material recommended method, device, storage medium, terminal Download PDFInfo
- Publication number
- CN109409928A CN109409928A CN201811073786.4A CN201811073786A CN109409928A CN 109409928 A CN109409928 A CN 109409928A CN 201811073786 A CN201811073786 A CN 201811073786A CN 109409928 A CN109409928 A CN 109409928A
- Authority
- CN
- China
- Prior art keywords
- class
- subscriber
- user
- style
- style 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.)
- Withdrawn
Links
- 239000000463 material Substances 0.000 title claims abstract description 367
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000003860 storage Methods 0.000 title claims abstract description 11
- 239000013598 vector Substances 0.000 claims abstract description 129
- 241001269238 Data Species 0.000 claims description 10
- 230000001186 cumulative effect Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 description 8
- 101150004367 Il4i1 gene Proteins 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 208000001491 myopia Diseases 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 4
- 230000006399 behavior Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 239000002699 waste material Substances 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005304 joining Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 239000011435 rock Substances 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
- G06Q30/0269—Targeted advertisements based on user profile or attribute
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0277—Online advertisement
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0273—Determination of fees for advertising
- G06Q30/0275—Auctions
Abstract
The present invention relates to a kind of material recommended method, device, storage mediums, terminal.The material recommended method includes: to determine at least one class of subscriber according to the historical traffic data of multiple users；Count the respective typical material of each class of subscriber, and the style vector of each class of subscriber is determined according to the style vector of the typical material of each class of subscriber, show the material that number is greater than preset threshold in the historical traffic data of typical case's material selected from each class of subscriber；Class of subscriber belonging to the active user is determined according to the real-time traffic data of active user；Deposit material in material database is ranked up according to the similarity of its style vector and the style vector of class of subscriber belonging to the active user, to determine candidate material according to ranking results；The candidate material is recommended into the active user.Technical solution of the present invention can more accurately recommend material.
Description
Technical field
The present invention relates to Internet advertising distribution technology field more particularly to a kind of material recommended method, device, storage Jie
Matter, terminal.
Background technique
In current Internet application, accurate recommended technology is usually used, such as the recommendation of advertisement, content push away
Recommend etc..By taking advertisement is recommended as an example, in Internet advertising industry, real time bid (Real Time Bidding, abbreviation RTB)
It is that gradually mode is selected and purchased in a kind of advertisement of prevalence in recent years.It is different that mode is selected and purchased from traditional contract advertisement, and RTB allows advertisement
Material publisher will do it for the display machine of each ad material and bid, to be bought with crowd instead of ad material
Show position purchase.
On this basis, party in request's platform (Demand Side Platform, abbreviation DSP) is used as ad material publisher
Agency, in order to assist ad material publisher complete RTB operation, need in 100ms to each ad material posting request
Decision is carried out, to judge whether to participate in bid, and how much is bid if participating in bidding.And ad material publisher would generally
It is to refer to click-through-rate (Click Through Rate, abbreviation CTR) and conversion ratio (Conversion Rate, abbreviation CVR)
The agency that mark investigates DSP launches effect.
Current DSP is usually higher by clicking rate in historical traffic or conversion ratio when to user's recommended advertisements material
Material recommend user, can not according to the material style that user is liked come to its recommend material；In addition, for material database
In the new material never launched, it usually needs first tentative is delivered to user, according to the click condition of feedback, then adjusts again
It is whole may the interested dispensing crowd of style to new material, cause to launch increased costs.
The style how to be liked according to user more accurately recommends material to user, is a problem to be solved.
Summary of the invention
Present invention solves the technical problem that being how more accurately to recommend material.
In order to solve the above technical problems, including the following steps: root the embodiment of the invention provides a kind of material recommended method
According to the historical traffic data of multiple users, at least one class of subscriber is determined；Count the respective typical case of each class of subscriber
Material, and determine according to the style vector of the typical material of each class of subscriber the style vector of each class of subscriber, the allusion quotation
Type material shows the material that number is greater than preset threshold, typical case's material in the historical traffic data selected from each class of subscriber
Style vector be to be determined based on predefined multiple style dimensions；According to the determination of the real-time traffic data of active user
Class of subscriber belonging to active user；By the deposit material in material database according to belonging to its style vector and the active user
The similarity of the style vector of class of subscriber is ranked up, to determine candidate material according to ranking results；By the candidate material
Recommend the active user.
Optionally, according to the historical traffic data of multiple users, determine that at least one class of subscriber includes: from the multiple
Data centered on the historical traffic data of K user of selection in the historical traffic data of user, wherein each centre data is equal
A corresponding class of subscriber, K is positive integer；According to the historical traffic data of the multiple user and a centre data of K
The multiple user is divided into K class of subscriber by similarity.
It optionally, will be described according to the similarity of the historical traffic data of the multiple user and the K centre datas
Multiple users are divided into the phase that K class of subscriber includes: the historical traffic data and the K centre datas of more each user
Like degree, record and each highest centre data of user's similarity；Each user is subdivided into the highest middle calculation of similarity
According in corresponding class of subscriber.
Optionally, the respective typical material of each class of subscriber is counted, and according to typical case's element of each class of subscriber
The style vector of material determines that the style vector of each class of subscriber includes: statistics in the historical traffic data of each class of subscriber
Show the material that number is greater than preset threshold, and material of the number greater than preset threshold will be showed and arranged from high to low according to clicking rate
Sequence chooses the n material that clicking rate sorts forward and is used as typical material, wherein n is positive integer；For each class of subscriber, divide
The style vector of wherein each typical material is not calculated, and the class of subscriber is calculated according to the style vector of each typical material
Style vector.
Optionally, calculating the style vector of the class of subscriber according to the style vector of each typical material includes: to calculate
In the class of subscriber style vector of each typical case's material cumulative and, to obtain the style vector of the class of subscriber.
Optionally, by the deposit material in material database according to class of subscriber belonging to its style vector and the active user
The similarity of style vector be ranked up, with according to ranking results determine candidate material include: according to similarity from high to low
The deposit material is ranked up, chooses the forward m material that sort as candidate material.
