CN105869016A - Method for estimating click through rate based on convolution neural network - Google Patents
Method for estimating click through rate based on convolution neural network Download PDFInfo
- Publication number
- CN105869016A CN105869016A CN201610186350.0A CN201610186350A CN105869016A CN 105869016 A CN105869016 A CN 105869016A CN 201610186350 A CN201610186350 A CN 201610186350A CN 105869016 A CN105869016 A CN 105869016A
- Authority
- CN
- China
- Prior art keywords
- click
- pond
- layer
- convolutional neural
- parameter
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; 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
- G06Q30/0242—Determining effectiveness of advertisements
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
- G06F17/153—Multidimensional correlation or convolution
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Business, Economics & Management (AREA)
- Finance (AREA)
- Strategic Management (AREA)
- Computing Systems (AREA)
- Development Economics (AREA)
- Databases & Information Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Algebra (AREA)
- Accounting & Taxation (AREA)
- Marketing (AREA)
- Probability & Statistics with Applications (AREA)
- Economics (AREA)
- Game Theory and Decision Science (AREA)
- Entrepreneurship & Innovation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Operations Research (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- General Business, Economics & Management (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for estimating a click through rate based on a convolution neural network. The method comprises the following steps: establishing Hash tables for all elements of click instances to correspond with potential semantic vectors; for one specific click instance, indexing a corresponding potential semantic vector in the Hash table to obtain a click instance matrix which serves as an input matrix of the convolution neural network; conducting convolution and pooling of the convolution neural network to obtain a multi-layer convolution neural network; finally, multiplying a pooling layer by a full connection matrix, conducting a flexible maximum transfer function calculation to obtain an output layer; optimizing model parameters and inputting the model parameters, adopting a logistic cost function to measure performances of the model, and finally outputting an estimation probability of the click instance being put into each type. According to the invention, the method can mine global semantic interaction information which is important in a single advertisement and regional and global dynamic features in a sequence advertisement, which addresses the problem of current models that only regional or static features can be extracted.
Description
Technical field
The present invention relates to Internet advertising click and commodity interaction technique field, especially a kind of based on
The click-through-rate predictor method of convolutional neural networks.
Background technology
Along with Internet advertising also increases with electricity business blowout formula, advertisement and electricity business's platform are all remembered every day
The user click data of the lower magnanimity of record.For single click behavior, often there is many contexts
Background information, such as user profile, facility information, site information, content information etc., on these
The complex interaction effect of context information often produces tremendous influence to the click behavior of user.
On the other hand, behavior is clicked on for sequence, i.e. according to the time in a certain predetermined time period
The user of sequencing arrangement clicks on the set of behavior, comprises abundant sequence information, such as user
One click behavior before result in current click behavior, the phase clicking on behavior of this sequence
User be clicked on estimating of behavior future and be very helpful by impact and interaction meeting mutually.
Current more popular click-through-rate prediction model is to use this special regression algorithm of logic, or because of
Sub-disassembler algorithm extracts the interaction semantics feature in single click example context background information,
Carry out abstraction sequence with Periodic Neural Networks and click on the sequence signature in example.But matrix decomposition algorithm
Or factorisation machine algorithm all can only extract the interaction semantics feature of two kinds of context information.
On the other hand, in Periodic Neural Networks, periodic signal transfer matrix is always to maintain constant, this cycle
It is special that the hypothesis that feature changes in the way of invariable limits real sequence in reality complex scene
The extraction levied.Accordingly, it would be desirable to excavate the overall situation friendship of context information in single click behavior
Dynamic sequence feature in mutual semantic feature and sequence click behavior, but existing click-through-rate
(Click through rate, CTR) prediction model can not efficiently extract these features.
Summary of the invention
In consideration of it, the present invention proposes a kind of click-through-rate side of estimating based on convolutional neural networks
Method.The method is based on convolutional neural networks can extract in click example multiple context conditions it
Between complicated Semantic interaction information and sequence information in sequential click behavior, it is possible to well should
Push and the scene of all kinds of commending system for Internet advertising.
