Method for optimizing feature data extraction and related products

Abstract

The embodiment of the present application discloses a method for optimizing feature data extraction and related products thereof, the method comprising: obtaining a plurality of first feature data; grouping the plurality of first feature data to obtain a plurality of first feature data sets; calculating the correlation between every two first feature data in each of the plurality of first feature data sets to obtain a plurality of correlations; sorting the first feature data in each first feature data set according to the plurality of correlations to obtain a first ordered data set; inputting the first ordered data set into a pre-selected and trained first feature extraction model to obtain a second ordered data set; inputting the second ordered data set into a pre-trained neural network model to obtain target data. The embodiment of the present application groups and sorts a plurality of feature data, and then sequentially inputs the sorted feature data into a feature extraction model and a neural network model, thereby reducing the processing time of the feature extraction model on the feature data and improving the efficiency of conversion rate prediction.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a method for optimizing feature data extraction and related products thereof. Background Art

[0002] With the development of science and technology, people receive more and more information every day. Personalized recommendation systems can effectively provide users with personalized decision support. Personalized recommendation systems are an intelligent platform based on massive data mining. Conversion rate prediction is an important task in recommendation systems, which aims to estimate the probability of users purchasing a given item. The current conversion rate prediction pushes relevant content to users based on the correlation between user data.

[0003] In recent years, many ranking models (WDL, AFN, xDeepFM, etc.) have been proposed to learn low-order and high-order interactive features from raw features. For the big data team modeling application platform, in the complete deep learning rank recommendation framework, many different features need to be constructed for different scenarios, so how to use the rank model to improve the efficiency of conversion rate prediction is very necessary. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to provide a method for optimizing feature data extraction and related products, which can effectively further improve the efficiency of conversion rate prediction.

[0005] In a first aspect, an embodiment of the present application provides a method for optimizing feature data extraction, which is applied to an electronic device, and the method includes:

[0006] Acquire a plurality of first feature data;

[0007] Grouping the plurality of first feature data to obtain a plurality of first feature data sets;

[0008] Calculating the correlation between every two first feature data in each of the plurality of first feature data sets to obtain a plurality of correlations;

[0009] Sort the first feature data in each of the first feature data sets according to the multiple correlations to obtain a first ordered data set;

[0010] Determine a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model, where the second ordered data set corresponds to the first ordered data set;

[0011] The second ordered data set is input into a pre-trained second neural network model to obtain target data.

[0012] Optionally, grouping the multiple first feature data to obtain multiple first feature data sets includes: grouping the multiple first feature data according to a cluster analysis algorithm to obtain the multiple first feature data sets.

[0013] Optionally, the clustering analysis algorithm includes a K-means clustering (kmeans) clustering algorithm; grouping the multiple first feature data according to the clustering analysis algorithm to obtain the multiple first feature data sets includes: calculating the similarity between the features of the multiple first feature data according to the kmeans clustering method; grouping the multiple first feature data according to the similarity between the features to obtain the multiple first feature data sets.

[0014] Optionally, sorting the first feature data in each first feature data set according to the multiple relevances to obtain a first ordered data set includes: sorting the first feature data in each first feature data set in descending order according to the multiple relevances to obtain the first ordered data set.

[0015] Optionally, determining the second ordered data set based on the first ordered data set and the trained first feature extraction model includes: converting the first feature data in the first ordered data set into a preprocessing matrix; inputting the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set, wherein the feature pyramid network model is the pretrained first feature extraction model.

[0016] Optionally, the inputting the preprocessing matrix into the feature pyramid network model to obtain the second ordered data set includes: inputting the preprocessing matrix into a first convolution device and performing a maximum pooling operation to obtain a first feature matrix; inputting the first feature matrix into a first transformation module and performing a maximum pooling operation to obtain a first target feature matrix; inputting the first feature matrix into a second convolution device and performing a maximum pooling operation to obtain a second feature matrix; inputting the second feature matrix into a second transformation module and performing a maximum pooling operation to obtain a second target feature matrix; and splicing and fusing the preprocessing matrix, the first target feature matrix, and the second target feature matrix to obtain the second ordered data set.

[0017] Optionally, the preprocessing matrix is ​​input into the feature pyramid network model to obtain the second ordered data set, including: inputting the preprocessing matrix into a first convolution device and performing a maximum pooling operation to obtain a first feature matrix; inputting the first feature matrix into a first transformation module and performing a maximum pooling operation to obtain a first target feature matrix; inputting the first feature matrix into a second convolution device and performing a maximum pooling operation to obtain a second feature matrix; inputting the second feature matrix into a second transformation module and performing a maximum pooling operation to obtain a second target feature matrix; inputting the second feature matrix into a third convolution device and performing a maximum pooling operation to obtain a third feature matrix; inputting the third feature matrix into a third transformation module and performing a maximum pooling operation to obtain a third target feature matrix; splicing and fusing the preprocessing matrix, the first target feature matrix, the second target feature matrix and the third target feature matrix to obtain the second ordered data set.