In order to solve the above technical problems, the embodiment of the invention also provides a kind of ad material recommendation apparatus, including user
Cluster module determines at least one class of subscriber suitable for the historical traffic data according to multiple users；Material style counts mould
Block is suitable for counting the respective typical material of each class of subscriber, and the style of the typical material according to each class of subscriber
Vector determines the style vector of each class of subscriber, opens up in the historical traffic data of typical case's material selected from each class of subscriber
Occurrence number is greater than the material of preset threshold, and the style vector of typical case's material is determined based on predefined multiple style dimensions
's；Real-time traffic determining module determines user belonging to the active user suitable for the real-time traffic data according to active user
Classification；Material sorting module, suitable for by the deposit material in material database according to belonging to its style vector and the active user
The similarity of the style vector of class of subscriber is ranked up, to determine candidate material according to ranking results；Material recommending module is fitted
In the candidate material is recommended the active user.
Optionally, user's cluster module includes: that cluster centre chooses submodule, suitable for going through from the multiple user
Data centered on the historical traffic data of K user of selection in history data on flows, wherein each centre data is one corresponding
Class of subscriber, K are positive integer；User divides submodule, a described suitable for the historical traffic data and K according to the multiple user
The multiple user is divided into K class of subscriber by the similarity of centre data.
Optionally, it includes: comparing unit that the user, which divides submodule, is adapted to compare the historical traffic data of each user
With the similarity of the K centre datas, record and each highest centre data of user's similarity；Division unit is suitable for
Each user is subdivided into the corresponding class of subscriber of the highest centre data of similarity.
Optionally, the material style statistical module includes: typical story extraction submodule, is counted in each class of subscriber
Historical traffic data in show number be greater than preset threshold material, and by show number greater than preset threshold material according to
Clicking rate sorts from high to low, chooses the n material that clicking rate sorts forward and is used as typical material, wherein n is positive integer；Element
Material style extracting sub-module, for each class of subscriber, suitable for calculating separately the style vector of wherein each typical material, and root
The style vector of the class of subscriber is calculated according to the style vector of each typical material.
Optionally, the material style extracting sub-module includes: summing elements, is suitable for calculating each in the class of subscriber
The style vector of typical material cumulative and, to obtain the style vector of the class of subscriber.
Optionally, the material sorting module includes: sorting sub-module, is suitable for according to similarity from high to low to the storage
Standby material is ranked up, and chooses the forward m material that sort as candidate material, wherein m is positive integer.
In order to solve the above technical problems, being stored thereon with computer the embodiment of the invention also provides a kind of storage medium
Instruction, when the computer instruction is run the step of material recommended method described in execution.
In order to solve the above technical problems, the embodiment of the invention also provides a kind of terminal, including memory and processor, institute
The computer instruction for being stored with and capable of being run on memory on the processor is stated, the processor runs the computer and refers to
The step of material recommended method is executed when enabling.
Compared with prior art, the technical solution of the embodiment of the present invention has the advantages that
The material recommended method of technical solution of the present invention can determine at least one according to the historical traffic data of multiple users
A class of subscriber；Then, the respective typical material of each class of subscriber is counted, and according to typical case's element of each class of subscriber
The style vector of material determines that the style vector of each class of subscriber, typical case's material are selected from the historical traffic of each class of subscriber
Show the material that number is greater than preset threshold in data, the style vector of typical case's material is based on predefined multiple styles
What dimension determined；Next, can be determined according to the real-time traffic data of active user described current when this method application on site
Class of subscriber belonging to user；Then by the deposit material in material database according to belonging to its style vector and the active user
The similarity of the style vector of class of subscriber is ranked up, to determine candidate material according to ranking results；Finally again by the time
Material is selected to recommend the active user.Thus, it is possible to recommend material to it according to the style that user likes, be conducive to improve element
The clicking rate of material.For advertisement recommendation, conversion ratio is also advantageously improved, is obtained more preferably to realize with less cost
Ad material launches effect.For content (such as news, short-sighted frequency etc.) recommendation, it is more accurate to recommend, and has
Conducive to improvement user experience.
Further, technical solution of the present invention can choose K user's from the historical traffic data of the multiple user
Data centered on historical traffic data, then according to the phase of the historical traffic data of the user and the K centre datas
Like degree, the multiple user is divided into K class of subscriber.Classified as a result, using cluster mode to multiple users, just
It is accurately analyzed in the subsequent style to class of subscriber.
Further, technical solution of the present invention can count that show number in the historical traffic data of each class of subscriber big
In the material of preset threshold, and material of the number greater than preset threshold will be showed and sorted from high to low according to clicking rate, selected point
The n material that the rate of hitting sorts forward is as typical material, next, calculating separately wherein each allusion quotation for each class of subscriber
The style vector of type material, and calculate according to the style vector of each typical material the style vector of the class of subscriber.Due to
The high material of clicking rate can the material lower than clicking rate more accurately reflect the style that associated user likes, therefore,
The high material of clicking rate is used as typical material, may improve the accuracy calculated, thus when being conducive to increase the recommendation of subsequent material
Precision.
Further, technical solution of the present invention can be by the deposit material in material database according to its style vector and described current
The similarity of the style vector of class of subscriber belonging to user is ranked up from high to low, is chosen the forward m material that sort and is made
For candidate material.Accordingly, for the material never launched newly being added in material database, can prejudge may feel emerging to it
The user of interest, prevents the cold start-up of new material, effectively prevents the waste that material launches cost.
Detailed description of the invention
Fig. 1 is a kind of flow chart of material recommended method of the embodiment of the present invention；
Fig. 2 is a kind of flow chart of specific embodiment of step S11 in Fig. 1；
Fig. 3 is a kind of flow chart of specific embodiment of step S12 in Fig. 1；
Fig. 4 is a kind of structural schematic diagram of material recommendation apparatus of the embodiment of the present invention；
Fig. 5 is the structural schematic diagram of user's cluster module in Fig. 4；
Fig. 6 is the structural schematic diagram of material style statistical module in Fig. 4.
Specific embodiment
It will be understood by those skilled in the art that current DSP is when to user's recommended advertisements material, usually by history stream
Clicking rate or the higher material of conversion ratio recommend user in amount, can not push away according to the material style that user is liked to it
Recommend material；In addition, for the new material never launched in material database, it usually needs first tentative is delivered to user, according to
The click condition of feedback, then adjust again may the interested dispensing crowd of style to new material, cause to launch increased costs.