The present invention is achieved in that a kind of click-through-rate side of estimating based on convolutional neural networks
Method, including step:
Step S1, to clicking on all elements in example, sets up Hash table, makes all elements with potential
Semantic vector one_to_one corresponding；
Step S2, to a certain concrete click example, indexes out the potential applications of correspondence in Hash table
Vector, is formed click example matrix by the potential applications vector that all elements in this click example is corresponding,
Input matrix as convolutional neural networks；
Step S3, is applied to the convolution operation of convolutional neural networks click on the difference in example matrix
On the same dimension of element, obtain first convolutional layer, the i.e. feature of the information of local neighborhood；
Step S4, is applied to previous convolutional layer by the pond operation of convolutional neural networks, extracts institute
Need feature, obtain first pond layer；
Step S5, repeats step S3-S4 convolution operation and pond operation obtains multilamellar convolutional Neural
Network；
Step S6, makes last pond layer of multilamellar convolutional neural networks carry out with full connection matrix
Product calculation, then it is calculated output layer by flexible maximum delivered function, the click i.e. inputted is real
What example was assigned to each class estimates probability；
Step S7, gives initialized input and parameter to model, Optimized model ginseng on data set
Number and output, use this special loss function measurement model performance of logic, finally give the click of input
What example was assigned to each class estimates probability.
The present invention is by order to click in example, each treats mutual element one potential applications of study
Vector is expressed, and can extract list by the convolution operation in convolutional neural networks and pond operation simultaneously
Between the context information in secondary click example, complex interaction and the sequence of the overall situation click on example
In important local or the dynamic sequence feature of the overall situation such that it is able to simultaneously extract single click on reality
Between the context information in example, complex interaction and the sequence of the overall situation are clicked in example important
Local or the dynamic sequence feature of the overall situation, lay a good foundation for accurate output estimation probability, and mould
Type is clicked on the performance estimated and can be measured by this special loss function of logic so that estimate probability the most smart
Really.
Accompanying drawing explanation
The flow chart of Fig. 1 present invention；
Fig. 2 on YOOCHOOSE data set between different models click-through-rate estimate performance
Lateral comparison schematic diagram.
Detailed description of the invention
For making the object, technical solutions and advantages of the present invention clearer, below in conjunction with specifically
Embodiment, and referring to the drawings, the present invention is described in more detail.
Step S1, for clicking on the mutual information element that needed occurred in example, sets up Hash
Table, makes all information elements and potential applications vector one_to_one corresponding；
In the present invention, described click example both can be to single click on time, the website that behavior comprises
The context semantic information such as kind, device category, it is also possible to be that a user is in certain time period
All click behaviors composition click sequence information.
Described treat context information that mutual information element includes single clicing in example with
And the one click example in sequence click example.
Step S2, for a certain concrete click example, indexes out the potential of correspondence in Hash table
Semantic vector, obtains clicking on example matrix, as the input matrix of convolutional neural networks；
It should be noted that in the present invention, the number of the information element in described click example is not
Fixing, input length-flexible is variable, is thus applicable to the click application scenarios of reality, this
Sample is so that the range of application of model is greatly increased.
Step S3, is applied to the convolution operation of convolutional neural networks click on the difference in example matrix
On the same dimension of information element, obtain first convolutional layer, the i.e. feature of the information of local neighborhood；
Wherein, convolution kernel, element expression vector, Quan Lian are comprised with error reverse conduction Algorithm Learning
Connect matrix norm shape parameter；
Step S4, is applied to previous convolutional layer, Chi Huati by the pond operation of convolutional neural networks
Take required feature, obtain first pond layer；
Concrete, it is achieved on, the operation of described pond can be maximum pond or average Chi Hua etc., pond
Changing the function that parameter is the length of previous convolutional layer, described pond parameter uses segmentation power-exponent function；
Have employed the foundation that segmentation power-exponent function selects as pond parameter, enable model to process the longest
The input of degree, increases the range of application of model.
Step S5, repeats above convolution operation and pond operation and obtains multilamellar convolutional neural networks；
In the present invention, the number of plies of the multilamellar convolutional neural networks that step S5 obtains, each layer of feature
The number of figure, convolution kernel size are all hyper parameter；
The scale of hyper parameter decision model and the quantity of parameter, need according to inputting the concrete big of data
Little and kind is adjusted, can be by repeatedly trying in a parameter interval set in advance in experiment
Test contrast to obtain.
