CN110619540A - Click stream estimation method of neural network - Google Patents

Click stream estimation method of neural network Download PDF

Info

Publication number
CN110619540A
CN110619540A CN201910746636.3A CN201910746636A CN110619540A CN 110619540 A CN110619540 A CN 110619540A CN 201910746636 A CN201910746636 A CN 201910746636A CN 110619540 A CN110619540 A CN 110619540A
Authority
CN
China
Prior art keywords
neural network
weight
data
vector
word embedding
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
Application number
CN201910746636.3A
Other languages
Chinese (zh)
Inventor
邹大方
俞辉
毛家发
盛伟国
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN201910746636.3A priority Critical patent/CN110619540A/en
Publication of CN110619540A publication Critical patent/CN110619540A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • 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
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G06Q30/0242—Determining effectiveness of advertisements
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Strategic Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Game Theory and Decision Science (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Biomedical Technology (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Tourism & Hospitality (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A click stream estimation method of a neural network comprises the following steps: 1) a large amount of user historical behavior data is collected, and the data comprises characteristics possibly helpful for click stream estimation, such as 4 types of data including advertisement commodity information, user information, context information, shop information and the like, so that a data set is constructed on the basis of the characteristics. 2) And constructing a weight matrix decomposition cross neural network model of the associated features, wherein the model comprises a logistic block, a word embedding vector weight intersection block and a hidden layer block. 3) And cutting the data set into small blocks, sequentially inputting the small blocks into a weight matrix decomposition cross neural network of the associated features, and updating parameters by using a back propagation Adam algorithm until the parameters meet the early-stop condition and are converged. 4) And the trained model is used for completing the click prediction of the user on the advertisement commodity in an actual system. According to the invention, the relatively high-dimensional data is mapped into the low-dimensional word vector through the word embedding technology, so that the calculation amount is reduced, and the learning of a neural network is facilitated.