Material recommended method in the embodiment of the present invention can determine at least one according to the historical traffic data of multiple users
A class of subscriber；Then, the respective typical material of each class of subscriber is counted, and according to typical case's element of each class of subscriber
The style vector of material determines that the style vector of each class of subscriber, typical case's material are selected from the historical traffic of each class of subscriber
Show the material that number is greater than preset threshold in data, the style vector of typical case's material is based on predefined multiple styles
What dimension determined；Next, can be determined according to the real-time traffic data of active user described current when this method application on site
Class of subscriber belonging to user；Then by the deposit material in material database according to belonging to its style vector and the active user
The similarity of the style vector of class of subscriber is ranked up, to determine candidate material according to ranking results；Finally again by the time
Material is selected to recommend the active user.Thus, it is possible to recommend material to it according to the style that user likes, be conducive to improve element
The clicking rate of material.For advertisement recommendation, conversion ratio is also advantageously improved, is obtained more preferably to realize with less cost
Ad material launches effect.For content (such as news, short-sighted frequency etc.) recommendation, it is more accurate to recommend, and has
Conducive to improvement user experience.
It is understandable to enable above-mentioned purpose of the invention, feature and beneficial effect to become apparent, with reference to the accompanying drawing to this
The specific embodiment of invention is described in detail.Obviously, described embodiments are only a part of the embodiments of the present invention, without
It is whole embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not making creative work
Under the premise of every other embodiment obtained, belong to protection scope of the present invention.
Fig. 1 is a kind of flow chart of material recommended method of the embodiment of the present invention.
Material recommended method in the present embodiment can be applied to advertisement recommendation, such as assisted by party in request platform DSP
Ad material publisher uses when executing RTB operation, to improve the clicking rate and conversion ratio of ad material.
Referring to FIG. 1, the material recommended method may include steps of:
Step S11: according to the historical traffic data of multiple users, at least one class of subscriber is determined；
Step S12: the respective typical material of statistics each class of subscriber, and according to typical case's element of each class of subscriber
The style vector of material determines that the style vector of each class of subscriber, typical case's material are selected from the historical traffic of each class of subscriber
Show the material that number is greater than preset threshold in data, the style vector of typical case's material is based on predefined multiple styles
What dimension determined；
Step S13: class of subscriber belonging to the active user is determined according to the real-time traffic data of active user；
Step S14: by the deposit material in material database according to user class belonging to its style vector and the active user
The similarity of other style vector is ranked up, to determine candidate material according to ranking results；
Step S15: the candidate material is recommended into the active user.
In the present embodiment, the material can be picture, may include advertising information in the picture, such as: quotient for sale
Product information.
Specifically, the format of the picture can be the formats such as jpg, jpeg, tif, gif, bmp, the embodiment of the present invention pair
This is with no restrictions.
Further, the historical traffic data may include identification information (Identification, the abbreviation of equipment
ID), hereinafter referred to as device id.As a non-limiting embodiment, can be associated with by the device id user and its
The internet access trace that internet behavior generates.Alternatively, for being based on PC (Personal Computer, abbreviation PC)
The user's internet access trace generated is held, the user can also be associated with according to the ID for the browser installed on the end PC.
Preferably, internet access trace associated with the user may include the user in material publisher
And/or on the website of material display platform browsing record, it is described browsing record in may include browse duration, browsing it is specific
Material in webpage etc..Wherein, the material publisher can refer to the provider of the material to be released；The material exhibition
The platform for providing for the material to be presented and showing position can be referred to by showing platform.
In addition, the internet access trace that the internet behavior based on the user generates, not only available user is browsed
Specific webpage in material information, the basic information of the user can also be obtained.
Specifically, the basic information of the user may include the city of residence of the user, age, gender, purchasing power
Etc. information.
Preferably, the basic information of the user can collected from the material publisher (and/or material display platform,
Such as website, webpage), it can also directly acquire from the historical internet of the user and access data.For example, user can be based on
The data provided to the material publisher and/or material display platform directly acquire the basic information of the user；In another example also
It can be based on the user in the material publisher and/or the internet access trace of material display platform, thus it is speculated that obtain institute
State the basic information of user.Wherein, thus it is speculated that operation can be executed by the material publisher and/or material display platform, or
Person can also be executed as the terminal (such as DSP) for executing scheme described in the present embodiment.
Further, the DSP can at least one pre-recorded user basic information and associated device id,
After receiving real-time traffic data, can based on the device id in the data on flows from prestore record in search corresponding base
Then plinth information carries out category division to active user according to the basic information.
When specific execution step S11, user can be divided by clustering algorithm by different class of subscribers.
Specifically, the clustering algorithm in the present embodiment can using partitioning algorithm (such as: K-means algorithm), by different level
Algorithm (such as: ROCK algorithm), Name-based Routing (such as: DBSCAN algorithm), based on grid algorithm (such as: STING
Algorithm) etc..Those skilled in the art can according to specific application the suitable clustering algorithm of selection of adaptability, this hair
Bright embodiment is without limitation.
Fig. 2 is referred to, Fig. 2 is a kind of flow chart of specific embodiment of step S11 in Fig. 1.
In a non-limiting embodiment, such as: user is divided into using K-means algorithm by different user class
Not, it may include steps of:
Step S21: from the historical traffic data conduct for choosing K user in the historical traffic data of the multiple user
Calculation evidence, wherein each centre data corresponds to a class of subscriber, and K is positive integer；
Step S22:, will be described more according to the similarity of the historical traffic data of the user and the K centre datas
A user is divided into K class of subscriber.
In the present embodiment, the historical traffic data of the user can be subjected to digitized processing, each historical traffic
Data all can serve as the mark of a user, below specifically the holding to step S21 and step S22 with a concrete application scene
Row process is described in detail.
In the concrete application scene, by taking one shares 100 users as an example, need 100 users being divided into 5 use
Family classification, the user group in each class of subscriber have similar purchase preference and information browse preference, different user classification
In user group have different purchase preference and information browse preference.Diversity factor between different user classification is bigger, then
Presentation class is more clear, and classifying quality is more excellent.
During machine learning, first with number a1, a2, a3, a4, a5, a6, a7, a8 ... a100 respectively as
The mark of this 100 users, this 100 data are exactly historical traffic data corresponding with the user.When initial launch, K-
Means algorithm can randomly choose 5 historical traffic datas, as the centre data of 5 class of subscribers, for example, a4, a26, a59,
a73、a89。
Next, needing to compare the historical traffic data of 100 users and the similarity of 5 centre datas, record
Then 100 users are subdivided into the highest centre data pair of similarity by lower and each highest centre data of user's similarity
In the class of subscriber answered.