Step S6, carries out product calculation by last pond layer and full connection matrix, then by soft
Property maximum delivered function be calculated output layer, i.e. input click on example be assigned to the pre-of each class
Estimate Probability pm；
Wherein, the pond parameter of last pond layer is fixing, keeps with the classification number of output
Unanimously.
Step S7, data set divides training set, checking collection, test set；In training set anti-by error
The model parameters such as vectorial, full connection matrix are expressed to conduction algorithm regulation model convolution kernel, element；
Checking collection upper by test of many times contrast in parameter interval regulate the convolutional network number of plies, convolution kernel size,
The hyper parameter such as characteristic pattern number；On test set, this special loss of the logic of computation model measures click in advance
Estimate the performance of model, use this special loss function tolerance of logic to click on the performance of prediction model；
This special loss function of logic is expressed as:
Wherein, logloss represents this special loss of logic, ntestFor the number of samples all on test set,
tmRepresenting the desired value clicking on example on test set, whether the click behavior the most actually estimated is sent out
Raw, value is 0 or 1, and 0 represents that not click, 1 expression are clicked on, pmRepresent estimated value, i.e.
The probability that model calculated click behavior can occur, value is real-valued on closed interval [0,1].
For click model, this special loss function of logic more closes than general accuracy rate isometry function
It is suitable, even if ratio also is able to measure well under the serious unbalanced situation of the positive negative sample of test set
The performance estimated clicked on by model.
Embodiment
In order to verify the implementation result of the present invention, say as a example by YOOCHOOSE data set below
Bright.This YOOCHOOSE data set contains click record and the purchase of user on electric business's platform
Record, these records were separated by the time period set in advance.Meanwhile, former data set is carried out
Suitable pretreatment operation, all possible click sequence of certain time period, if last point
Hit example correspondence purchasing behavior, then its label is 1, and otherwise its label is 0.On this data set
The task that click-through-rate is estimated can be carried out.Specifically comprise the following steps that
Clicked commodity all in YOOCHOOSE data set are extracted by step S1, set up
One Hash table, makes each commodity and a potential applications vector one_to_one corresponding.
Step S2, by pretreatment operation, is converted into all possible by YOOCHOOSE data set
Click on sequence, and click on one corresponding label of sequence according to purchaser record to each, for certain
One concrete click example, indexes out the potential applications vector of correspondence in Hash table, obtains a little
Hit example matrix, as the input matrix of convolutional neural networks.
Step S3, is applied to the convolution operation in convolutional neural networks click in example matrix not
With on the same dimension of element, obtain first convolutional layer, the i.e. feature of the information of local neighborhood,
By the parameter of error reverse conduction Algorithm Learning model, regulation includes convolution kernel, and element expresses vector,
The model parameter of full connection matrix.
Step S4, is applied to previous convolutional layer, Chi Hua by the pond operation in convolutional neural networks
Operation is maximum pond, and pond parameter is the function of the length of previous convolutional layer, is chosen as segmentation power
Refer to that function, pondization operation can efficiently extract required feature, obtain first pond layer.
Step S5, repeats convolution operation above and pond operation obtains multilamellar convolutional neural networks,
The number of plies of network, the number of each layer of characteristic pattern, convolution kernel size is all hyper parameter, according to tool
Size and the kind of the input data of body are adjusted, can be by a parameter set in advance
Interval test of many times contrast obtains.
Step S6, last pond layer and full connection matrix carry out product calculation, then pass through flexibility
Maximum delivered function is calculated output layer, and the example of clicking on i.e. inputted is assigned to estimating of each class
Probability, the pond parameter of last pond layer is fixed value 2, keeps one with the classification number of output
Cause.
Step S7, data set divides training set, checking collection, test set；In training set, use error
Reverse conduction algorithm regulation model convolution kernel, element express the model parameters such as vectorial, full connection matrix；
On checking collection, by test of many times contrast regulation the convolutional network number of plies, convolution kernel in parameter interval
The hyper parameter such as size, characteristic pattern number；On test set, this special loss of the logic of computation model is spent
Amount clicks on the performance of prediction model.