Description

Click stream estimation method of neural network
Technical Field
The invention relates to user behavior prediction and analysis in the field of computational advertisement, in particular to a large-scale data prediction classification method for a weight matrix decomposition cross neural network based on correlation characteristics, and belongs to the field of click stream prediction.
Background
Click-through rate (CTR) predicts one of the most challenging and valuable techniques in the field of computing advertisements. The purpose of which is to predict the likelihood of a certain advertisement being clicked on based on historical data.
Online advertising is a major source of revenue for most internet companies. In an advertising system, the ranking of advertisements is determined by the bid and click-through rate of the advertisements. It is important to estimate the flow correctly, which determines the revenue for most internet companies. We need to predict in a given scenario (such as an e-commerce platform) the likelihood of a user clicking on an item of an advertised good (such as an advertised good and a store). Unlike continuous data such as video and audio, it is usually discrete to obtain user historical behavior data. While most of these data (such as the user's age, gender, city of residence, etc.) are multi-category and non-continuous. When we want to construct a machine learning system to process these data, it is common practice to convert these discrete data into a multidimensional sparse representation using one-hot encoding. In practical application, the association of learning features can effectively improve the performance of a machine learning model, but professional domain knowledge and a large amount of manpower are required to construct artificial features.
The traditional click stream estimation method is mainly based on a machine learning method: logistic Regression (LR), naive bayes (naive bayes), and Gradient Boosting Decision Tree (GBDT). However, the conventional shallow machine learning model requires a large amount of artificial feature engineering to achieve a relatively ideal accuracy. In recent years, with the development of deep learning technology, the deep learning technology based on the multi-layer neural network is also applied to the field of click stream estimation. The deep learning model can automatically fit data high-dimensional information, so that the artificial characteristic engineering quantity is reduced, and meanwhile, the prediction accuracy can be improved. However, after one-hot coding, the multidimensional sparse data is not suitable for being directly used as the input of the multilayer neural network.
In order to construct an effective machine learning model for predicting sparse features, the machine learning model needs to learn the correlation information between features, which is very important. In practice, feature crossing has proven to be an effective way to improve the accuracy of the prediction. In data mining competitions (Kaggle), many successful approaches rely heavily on artificial feature cross-engineering. However, manual feature engineering is time-consuming and labor-intensive and requires professional domain knowledge, and successful deployment of a model in practical applications requires a significant cost. Therefore, the machine learning model automatically learns the potential association information between the features, which has practical value.
Disclosure of Invention
The invention provides a click stream estimation method of a neural network, aiming at solving the problems that the existing click stream estimation technology is low in precision, needs a large amount of artificial characteristic engineering and the like.
The invention designs a multilayer fully-connected neural network model based on a word embedding (word embedding) method. A model width component (Cheng HT, Koc L, Harmsen J, Shaked T, Chandra T, Aradhye H, Anderson G, Corrado G, Chai W, Ispir M, et al. in Proceedings of the 1st Workshop on Deep Learning Systems for Recommender Systems,2016) was introduced on the basis of the depth model for better Learning the low-order information of the data. By carrying out weight intersection processing on the word embedding vectors based on the inner product of the word embedding vectors to improve the multilayer fully-connected neural network of the depth part of the model, the model can better capture potential correlation information between features to improve the prediction accuracy of the model.
The technical scheme adopted by the invention for solving the technical problems is as follows:
a click stream estimation method of a neural network utilizes word embedding vectors to represent each feature, then uses a weight cross layer to obtain associated information among the features, and finally uses the result of the weight cross layer as the input of the multilayer neural network to learn data high-dimensional information, and the method comprises the following steps:
step 1, collecting a large amount of user historical behavior data, wherein the data comprises characteristics possibly helpful for click stream estimation, such as 4 types of data including advertisement commodity information, user information, context information, shop information and the like, and constructing a data set on the basis of the data.
And 2, constructing a weight matrix decomposition cross neural network model of the associated features, wherein the model comprises a logistic block, a word embedding vector weight cross block and a hidden layer block.
The weight matrix decomposition cross neural network of the correlation characteristics comprises a width module and a depth module. The final output is the output of the width module plus the output of the depth module. The width module is a logistic regression method and mainly aims to make up for the deficiency of the depth part. The depth part is composed of a word embedding layer, a weight crossing layer and a multilayer neural network.
The word embedding layer comprises a series of low-dimensional continuous word embedding vectors corresponding to each feature. Typically, we define a series of corresponding index sets χ from the word vector z, the result of the computation of the word embedding layer being denoted ε ═ z { (z) }ixi}i∈χ。
The weight cross layer comprises a word embedding vector dot product module and a weight vector of a click vector, and the inner product of the weight vector is used as the weight of the word embedding vector dot product. By inner product we get some column vectors, which can be expressed as:
inner product operation that indicates two vectors, (v)i⊙vj)k＝vikvjk. The weight intersection can be expressed as:
in the formulaIs a vector of dimensions k to k, and,is a u-dimensional vector and is a vector,representing the weight of the feature intersection. The output result of the weight cross layer is expressed as:
h0＝[c1,c2,...,cm] (3)
the multi-layer neural network consists of a series of fully connected neural network layers, each block using the ReLu activation function. And taking the result obtained by the weight value cross layer as the input of the multilayer neural network. The neural network of each layer can be represented as:
hl+1＝σ(wlhl+bl) (4)
l denotes the number of layers of the neural network and σ denotes the activation function.Andrespectively representing the output result, the weight and the deviation of the l-th layer.