Specifically, for historical traffic data a1, the distance of a1 to a4, a26, a59, a73, a89 can be calculated separately.Away from
From closer, then similarity is higher, and in other words, distance is remoter, then similarity is lower.If the distance of a1 to a4 is nearest, can incite somebody to action
A1 incorporates the class of subscriber where centre data a4 into.Next, can calculate separately in the same way a2, a3, a4, a5,
A6, a7, a8 ... a100 to each centre data distance, and successively by a2, a3, a4, a5, a6, a7, a8 ... a100 represent
User be respectively subdivided into 5 different class of subscribers.
If for the first time divide after the completion of, using the a4 user that it includes as the class of subscriber of center data collect be combined into a1, a2,
A3, a4, a5, a6, a7, a8 }, then continue to calculate all data in user's set { a1, a2, a3, a4, a5, a6, a7, a8 }
Average, average is denoted as b1, i.e. b1=(a1+a2+a3+a4+a5+a6+a7+a8)/8, b1 is used to replace a4 as the use
New centre data in the classification of family.The new centre data of other four class of subscribers is calculated separately out in the same way.
After the completion of calculating, the centre data of five class of subscribers can be updated to by a4, a26, a59, a73, a89 b1, b2, b3, b4,
b5。
The new middle calculation of a100 to five classification next, calculating separately a1, a2, a3, a4, a5, a6, a7, a8 ...
According to the distance of b1, b2, b3, b4, b5, by use representated by above-mentioned data a1, a2, a3, a4, a5, a6, a7, a8 ... a100
Family carries out repartitioning classification.For each user set after repartitioning, continue to calculate its new centre data, and root
All users are divided again according to new centre data.
After successive ignition calculates, the centre data of relatively stable each class of subscriber, i.e., each user can be obtained
The centre data of classification no longer changes with the number of iterations.At this point it is possible to think that class of subscriber division terminates.
It should be noted that the quantity of the class of subscriber can be according to how much progress of specific requirements and number of users
Setting, for example, it may be 5 classifications, 7 classifications, 12 classifications etc., the embodiment of the present invention is without limitation.
After the category division for completing multiple users, the respective typical material of each class of subscriber can be counted, and
The style vector of each class of subscriber is determined according to the style vector of the typical material of each class of subscriber.
Referring to FIG. 3, described Fig. 3 is a kind of flow chart of specific embodiment of step S12.
May include following subdivided step in specific implementation step S12:
Step S31: statistics shows the material that number is greater than preset threshold in the historical traffic data of each class of subscriber,
And material of the number greater than preset threshold will be showed and sorted from high to low according to clicking rate, it chooses clicking rate and sorts forward n
Material is as typical material, wherein n is positive integer；
Step S32: for each class of subscriber, the style vector of wherein each typical material is calculated separately, and according to each
The style vector of a typical case's material calculates the style vector of the class of subscriber.
Further, the style vector of the class of subscriber can be the style of each typical case's material in the class of subscriber
Vector cumulative and.
Specifically, the style vector of the typical material is determined based on predefined multiple style dimensions.
In a non-limiting embodiment, the quantity of the style dimension can be 2, respectively pseudo-classic style and
Korean style style,.For example, if some material is pertaining only to pseudo-classic style, style vector can be expressed as (1,0), if some is plain
Material is pertaining only to Korean style style, then its style vector can be expressed as (0,1).Wherein, the number 1 in style vector is represented in two dimension
In dimension, some material has some style, and the number 0 in style vector represents in two-dimentional dimension, some material does not have
Some style.
In another non-limiting embodiment, the quantity of the style dimension can be more, such as can be
Pseudo-classic style, Korean style style and punk.For example, style vector can be with table if some material is pertaining only to pseudo-classic style
It is shown as (1,0,0), if some material is pertaining only to punk, style vector can be expressed as (0,0,1), if some material
Belong to pseudo-classic style and punk simultaneously, then its style vector can be expressed as (1,0,1).Wherein, the number in style vector
Word 1 represents in three dimensions, some material has some style, and the number 0 in style vector represents in three dimensions, certain
A material does not have some style.
It can be seen that for different typical materials, style vector is different.Thus, it is possible to by typical material
Style is digitized, and subsequent calculating and classification are convenient for.
It will be understood by those skilled in the art that in order to more accurately identify typical material, also can choose four dimension words to
The digital vectors of amount, five dimension word vectors or more, the embodiment of the present invention are without limitation.
In specific implementation step S31, each class of subscriber can regard a user group as, in the user group
The material that the historical traffic data of all users is showed has similar style.
On this basis, it can count and show number in the historical traffic data of each user group greater than preset threshold
Material, wherein the preset threshold can be set to 1000,2000,5000 etc..Preset threshold is higher, shows number and is greater than in advance
If the material of threshold value is fewer, it is possible to which the style that cannot represent the user group comprehensively causes to calculate inaccuracy；Preset threshold is got over
It is low, show number greater than preset threshold material it is more, cause calculate when noise it is excessive, also will affect the accurate of calculated result
Property.Those skilled in the art can reasonably select preset threshold according to specific requirements and considering for precision of calculating, the present invention
Embodiment is without limitation.
It sorts from high to low next, material of the number greater than preset threshold can will be showed according to clicking rate, selected point
For the n material that the rate of hitting sorts forward as typical material, n can be with value for 200,100,50 etc..N value can be seen as finally joining
With the numerical value of the sample value of calculating, n it is excessive or it is too small also will affect whole sample representativeness and calculate accuracy.This field
Technical staff can reasonably select the specific value of n, the embodiment of the present invention according to specific requirements and considering for precision of calculating
It is without limitation.
In specific implementation step S32, the VGG19 convolutional neural networks based on the training of ImageNet data set can be selected
" block2_conv1 " of (Convolutional Neural Networks, abbreviation CNN) model, " block3_conv1 ",
" block4_conv1 ", the layers such as " block5_conv1 " export result and characterize as the content characteristic of material image
(FeatureMap)。
Feature Map=VGG19 (IMG)
Image Style=Concat (Flatten (Graml))
Wherein,The matrix in Feature Map exported for l layers of VGG19,0 < i <=l,For VGG19
Another matrix in the Feature Map of l layers of output, 0 < j <=l；It indicates to pass through vectorization characteristic spectrum i
The style vector of the typical material generated with the inner product of j, GramlIndicate the sum of l layers of all style vectors, Image Style
Indicate style vector final as obtained from each layer style vector after connection flattening-out.