Fig. 1 is on YOOCHOOSE data set, and between different models, click-through-rate estimates performance
Lateral comparison schematic diagram, CCPM is that present invention click-through-rate based on convolutional neural networks is estimated
Model, RNN is Periodic Neural Networks model, and FM is disassembler model, and LR is this special recurrence of logic
Model.
Table 1 is the characteristic pattern number (f1, f2, f3) in the present invention each layer convolutional neural networks and volume
The model hyper parameter such as long-pending core size (k1, k2, k3) when the parameter set interval [2,9] changes,
Model click-through-rate estimates the longitudinal comparison of performance.
Table 1
Particular embodiments described above, is carried out the purpose of the present invention, technical scheme and effect
Further describe, be it should be understood that the foregoing is only the present invention specific embodiment and
, be not limited to the present invention, all within the spirit and principles in the present invention, that is done is any
Amendment, equivalent, improvement etc., should be included within the scope of the present invention.
Claims (8)
1. a click-through-rate predictor method based on convolutional neural networks, it is characterised in that
Including step:
Step S1, to clicking on all elements in example, sets up Hash table, makes all elements with potential
Semantic vector one_to_one corresponding；
Step S2, to a certain concrete click example, indexes out the potential applications of correspondence in Hash table
Vector, is formed click example matrix by the potential applications vector that all elements in this click example is corresponding,
Input matrix as convolutional neural networks；
Step S3, is applied to the convolution operation of convolutional neural networks click on the difference in example matrix
On the same dimension of element, obtain first convolutional layer, the i.e. feature of the information of local neighborhood；
Step S4, is applied to previous convolutional layer by the pond operation of convolutional neural networks, extracts institute
Need feature, obtain first pond layer；
Step S5, repeats step S3-S4 convolution operation and pond operation obtains multilamellar convolutional Neural
Network；
Step S6, makes last pond layer of multilamellar convolutional neural networks carry out with full connection matrix
Product calculation, then it is calculated output layer by flexible maximum delivered function, the click i.e. inputted is real
What example was assigned to each class estimates probability；
Step S7, gives initialized input and parameter to model, Optimized model ginseng on data set
Number and output, use this special loss function measurement model performance of logic, finally give the click of input
What example was assigned to each class estimates probability.
Method the most according to claim 1, it is characterised in that described click example both may be used
To be the context semantic information in single clicing on, it is also possible to be that a user is in certain time period
All click behaviors composition click sequence information.
Method the most according to claim 1, it is characterised in that in described click example
The number of element is change.
Method the most according to claim 1, it is characterised in that in described click example
It is reverse by error that element expresses vector, convolution kernel in convolution operation, full connection matrix model parameter
Conduction algorithm learns.
Method the most according to claim 1, it is characterised in that described convolutional neural networks
In pondization operation be maximum pond or average pond, pond parameter is the length of previous convolutional layer
Function；The function of pond parameter uses segmentation power-exponent function.
Method the most according to claim 1, it is characterised in that described multilamellar convolutional Neural
The network of network number of plies, the number of each layer of characteristic pattern, convolution kernel size are hyper parameter, according to defeated
Enter size and the kind regulation of data, can be by carrying out repeatedly in a parameter interval set in advance
Experimental Comparison model performance obtains.
Method the most according to claim 1, it is characterised in that last pond layer
Pond parameter is fixing, and keeps consistent with the classification number of output.