And adding the calculation result of the width part to the result of the multilayer neural network, and obtaining a final prediction result through a Sigmod activation function. The weight matrix factorization cross neural network of the correlation features can be expressed as:
step 3, network training, comprising the following steps:
the training data is divided into a training set T, a validation set V and a test set T.
And embedding the random initialization words into the characteristic vectors, the weight vectors and the weights and the deviations of the multilayer neural network.
And processing the data of the training set T, the verification set V and the test set T by one-hot and inputting the processed data into the network in batches.
The click stream prediction task generally adopts a log loss function as an objective function. The objective function is defined as follows:
χ represents the set of all training data x, y (x) represents the true value of each data prediction, and σ (y (x)) represents the predicted value of the weight matrix decomposition cross neural network model of the associated features. The parameters of the cross neural network decomposed by the weight matrix of the associated features comprise word embedding vectors, weight vectors and hyper-parameters in the multilayer fully-connected neural network.
And calculating a log loss function value of the training set T, and updating the word embedding characteristic vector, the weight vector and the weight and the deviation of the multilayer neural network according to a back propagation algorithm.
And calculating the log loss function value of the verification set V, judging whether the model is converged, and if the convergence model training is finished, entering data on the next training set T for training until the log loss function value on the verification set V tends to be converged. And finally, verifying the accuracy of the model by using the test set T.
Step 4, the user click stream estimation test comprises the following processes:
and inputting the data after one-hot coding into a weight matrix decomposition cross neural network of the associated features to obtain an output result. And setting a threshold T, decomposing the prediction result in the cross neural network by using the weight matrix of the associated features, and obtaining the predicted classification result according to the threshold T. And comparing the actual clicking behavior of the user with the prediction result, and calculating the average accuracy of the predicted clicking behavior of the user according to the evaluation criterion of the click stream prediction.
Through the operation of the steps, the prediction of the user click stream can be realized.
The invention has the advantages that: the invention provides a click stream estimation method for decomposing a cross neural network by a weight matrix based on associated features, which is characterized in that after data is subjected to one-hot coding, relatively high-dimensional data is mapped into low-dimensional word vectors by a word embedding technology, so that the calculated amount is reduced, and the learning of the neural network is facilitated; the weight cross layer can effectively learn the associated information between the features, and finally, the multi-layer neural network is utilized to learn the high-level information between the features
Drawings
FIG. 1 is a diagram of a cross neural network model structure based on correlation feature weight matrix decomposition according to the present invention.
FIG. 2 is a step of the method for estimating click stream based on neural network method of the present invention.
FIG. 3 is a flow chart of the present invention of weight matrix factorization of correlation features across neural network deployments.
Detailed Description
In order to better explain the technical scheme of the invention, the invention is further explained by an embodiment with the accompanying drawings.
A weight matrix decomposition cross neural network method based on correlation characteristics comprises the following steps:
step 1, taking data in a MovieLens data set (Harper FM, Konstan JA. the movieels data sets: History and context. in actions on interactive systems) to respectively construct a training set and a testing set, wherein the data set comprises score data of a (internet movie database, IMDB) user on a movie and is widely applied as recommended system test data.
And 2, establishing a weight matrix decomposition cross neural network model of the correlation characteristics according to the graph shown in the figure 1. As shown in fig. 1, the mold includes a width portion and a depth portion. The width part of the model is the logistic regression model. The design concept of the depth part of the model is shown in fig. 2. And taking the sparse data subjected to one-hot coding as the input of the model, converting the high-dimensional sparse data into relatively low-dimensional word embedding vectors for representation, then fusing the word embedding vectors by using weight intersection, and finally inputting the fused result into a multilayer fully-connected neural network to obtain the final structure. The specific parameters are as follows: the dimension k of the word embedding vector is 128, the dimension u of the weight vector is 8, p is 1, the layer number l of the hidden layer is 3, and each layer sequentially comprises 128 units, 64 units and 64 units.
And 3, cutting the data set into small blocks, sequentially inputting the small blocks into a weight matrix decomposition cross neural network of the associated features, and updating parameters by using a back propagation algorithm, wherein the size (batch size) of each block comprises 4096 pieces of data. As shown in fig. 3, first, a word embedding vector, a weight vector, and a fully connected neural network are initialized to a positive distribution with an expected value μ of 0 and a variance σ of 1. And then, sequentially inputting the training data of each block into the model, and updating the word embedding vector, the weight vector and the parameters in the fully-connected neural network by using a back propagation algorithm. At the end of each training round, the validation errors on the validation set are calculated, and the training is stopped when the early stopping condition (early stopping) is met, i.e. the validation set errors rise for 5 consecutive rounds. In order to prevent overfitting, in the training process, in order to prevent overfitting, a dropout technology is adopted to process the word embedding vector and the multilayer fully-connected neural network respectively, and 70% of units are reserved when the dropout rate is 0.3. In the training process, the model is trained by using an Adam algorithm, and the learning rate is 0.001.
And 4, completing the preference prediction of the user on the movie by the trained model on the test set.
The embodiments described in this specification are merely illustrative of implementations of the inventive concept and the scope of the present invention should not be considered limited to the specific forms set forth in the embodiments but rather by the equivalents thereof as may occur to those skilled in the art upon consideration of the present inventive concept.