The style vector of the class of subscriber can be calculated by following formula:
Wherein, M is the style vector of class of subscriber, ctr 'h*Image StylehFor h-th of typical material style to
Amount, n are that the clicking rate selected sorts the quantity of forward material in step S31, ctr 'hRepresent the clicking rate of h-th of typical material
Rate value is obtained after weight proportion is handled, clicking rate ctr is higher, then ctr 'hValue it is bigger, for example, for clicking rate
The respectively material of A, B, C can be expressed as A/ (A+B+C), B/ (A+B+C), C/ (A+B+ respectively after weight proportion is handled
C)。
Certainly, in specific calculate, the specific value of clicking rate can also be directly utilized according to different applications
The calculating for carrying out style vector, without obtaining rate value using after weight proportion is handled.
The present embodiment can examine the weight of the clicking rate of each typical material when calculating the style vector of class of subscriber
Including worry, the higher material of clicking rate occupies larger specific gravity, and the lower material of clicking rate occupies lesser specific gravity.Due to clicking rate
High material can the material lower than clicking rate more accurately reflect the style that associated user likes, thus, it is possible to effectively
The accuracy calculated is improved, to be conducive to increase precision when subsequent material is recommended.
With continued reference to FIG. 1, next, execution step S13 can be according to the reality for the real-time traffic detected
When data on flows determine class of subscriber belonging to active user.
When it is implemented, the internet access behavior of the user can generate when user's displaying live view a certain webpage
Real-time traffic data.For example, when user accesses material display platform (such as webpage), if there is an ad material on the webpage
Show position, website (i.e. background server) belonging to the webpage is notified that DSP someone is browsing the webpage, and whether inquires DSP
Need to bid the displaying position of the ad material.The DSP can use the present embodiment technical solution, according to active user phase
Associated device id data or active user basic information data come calculate and each class of subscriber in centre data between
Similarity, and according to highest similarity determine active user belonging to particular user classification.
After determining specific class of subscriber for active user, step S14 can be continued to execute, by the deposit element in material database
Material is ranked up according to the similarity of its style vector and the style vector of class of subscriber belonging to the active user, with basis
Ranking results determine candidate material.
It is possible to further be ranked up from high to low to the deposit material according to similarity, the forward m that sorts is chosen
Zhang Sucai is as candidate material.
In a concrete application scene, such as: class of subscriber shares 5, then the wind direction vector of class of subscriber shares 5
(such as: H1, H2, H3, H4, H5) has 7 deposit materials in material database, this 7 deposit materials share 7 style vector (examples
Such as: Fig1, Fig2, Fig3, Fig4, Fig5, Fig6, Fig7).
7 deposit materials its style vectors is respectively compared with the similarity of the style vector of 5 class of subscribers is in table 1
It is as follows:
Table 1
| H1 | H2 | H3 | H4 | H5 | |
| Fig1 | 1.2 | 1.9 | 3 | 6 | 0.4 |
| Fig2 | 1.5 | 3.6 | 4.7 | 2 | 1 |
| Fig3 | 0.8 | 4.4 | 7 | 5 | 2.6 |
| Fig4 | 3 | 6 | 1 | 3.5 | 4.1 |
| Fig5 | 2 | 2.8 | 5 | 0.6 | 3.2 |
| Fig6 | 6.9 | 1 | 2.4 | 4.6 | 5 |
| Fig7 | 4.5 | 0.4 | 3.9 | 7 | 1.7 |
In upper table, H1, H2, H3, H4, H5 in the first row are style vectors corresponding to five class of subscribers, first
What the Fig1-Fig7 in column was represented is the style vector of the deposit material in material database, and the number in table is deposit material
The distance between style vector and 5 unused class of subscriber style vectors, number are smaller, then it represents that similarity is higher.
The style vector of the deposit material can be obtained by calculation identical with typical material style vector,
This is repeated no more.
If real-time traffic statistics indicate that, the style vector that class of subscriber belonging to active user has be H5, then according to
The size of similarity is ranked up between the style vector and H5 of Fig1-Fig7.For example, according to upper table be ranked up for Fig1,
Fig2、Fig7、Fig3、Fig5、Fig4、Fig6。
Next, can be candidate according to the preset each material number preparation for needing to recommend to active user of system
Material.If necessary to recommend 3 materials to user, then the candidate material recommended can be Fig1, Fig2, Fig7；If necessary
Recommend 5 materials to user, then the candidate material recommended can be Fig1, Fig2, Fig7, Fig3, Fig5；If necessary to
7 materials are recommended at family, then the candidate material recommended can be Fig1, Fig2, Fig7, Fig3, Fig5, Fig4, Fig6.
After determining candidate material, step S15 can be continued to execute, the candidate material is recommended into the current use
Family.
It, can not also be according to the sequence of similarity (i.e. distance from small to large) from high to low in above-mentioned concrete application scene
To determine the candidate material recommended to active user.It in other words, can also be from upper if necessary to recommend 3 materials to user
It states and randomly selects 3 materials in 7 materials as the candidate material recommended, such as: it can be Fig2, Fig5, Fig4, at this point,
The similarity of Fig2, Fig5, Fig4 and class of subscriber style vector H5 are not to be arranged in front 3.
Technical solution of the present invention can be by the deposit material in material database according to its style vector and the active user institute
The similarity of the style vector of the class of subscriber of category is ranked up from high to low, chooses the forward m material that sort as candidate
Material.Accordingly, for the material never launched newly being added in material database, can prejudge may be to its interested use
Family prevents the cold start-up of new material, effectively prevents the waste that material launches cost.
Although above embodiments are illustrated so that advertisement is recommended as an example, but it should be recognized that the technical side of the present embodiment
Case is not limited to advertisement recommendation, can be also used for commending contents or the recommendation of other objects.It, can be with for example, in commending contents
The materials such as news, short-sighted frequency are recommended.When carrying out commending contents, news, the style dimension of short-sighted frequency be can according to need
It is set.In a unrestricted embodiment, for example, the keyword according to included in news will be described new
The material style of news is divided into report style, comment style, analysis style etc..