Method the most according to claim 1, it is characterised in that this special loss of described logic
Function representation is:
Wherein, log loss represents this special loss of logic, ntestFor numbers of samples all on test set, tm
Representing and click on example goal value on test set, value is 0 or 1, and 0 represents not click, and 1 represents
Click on；pmRepresenting the probability that calculated click behavior occurs, value is on closed interval [0,1]
Real-valued.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610186350.0A CN105869016A (en) | 2016-03-28 | 2016-03-28 | Method for estimating click through rate based on convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610186350.0A CN105869016A (en) | 2016-03-28 | 2016-03-28 | Method for estimating click through rate based on convolution neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN105869016A true CN105869016A (en) | 2016-08-17 |
Family
ID=56626121
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610186350.0A Pending CN105869016A (en) | 2016-03-28 | 2016-03-28 | Method for estimating click through rate based on convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105869016A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106844294A (en) * | 2016-12-29 | 2017-06-13 | 华为机器有限公司 | Convolution operation chip and communication equipment |
| CN106951783A (en) * | 2017-03-31 | 2017-07-14 | 国家电网公司 | A kind of Method for Masquerade Intrusion Detection and device based on deep neural network |
| CN108280511A (en) * | 2018-01-10 | 2018-07-13 | 北京掌阔移动传媒科技有限公司 | A method of network access data is carried out based on convolutional network and is handled |
| CN108537624A (en) * | 2018-03-09 | 2018-09-14 | 西北大学 | A kind of tourist service recommendation method based on deep learning |
| CN109615060A (en) * | 2018-11-27 | 2019-04-12 | 深圳前海微众银行股份有限公司 | CTR estimation method, apparatus and computer readable storage medium |
| CN109815916A (en) * | 2019-01-28 | 2019-05-28 | 成都蝉远科技有限公司 | A kind of recognition methods of vegetation planting area and system based on convolutional neural networks algorithm |
| CN109815914A (en) * | 2019-01-28 | 2019-05-28 | 成都蝉远科技有限公司 | A kind of convolutional neural networks model training method and system based on vegetation area identification |
| CN109816542A (en) * | 2019-01-28 | 2019-05-28 | 中国平安财产保险股份有限公司四川分公司 | A kind of crop production reduction Claims Resolution method and system |
| CN110263982A (en) * | 2019-05-30 | 2019-09-20 | 百度在线网络技术（北京）有限公司 | The optimization method and device of ad click rate prediction model |
| CN111325579A (en) * | 2020-02-25 | 2020-06-23 | 华南师范大学 | Advertisement click rate prediction method |
| CN112435035A (en) * | 2019-08-09 | 2021-03-02 | 阿里巴巴集团控股有限公司 | Data auditing method, device and equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104199822A (en) * | 2014-07-11 | 2014-12-10 | 五八同城信息技术有限公司 | Method and system for identifying demand classification corresponding to searching |
| CN104217214A (en) * | 2014-08-21 | 2014-12-17 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Configurable convolutional neural network based red green blue-distance (RGB-D) figure behavior identification method |
| CN104700099A (en) * | 2015-03-31 | 2015-06-10 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing traffic signs |
| CN104915399A (en) * | 2015-05-29 | 2015-09-16 | 百度在线网络技术（北京）有限公司 | Recommended data processing method based on news headline and recommended data processing method system based on news headline |
| CN109165350A (en) * | 2018-08-23 | 2019-01-08 | 成都品果科技有限公司 | A kind of information recommendation method and system based on deep knowledge perception |
-
2016
- 2016-03-28 CN CN201610186350.0A patent/CN105869016A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104199822A (en) * | 2014-07-11 | 2014-12-10 | 五八同城信息技术有限公司 | Method and system for identifying demand classification corresponding to searching |
| CN104217214A (en) * | 2014-08-21 | 2014-12-17 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | Configurable convolutional neural network based red green blue-distance (RGB-D) figure behavior identification method |
| CN104700099A (en) * | 2015-03-31 | 2015-06-10 | 百度在线网络技术（北京）有限公司 | Method and device for recognizing traffic signs |
| CN104915399A (en) * | 2015-05-29 | 2015-09-16 | 百度在线网络技术（北京）有限公司 | Recommended data processing method based on news headline and recommended data processing method system based on news headline |
| CN109165350A (en) * | 2018-08-23 | 2019-01-08 | 成都品果科技有限公司 | A kind of information recommendation method and system based on deep knowledge perception |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106844294A (en) * | 2016-12-29 | 2017-06-13 | 华为机器有限公司 | Convolution operation chip and communication equipment |