Claims (3)

1. A click stream estimation method of a neural network comprises the following steps:
step 1, collecting a large amount of user historical behavior data, wherein the data comprises characteristics possibly helpful for click stream estimation, namely 4 types of data of advertisement commodity information, user information, context information and shop information, and constructing a data set on the basis of the data;
step 2, constructing a weight matrix decomposition cross neural network model of the associated features, wherein the model comprises a logistic block, a word embedding vector weight value cross block and a hidden layer block;
the weight matrix decomposition cross neural network of the correlation characteristics comprises a width module and a depth module; the final output result is the output of the width module plus the output of the depth module; the width module is a logistic regression method and mainly aims to make up for the deficiency of the depth part; the depth part consists of a word embedding layer, a weight crossing layer and a multilayer neural network;
the word embedding layer comprises a series of low-dimensional continuous word embedding vectors corresponding to each feature; defining a series of corresponding sets of indices of the word vector zThe calculation result of the word embedding layer is expressed as
The weight cross layer comprises a word embedding vector dot product module and a weight vector of a click vector, and the inner product of the weight vector is used as the weight of the word embedding vector dot product; the inner product operation yields a number of column vectors, which can be expressed as:
inner product operation that indicates two vectors, (v)i⊙vj)k＝vikvjk(ii) a The weight intersection can be expressed as:
in the formulaIs a vector of dimensions k to k, and,is a u-dimensional vector and is a vector,representing the weight of the feature intersection; the output result of the weight cross layer is expressed as:
h0＝[c1,c2,...,cm] (3)
the multilayer neural network consists of a series of fully connected neural network layers, and each block uses a ReLu activation function; taking the result obtained by the weight value cross layer as the input of the multilayer neural network; the neural network of each layer can be represented as:
hl+1＝σ(wlhl+bl) (4)
l represents the number of layers of the neural network, and σ represents the activation function;andrespectively representing the output result, the weight and the deviation of the ith layer;
adding the calculation result of the width part to the result of the multilayer neural network, and obtaining a final prediction result through a Sigmod activation function; the weight matrix factorization cross neural network of the correlation features can be expressed as:
step 3, network training, comprising the following steps:
dividing training data into a training set T, a verification set V and a test set T;
embedding a random initialization word into the characteristic vector, the weight vector and the weight and the deviation of the multilayer neural network;
the data of the training set T, the verification set V and the test set T are processed by one-hot and then input to the network in batches;
a click stream estimation task generally adopts a logloss function as a target function; the objective function is defined as follows:
χ represents the set of all training data x, y (x) represents the true value of each data prediction, and σ (y (x)) represents the predicted value of the weight matrix decomposition cross neural network model of the associated features; parameters of the cross neural network decomposed by the weight matrix of the associated features comprise word embedding vectors, weight vectors and hyper-parameters in the multilayer fully-connected neural network;
calculating a log loss function value of the training set T, and updating word embedding characteristic vectors, weight vectors and weights and deviations of the multilayer neural network according to a back propagation algorithm;
calculating a logloss function value of the verification set V, judging whether the model is converged, and if the convergence model training is finished, entering data on the next training set T for training until the log loss function value on the verification set V tends to be converged; finally, verifying the accuracy of the model by using a test set T;
step 4, the user click stream estimation test comprises the following processes:
the data is input into a weight matrix decomposition cross neural network of the associated characteristics after one-hot coding to obtain an output result; setting a threshold T, decomposing the prediction result in the cross neural network by using the weight matrix of the associated features, and obtaining the predicted classification result according to the threshold T; and comparing the actual clicking behavior of the user with the prediction result, and calculating the average accuracy of the predicted clicking behavior of the user according to the evaluation criterion of the click stream prediction.
2. The click stream prediction method of a neural network according to claim 1, characterized in that: the dimension of the word embedding vector in the step 1 is 128, and the dimension of the weight vector is 8.
3. The method for predicting the click stream of a neural network according to claim 1, wherein: in the training network described in step 3, the depth of the hidden layer is 3, and each layer contains 128, 64, and 64 units.
CN201910746636.3A 2019-08-13 2019-08-13 Click stream estimation method of neural network Pending CN110619540A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910746636.3A CN110619540A (en) 2019-08-13 2019-08-13 Click stream estimation method of neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910746636.3A CN110619540A (en) 2019-08-13 2019-08-13 Click stream estimation method of neural network

Publications (1)

Publication Number Publication Date
CN110619540A true CN110619540A (en) 2019-12-27

Family

ID=68921916

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910746636.3A Pending CN110619540A (en) 2019-08-13 2019-08-13 Click stream estimation method of neural network

Country Status (1)

Country Link
CN (1) CN110619540A (en)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111275479A (en) * 2020-01-07 2020-06-12 北京爱笔科技有限公司 People flow prediction method, device and system
CN111898756A (en) * 2020-08-11 2020-11-06 中国人民解放军海军航空大学 Multi-target information associated neural network loss function calculation method and device
CN112270568A (en) * 2020-11-02 2021-01-26 重庆邮电大学 Social e-commerce platform marketing activity order rate prediction method facing hidden information
CN112819523A (en) * 2021-01-29 2021-05-18 上海数鸣人工智能科技有限公司 Marketing prediction method combining inner/outer product feature interaction and Bayesian neural network
CN113129028A (en) * 2020-01-10 2021-07-16 联洋国融(北京)科技有限公司 Rogue user detection system based on time sequence neural network model
CN113222647A (en) * 2021-04-26 2021-08-06 西安点告网络科技有限公司 Advertisement recommendation method, system and storage medium based on click rate estimation model
CN113706211A (en) * 2021-08-31 2021-11-26 平安科技（深圳）有限公司 Advertisement click rate prediction method and system based on neural network