Further, referring to Fig. 4, Fig. 4 is a kind of structural schematic diagram of material recommendation apparatus of the embodiment of the present invention.The present invention
Embodiment additionally provides a kind of material recommendation apparatus 1, and the material recommendation apparatus 1 may include: user's cluster module 2, is suitable for
According to the historical traffic data of multiple users, at least one class of subscriber is determined；Material style statistical module 3 is suitable for statistics institute
The respective typical material of each class of subscriber is stated, and each use is determined according to the style vector of the typical material of each class of subscriber
The style vector of family classification, show in the typical case's historical traffic data of the material selected from each class of subscriber number be greater than it is default
The style vector of the material of threshold value, typical case's material is determined based on predefined multiple style dimensions；Real-time traffic is true
Cover half block 4 determines class of subscriber belonging to the active user suitable for the real-time traffic data according to active user；Material sequence
Module 5, suitable for the wind by the deposit material in material database according to class of subscriber belonging to its style vector and the active user
The similarity of lattice vector is ranked up, to determine candidate material according to ranking results；Material recommending module 6 is suitable for the time
Material is selected to recommend the active user.
The material recommendation apparatus 1 can recommend its interested ad material to user, be conducive to the point for improving material
Hit rate.For advertisement recommendation, conversion ratio is also advantageously improved, more preferably advertisement element is obtained with less cost to realize
Material launches effect.For content (such as news, short-sighted frequency etc.) recommendation, it is more accurate to recommend, and is conducive to change
Kind user experience.
Further, referring to Fig. 5, Fig. 5 is the structural schematic diagram of user's cluster module in Fig. 4.User's cluster module 2
It may include: that cluster centre chooses submodule 21, suitable for choosing K user's from the historical traffic data of the multiple user
Data centered on historical traffic data, wherein each centre data corresponds to a class of subscriber, and K is positive integer；User draws
Molecular modules 22 will be described suitable for the similarity of a centre data of historical traffic data and K according to the multiple user
Multiple users are divided into K class of subscriber.
User's cluster module in the material recommendation apparatus can classify to multiple users using cluster mode,
It is accurately analyzed convenient for the subsequent style to class of subscriber.
Further, it may include: comparing unit 221 that the user, which divides submodule 22, be adapted to compare each user's
The similarity of historical traffic data and the K centre datas, is recorded and each highest centre data of user's similarity；It draws
Sub-unit 222, suitable for each user to be subdivided into the corresponding class of subscriber of the highest centre data of similarity.
Further, referring to Fig. 6, Fig. 6 is the structural schematic diagram of material style statistical module in Fig. 4.The material style
Statistical module 3 may include: typical story extraction submodule 31, and statistics shows in the historical traffic data of each class of subscriber
Number is greater than the material of preset threshold, and will show material of the number greater than preset threshold and sort from high to low according to clicking rate,
It chooses the n material that clicking rate sorts forward and is used as typical material, wherein n is positive integer；Material style extracting sub-module 32,
For each class of subscriber, suitable for calculating separately the style vector of wherein each typical material, and according to each typical material
Style vector calculates the style vector of the class of subscriber.Due to the high material of clicking rate can be lower than clicking rate material more
The style that associated user likes accurately is reflected, therefore, using the high material of clicking rate as typical material, may improve and calculate
Accuracy, to be conducive to increase precision when subsequent material is recommended.
Further, the material style extracting sub-module 32 may include: summing elements 321, be suitable for calculating the use
In the classification of family the style vector of each typical case's material cumulative and, to obtain the style vector of the class of subscriber.
Further, the material sorting module 5 may include: sorting sub-module (not shown), be suitable for according to similarity
The deposit material is ranked up from high to low, chooses the forward m material that sort as candidate material, wherein m is positive whole
Number.The material recommendation apparatus 1, can be pre- in use, the material never launched for being newly added in material database as a result,
First judgement may prevent the cold start-up of new material to its interested user, effectively prevent the waste that material launches cost.
Working principle, more contents of working method about the material recommendation apparatus 1, are referred to Fig. 1 into Fig. 3
Associated description, which is not described herein again.
The embodiment of the invention also discloses a kind of storage mediums, are stored thereon with computer instruction, the computer instruction
The step of material recommended method shown in Fig. 1, Fig. 2 or Fig. 3 can be executed when operation.The storage medium may include ROM,
RAM, disk or CD etc..The storage medium can also include non-volatility memorizer (non-volatile) or non-transient
(non-transitory) memory etc..
The embodiment of the invention also discloses a kind of terminal, the terminal may include memory and processor, the storage
The computer instruction that can be run on the processor is stored on device.The processor can be with when running the computer instruction
The step of executing material recommended method shown in Fig. 1, Fig. 2 or Fig. 3.The terminal includes but is not limited to mobile phone, computer, puts down
The terminal devices such as plate computer.
Although present disclosure is as above, present invention is not limited to this.Anyone skilled in the art are not departing from this
It in the spirit and scope of invention, can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
Subject to the range of restriction.
Claims (14)
1. a kind of material recommended method characterized by comprising
According to the historical traffic data of multiple users, at least one class of subscriber is determined；
Count the respective typical material of each class of subscriber, and the style vector of the typical material according to each class of subscriber
It determines the style vector of each class of subscriber, shows in the historical traffic data of typical case's material selected from each class of subscriber secondary
Number is greater than the material of preset threshold, and the style vector of typical case's material is determined based on predefined multiple style dimensions；
Class of subscriber belonging to the active user is determined according to the real-time traffic data of active user；
By the deposit material in material database according to the style vector of class of subscriber belonging to its style vector and the active user
Similarity be ranked up, to determine candidate material according to ranking results；
The candidate material is recommended into the active user.
2. material recommended method according to claim 1, which is characterized in that according to the historical traffic data of multiple users,
Determine that at least one class of subscriber includes:
From data centered on the historical traffic data for choosing K user in the historical traffic data of the multiple user, wherein
Each centre data corresponds to a class of subscriber, and K is positive integer；
According to the similarity of the historical traffic data of the multiple user and the K centre datas, the multiple user is drawn
It is divided into K class of subscriber.