| CN106844294B (en) * | 2016-12-29 | 2019-05-03 | 华为机器有限公司 | Convolution arithmetic chips and communication equipment |
| CN106951783B (en) * | 2017-03-31 | 2021-06-01 | 国家电网公司 | A method and device for camouflage intrusion detection based on deep neural network |
| CN106951783A (en) * | 2017-03-31 | 2017-07-14 | 国家电网公司 | A kind of Method for Masquerade Intrusion Detection and device based on deep neural network |
| CN108280511A (en) * | 2018-01-10 | 2018-07-13 | 北京掌阔移动传媒科技有限公司 | A method of network access data is carried out based on convolutional network and is handled |
| CN108537624A (en) * | 2018-03-09 | 2018-09-14 | 西北大学 | A kind of tourist service recommendation method based on deep learning |
| CN108537624B (en) * | 2018-03-09 | 2021-08-03 | 西北大学 | A travel service recommendation method based on deep learning |
| CN109615060A (en) * | 2018-11-27 | 2019-04-12 | 深圳前海微众银行股份有限公司 | CTR estimation method, apparatus and computer readable storage medium |
| CN109615060B (en) * | 2018-11-27 | 2023-06-30 | 深圳前海微众银行股份有限公司 | CTR estimation method, device and computer-readable storage medium |
| CN109816542A (en) * | 2019-01-28 | 2019-05-28 | 中国平安财产保险股份有限公司四川分公司 | A kind of crop production reduction Claims Resolution method and system |
| CN109815914A (en) * | 2019-01-28 | 2019-05-28 | 成都蝉远科技有限公司 | A kind of convolutional neural networks model training method and system based on vegetation area identification |
| CN109815916A (en) * | 2019-01-28 | 2019-05-28 | 成都蝉远科技有限公司 | A kind of recognition methods of vegetation planting area and system based on convolutional neural networks algorithm |
| CN109816542B (en) * | 2019-01-28 | 2023-04-07 | 中国平安财产保险股份有限公司四川分公司 | Method and system for reducing yield and settling claim of crops |
| CN110263982A (en) * | 2019-05-30 | 2019-09-20 | 百度在线网络技术（北京）有限公司 | The optimization method and device of ad click rate prediction model |
| CN112435035A (en) * | 2019-08-09 | 2021-03-02 | 阿里巴巴集团控股有限公司 | Data auditing method, device and equipment |
| CN111325579A (en) * | 2020-02-25 | 2020-06-23 | 华南师范大学 | Advertisement click rate prediction method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105869016A (en) | Method for estimating click through rate based on convolution neural network | |
| CN111339415B (en) | Click rate prediction method and device based on multi-interactive attention network | |
| CN105677769B (en) | One kind is based on latent Dirichletal location（LDA）The keyword recommendation method and system of model | |
| CN105677780A (en) | Scalable user intent mining method and system thereof | |
| CN111460290A (en) | Information recommendation method, device, equipment and storage medium | |
| CN105868847A (en) | Shopping behavior prediction method and device | |
| CN106296242A (en) | A kind of generation method of commercial product recommending list in ecommerce and the system of generation | |
| Abdullahi et al. | Time-series large language models: A systematic review of state-of-the-art | |
| WO2021159787A1 (en) | Content processing method and apparatus, computer-readable storage medium and computer device | |
| CN114329193B (en) | Click-through rate prediction method based on time-aware interest evolution | |
| CN104462327A (en) | Computing method, search processing method, computing device and search processing device for sentence similarity | |
| CN110245285A (en) | A Personalized Recommendation Method Based on Heterogeneous Information Network | |
| CN108776864A (en) | The appraisal procedure and device of user comment value | |
| CN111414538A (en) | Artificial intelligence-based text recommendation method, device and electronic device | |
| CN116257798A (en) | Training of click-through rate prediction model, click-through rate prediction method, system and device | |
| CN106126567A (en) | Method based on trust data recommendation service | |
| Mukhtar et al. | Deep Learning With Processing Algorithms for Forecasting Tourist Arrivals | |
| CN108829908A (en) | A kind of circuit structure Reliability Prediction Method based on depth self-encoding encoder network | |
| CN117151247B (en) | Method, apparatus, computer device and storage medium for modeling machine learning task | |
| CN115858933B (en) | Click-rate prediction methods, devices, electronic equipment, and media based on click models | |
| Riandari et al. | Forecasting the Number of Students in Multiple Linear Regressions | |
| Lu et al. | Product recommendation system based on deep learning | |
| CN118297479A (en) | Evaluation index determination method and device for target application | |
| CN116992373B (en) | Data processing method, apparatus, device, storage medium and computer program product | |
| CN113742532B (en) | User portrayal method, apparatus and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20160817 |