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109960759A (en) * 2019-03-22 2019-07-02 中山大学 Recommender system clicking rate prediction technique based on deep neural network

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109960759A (en) * 2019-03-22 2019-07-02 中山大学 Recommender system clicking rate prediction technique based on deep neural network

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
陈巧红等: "基于递归神经网络的广告点击率预估研究", 《浙江理工大学学报(自然科学版)》 *

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111275479A (en) * 2020-01-07 2020-06-12 北京爱笔科技有限公司 People flow prediction method, device and system
CN111275479B (en) * 2020-01-07 2023-11-10 北京爱笔科技有限公司 People flow prediction method, device and system
CN113129028A (en) * 2020-01-10 2021-07-16 联洋国融(北京)科技有限公司 Rogue user detection system based on time sequence neural network model
CN111898756A (en) * 2020-08-11 2020-11-06 中国人民解放军海军航空大学 Multi-target information associated neural network loss function calculation method and device
CN111898756B (en) * 2020-08-11 2022-10-11 中国人民解放军海军航空大学 Multi-target information associated neural network loss function calculation method and device
CN112270568B (en) * 2020-11-02 2022-07-12 重庆邮电大学 Order rate prediction method for social e-commerce platform marketing campaign facing hidden information
CN112270568A (en) * 2020-11-02 2021-01-26 重庆邮电大学 Social e-commerce platform marketing activity order rate prediction method facing hidden information
CN112819523A (en) * 2021-01-29 2021-05-18 上海数鸣人工智能科技有限公司 Marketing prediction method combining inner/outer product feature interaction and Bayesian neural network
CN112819523B (en) * 2021-01-29 2024-03-26 上海数鸣人工智能科技有限公司 Marketing prediction method combining inner/outer product feature interaction and Bayesian neural network
CN113222647A (en) * 2021-04-26 2021-08-06 西安点告网络科技有限公司 Advertisement recommendation method, system and storage medium based on click rate estimation model
CN113222647B (en) * 2021-04-26 2023-11-28 西安点告网络科技有限公司 Advertisement recommendation method, system and storage medium based on click rate estimation model
CN113706211A (en) * 2021-08-31 2021-11-26 平安科技（深圳）有限公司 Advertisement click rate prediction method and system based on neural network
CN113706211B (en) * 2021-08-31 2024-04-02 平安科技（深圳）有限公司 Advertisement click rate prediction method and system based on neural network

Similar Documents

Publication Publication Date Title
CN111538912B (en) Content recommendation method, device, equipment and readable storage medium
CN110619540A (en) Click stream estimation method of neural network
CN112381581B (en) Advertisement click rate estimation method based on improved Transformer
CN111222332B (en) Commodity recommendation method combining attention network and user emotion
CN111209386B (en) Personalized text recommendation method based on deep learning
CN111737578B (en) Recommendation method and system
CN110263235A (en) Information pushes object updating method, device and computer equipment
CN113344615B (en) Marketing campaign prediction method based on GBDT and DL fusion model
CN112967088A (en) Marketing activity prediction model structure and prediction method based on knowledge distillation
CN110955826A (en) Recommendation system based on improved recurrent neural network unit
Wang A survey of online advertising click-through rate prediction models
CN112819523B (en) Marketing prediction method combining inner/outer product feature interaction and Bayesian neural network
CN111177579A (en) Integrated diversity enhanced ultra-deep factorization machine model and construction method and application thereof
CN111563770A (en) Click rate estimation method based on feature differentiation learning
CN113591971A (en) User individual behavior prediction method based on DPI time series word embedded vector
CN116051175A (en) Click rate prediction model and prediction method based on depth multi-interest network
CN115840853A (en) Course recommendation system based on knowledge graph and attention network
CN116680363A (en) Emotion analysis method based on multi-mode comment data
CN115080868A (en) Product pushing method, product pushing device, computer equipment, storage medium and program product
CN113449182B (en) Knowledge information personalized recommendation method and system
CN113888238A (en) Advertisement click rate prediction method and device and computer equipment
CN111274359A (en) Query recommendation method and system based on improved VHRED and reinforcement learning
CN116757747A (en) Click rate prediction method based on behavior sequence and feature importance
Kwon et al. Improving RNN based recommendation by embedding-weight tying
CN115689639A (en) Commercial advertisement click rate prediction method based on deep learning

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20191227

RJ01 Rejection of invention patent application after publication