3. material recommended method according to claim 2, which is characterized in that according to the historical traffic number of the multiple user
According to the similarity with the K centre datas, the multiple user, which is divided into K class of subscriber, includes:
The similarity of historical traffic data and the K centre data of more each user, is recorded similar to each user
Spend highest centre data；
Each user is subdivided into the corresponding class of subscriber of the highest centre data of similarity.
4. material recommended method according to claim 1, which is characterized in that statistics each respective allusion quotation of class of subscriber
Type material, and determine that the style vector of each class of subscriber includes: according to the style vector of the typical material of each class of subscriber
Statistics shows the material that number is greater than preset threshold in the historical traffic data of each class of subscriber, and will show number
Material greater than preset threshold sorts from high to low according to clicking rate, chooses the n material that clicking rate sorts forward and is used as typical case
Material, wherein n is positive integer；
For each class of subscriber, the style vector of wherein each typical material is calculated separately, and according to each typical material
Style vector calculates the style vector of the class of subscriber.
5. material recommended method according to claim 4, which is characterized in that according to the style of each typical material to meter
The style vector for calculating the class of subscriber includes: the cumulative of the style vector of each typical material in the calculating class of subscriber
With to obtain the style vector of the class of subscriber.
6. material recommended method according to claim 1, which is characterized in that by the deposit material in material database according to its wind
Lattice vector and the similarity of the style vector of class of subscriber belonging to the active user are ranked up, with true according to ranking results
Determining candidate material includes:
The deposit material is ranked up from high to low according to similarity, chooses the forward m material that sort as candidate element
Material, wherein m is positive integer.
7. a kind of material recommendation apparatus characterized by comprising
User's cluster module determines at least one class of subscriber suitable for the historical traffic data according to multiple users；
Material style statistical module is suitable for counting the respective typical material of each class of subscriber, and according to each user class
The style vector of other typical case's material determines that the style vector of each class of subscriber, typical case's material are selected from each class of subscriber
Historical traffic data in show the material that number is greater than preset threshold, the style vector of typical case's material is based on predefined
Multiple style dimensions determine；
Real-time traffic determining module determines user belonging to the active user suitable for the real-time traffic data according to active user
Classification；
Material sorting module, suitable for by the deposit material in material database according to use belonging to its style vector and the active user
The similarity of the style vector of family classification is ranked up, to determine candidate material according to ranking results；
Material recommending module, suitable for the candidate material is recommended the active user.
8. material recommendation apparatus according to claim 7, which is characterized in that user's cluster module includes:
Cluster centre chooses submodule, suitable for choosing the history stream of K user from the historical traffic data of the multiple user
Data centered on amount data, wherein each centre data corresponds to a class of subscriber, and K is positive integer；
User divides submodule, suitable for similar to a centre data of K according to the historical traffic data of the multiple user
Degree, is divided into K class of subscriber for the multiple user.
9. material recommendation apparatus according to claim 8, which is characterized in that the user divides submodule and includes:
Comparing unit is adapted to compare the historical traffic data of each user and the similarity of the K centre data, record with
Each highest centre data of user's similarity；
Division unit, suitable for each user to be subdivided into the corresponding class of subscriber of the highest centre data of similarity.
10. material recommendation apparatus according to claim 7, which is characterized in that the material style statistical module includes:
Typical story extraction submodule, statistics show number greater than preset threshold in the historical traffic data of each class of subscriber
Material, and number will be showed and sorted from high to low greater than the material of preset threshold according to clicking rate, and chosen clicking rate sequence and leans on
N preceding material is as typical material, wherein n is positive integer；
Material style extracting sub-module, for each class of subscriber, suitable for calculate separately the style of wherein each typical material to
It measures, and calculates the style vector of the class of subscriber according to the style vector of each typical material.
11. material recommendation apparatus according to claim 10, which is characterized in that the material style extracting sub-module packet
It includes:
Summing elements, suitable for calculate the cumulative of the style vector of each typical material in the class of subscriber and, it is described to obtain
The style vector of class of subscriber.
12. material recommendation apparatus according to claim 7, which is characterized in that the material sorting module includes:
Sorting sub-module is chosen and sorts forward m suitable for being ranked up from high to low to the deposit material according to similarity
Material is as candidate material, wherein m is positive integer.
13. a kind of storage medium, is stored thereon with computer instruction, which is characterized in that the computer instruction executes when running
The step of material recommended method described in any one of claims 1-6.
14. a kind of terminal, including memory and processor, be stored on the memory to run on the processor
Computer instruction, which is characterized in that perform claim requires described in any one of 1-6 when the processor runs the computer instruction
Material recommended method the step of.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811073786.4A CN109409928A (en) | 2018-09-14 | 2018-09-14 | A kind of material recommended method, device, storage medium, terminal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811073786.4A CN109409928A (en) | 2018-09-14 | 2018-09-14 | A kind of material recommended method, device, storage medium, terminal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109409928A true CN109409928A (en) | 2019-03-01 |
Family
ID=65464919
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811073786.4A Withdrawn CN109409928A (en) | 2018-09-14 | 2018-09-14 | A kind of material recommended method, device, storage medium, terminal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109409928A (en) |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109933700A (en) * | 2019-03-07 | 2019-06-25 | 王芃翰 | Students ' reading based on big data suggests generation method and relevant device |
| CN110851653A (en) * | 2019-11-08 | 2020-02-28 | 上海摩象网络科技有限公司 | Method and device for shooting material mark and electronic equipment |
| CN111324920A (en) * | 2020-03-17 | 2020-06-23 | 广东三维家信息科技有限公司 | Material sorting method and device, electronic equipment and computer readable medium |
| CN111368187A (en) * | 2020-02-26 | 2020-07-03 | 广东三维家信息科技有限公司 | Household material recommendation method and device and server |
| CN112819685A (en) * | 2019-11-15 | 2021-05-18 | 青岛海信移动通信技术股份有限公司 | Image style mode recommendation method and terminal |
| CN112836913A (en) * | 2019-11-25 | 2021-05-25 | 北京国双科技有限公司 | Service configuration method, device and equipment |
| CN114037599A (en) * | 2021-09-16 | 2022-02-11 | 福建大娱号信息科技股份有限公司 | Intelligent image matting method and device based on natural environment scene information and storage medium |
| CN114071382A (en) * | 2021-11-16 | 2022-02-18 | 中国银行股份有限公司 | Message sending method and device |
| CN114463673A (en) * | 2021-12-31 | 2022-05-10 | 深圳市东信时代信息技术有限公司 | Material recommendation method, device, equipment and storage medium |
| CN114491092A (en) * | 2022-01-26 | 2022-05-13 | 深圳市前海手绘科技文化有限公司 | Method and system for recommending materials according to document content and color matching |
| CN114662175A (en) * | 2022-04-01 | 2022-06-24 | 亿玛创新网络(天津)有限公司 | AI screening method, system, equipment and medium for intelligent design generated result |
| CN117648471A (en) * | 2024-01-30 | 2024-03-05 | 深圳鹏锐信息技术股份有限公司 | Three-dimensional industrial design material AI intelligent recommendation management system |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103886090A (en) * | 2014-03-31 | 2014-06-25 | 北京搜狗科技发展有限公司 | Content recommendation method and device based on user favorites |
| CN105224699A (en) * | 2015-11-17 | 2016-01-06 | Tcl集团股份有限公司 | A kind of news recommend method and device |
| CN108491529A (en) * | 2018-03-28 | 2018-09-04 | 百度在线网络技术（北京）有限公司 | Information recommendation method and device |
-
2018
- 2018-09-14 CN CN201811073786.4A patent/CN109409928A/en not_active Withdrawn
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103886090A (en) * | 2014-03-31 | 2014-06-25 | 北京搜狗科技发展有限公司 | Content recommendation method and device based on user favorites |
| CN105224699A (en) * | 2015-11-17 | 2016-01-06 | Tcl集团股份有限公司 | A kind of news recommend method and device |
| CN108491529A (en) * | 2018-03-28 | 2018-09-04 | 百度在线网络技术（北京）有限公司 | Information recommendation method and device |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109933700A (en) * | 2019-03-07 | 2019-06-25 | 王芃翰 | Students ' reading based on big data suggests generation method and relevant device |
| CN110851653A (en) * | 2019-11-08 | 2020-02-28 | 上海摩象网络科技有限公司 | Method and device for shooting material mark and electronic equipment |
| CN112819685B (en) * | 2019-11-15 | 2022-11-04 | 青岛海信移动通信技术股份有限公司 | Image style mode recommendation method and terminal |
| CN112819685A (en) * | 2019-11-15 | 2021-05-18 | 青岛海信移动通信技术股份有限公司 | Image style mode recommendation method and terminal |
| CN112836913A (en) * | 2019-11-25 | 2021-05-25 | 北京国双科技有限公司 | Service configuration method, device and equipment |
| CN111368187B (en) * | 2020-02-26 | 2023-05-26 | 广东三维家信息科技有限公司 | Household material recommendation method, device and server |
| CN111368187A (en) * | 2020-02-26 | 2020-07-03 | 广东三维家信息科技有限公司 | Household material recommendation method and device and server |
| CN111324920A (en) * | 2020-03-17 | 2020-06-23 | 广东三维家信息科技有限公司 | Material sorting method and device, electronic equipment and computer readable medium |
| CN111324920B (en) * | 2020-03-17 | 2023-06-23 | 广东三维家信息科技有限公司 | Material sorting method, apparatus, electronic device and computer readable medium |
| CN114037599A (en) * | 2021-09-16 | 2022-02-11 | 福建大娱号信息科技股份有限公司 | Intelligent image matting method and device based on natural environment scene information and storage medium |
| CN114071382A (en) * | 2021-11-16 | 2022-02-18 | 中国银行股份有限公司 | Message sending method and device |
| CN114463673A (en) * | 2021-12-31 | 2022-05-10 | 深圳市东信时代信息技术有限公司 | Material recommendation method, device, equipment and storage medium |
| CN114491092A (en) * | 2022-01-26 | 2022-05-13 | 深圳市前海手绘科技文化有限公司 | Method and system for recommending materials according to document content and color matching |
| CN114662175A (en) * | 2022-04-01 | 2022-06-24 | 亿玛创新网络(天津)有限公司 | AI screening method, system, equipment and medium for intelligent design generated result |
| CN117648471A (en) * | 2024-01-30 | 2024-03-05 | 深圳鹏锐信息技术股份有限公司 | Three-dimensional industrial design material AI intelligent recommendation management system |
| CN117648471B (en) * | 2024-01-30 | 2024-04-09 | 深圳鹏锐信息技术股份有限公司 | Three-dimensional industrial design material AI intelligent recommendation management system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109409928A (en) | A kind of material recommended method, device, storage medium, terminal | |
| US10095771B1 (en) | Clustering and recommending items based upon keyword analysis | |
| CN105224699B (en) | News recommendation method and device | |
| CN108363821A (en) | A kind of information-pushing method, device, terminal device and storage medium | |
| TWI648642B (en) | Data search processing method and system | |
| US8660901B2 (en) | Matching of advertising sources and keyword sets in online commerce platforms | |
| WO2017121251A1 (en) | Information push method and device | |
| CN106372249B (en) | A kind of clicking rate predictor method, device and electronic equipment | |
| US20140108190A1 (en) | Recommending product information | |
| CN111461841B (en) | Article recommendation method, device, server and storage medium | |
| CN107894998B (en) | Video recommendation method and device | |
| WO2018121700A1 (en) | Method and device for recommending application information based on installed application, terminal device, and storage medium | |
| CN110473040B (en) | Product recommendation method and device and electronic equipment | |
| CN108132963A (en) | Resource recommendation method and device, computing device and storage medium | |
| CN111444395A (en) | Method, system and equipment for obtaining relation expression between entities and advertisement recalling system | |
| KR20150023432A (en) | Method and apparatus for inferring user demographics | |
| CN107943910B (en) | Personalized book recommendation method based on combined algorithm | |
| CN109460519B (en) | Browsing object recommendation method and device, storage medium and server | |
| CN103136683A (en) | Method and device for calculating product reference price and method and system for searching products | |
| Zhou et al. | Relevance feature mapping for content-based multimedia information retrieval | |
| CN111815375B (en) | User portrayal method and device in advertisement putting | |
| CN106354867A (en) | Multimedia resource recommendation method and device | |
| CN110597987A (en) | Search recommendation method and device | |
| CN111626767B (en) | Resource data issuing method, device and equipment | |
| CN111078997B (en) | Information recommendation method and device |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190301 |