[0018] In a second aspect, an embodiment of the present application provides a device for optimizing feature data extraction, the device comprising:

[0019] An acquisition unit, used for acquiring a plurality of first feature data;

[0020] A grouping unit, used for grouping the plurality of first feature data to obtain a plurality of first feature data sets;

[0021] A calculation unit, used for calculating the correlation between every two first feature data in each of the plurality of first feature data sets to obtain a plurality of correlations;

[0022] A sorting unit, configured to sort the first feature data in each of the first feature data sets according to the multiple correlations to obtain a first ordered data set;

[0023] a determining unit, configured to determine a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model, wherein the second ordered data set corresponds to the first ordered data set;

[0024] An input unit is used to input the second ordered data set into a pre-trained neural network model to obtain target data.

[0025] Optionally, in the aspect of grouping the multiple first feature data to obtain multiple first feature data sets, the grouping unit is specifically used to: group the multiple first feature data according to a clustering analysis algorithm to obtain the multiple first feature data sets.

[0026] Optionally, the clustering analysis algorithm includes a K-means clustering (kmeans) clustering algorithm; in terms of grouping the multiple first feature data according to the clustering analysis algorithm to obtain the multiple first feature data sets, the grouping unit is specifically used to: calculate the similarity between the features of the multiple first feature data according to the kmeans clustering method; group the multiple first feature data according to the similarity between the features to obtain the multiple first feature data sets.

[0027] Optionally, in terms of sorting the first feature data in each first feature data set according to the multiple relevances to obtain the first ordered data set, the sorting unit is specifically used to: sort the first feature data in each first feature data set in descending order according to the multiple relevances to obtain the first ordered data set.

[0028] Optionally, in determining the second ordered data set based on the first ordered data set and the trained first feature extraction model, the determination unit is specifically used to: convert the first feature data in the first ordered data set into a preprocessing matrix; input the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set, wherein the feature pyramid network model is the pretrained first feature extraction model.

[0029] Optionally, in terms of inputting the preprocessing matrix into the feature pyramid network model to obtain the second ordered data set, the determination unit is specifically used to: input the preprocessing matrix into a first convolution device and perform a maximum pooling operation to obtain a first feature matrix; input the first feature matrix into a first transformation module and perform a maximum pooling operation to obtain a first target feature matrix; input the first feature matrix into a second convolution device and perform a maximum pooling operation to obtain a second feature matrix; input the second feature matrix into a second transformation module and perform a maximum pooling operation to obtain a second target feature matrix; and concatenate and fuse the preprocessing matrix, the first target feature matrix, and the second target feature matrix to obtain the second ordered data set.

[0030] Optionally, in the step of inputting the preprocessing matrix into the feature pyramid network model to obtain the second ordered data set, the determination unit is specifically used to: input the preprocessing matrix into the first convolution device and perform a maximum pooling operation to obtain a first feature matrix; input the first feature matrix into the first transformation module and perform a maximum pooling operation to obtain a first target feature matrix; input the first feature matrix into the second convolution device and perform a maximum pooling operation to obtain a second feature matrix; input the second feature matrix into the second transformation module and perform a maximum pooling operation to obtain a second target feature matrix; input the second feature matrix into the third convolution device and perform a maximum pooling operation to obtain a third feature matrix; input the third feature matrix into the third transformation module and perform a maximum pooling operation to obtain a third target feature matrix; and concatenate and fuse the preprocessing matrix, the first target feature matrix, the second target feature matrix and the third target feature matrix to obtain the second ordered data set.

[0031] In a third aspect, an embodiment of the present application provides an electronic device, comprising a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the program includes instructions for executing the steps of any method in the first aspect of the embodiment of the present application.

[0032] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps described in any method of the first aspect of the embodiment of the present application.

[0033] In a fifth aspect, an embodiment of the present application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to execute some or all of the steps described in any method of the first aspect of the embodiment of the present application. The computer program product may be a software installation package.

[0034] It can be seen that the embodiment of the present application obtains multiple first feature data; groups the multiple first feature data to obtain multiple first feature data sets; calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations; sorts the first feature data in each first feature data set according to the multiple correlations to obtain a first ordered data set; inputs the first ordered data set into a pre-selected and trained first feature extraction model to obtain a second ordered data set; inputs the second ordered data set into a pre-trained neural network model to obtain target data. The embodiment of the present application groups and sorts multiple feature data, and then inputs the sorted feature data into a feature extraction model and a neural network model in sequence to obtain target data, thereby reducing the processing time of the feature extraction model on the feature data and improving the efficiency of conversion rate prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0036] Figure 1 A schematic diagram of an operation for optimizing feature data extraction provided in an embodiment of the present application;

[0037] Figure 2A A flowchart of a method for optimizing feature data extraction provided in an embodiment of the present application;

[0038] Figure 2B A schematic diagram of an FPN model provided in an embodiment of the present application;

[0039] Figure 3 A flowchart of a method for optimizing feature data extraction provided in an embodiment of the present application;

[0040] Figure 4 It is a structural schematic diagram of an electronic device provided in an embodiment of the present application;

[0041] Figure 5 It is a block diagram of the functional units of a device for optimizing feature data extraction provided in an embodiment of the present application. DETAILED DESCRIPTION

[0042] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0043] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices.

[0044] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0045] like Figure 1 As shown, Figure 1 An operation diagram of optimizing feature data extraction provided in an embodiment of the present application. The electronic device groups a plurality of first feature data to obtain a plurality of first feature data sets; the electronic device sorts the first feature data in each first feature data set according to a plurality of correlations to obtain a first ordered data set; the first ordered data set is input into a pre-trained first feature extraction model to obtain a second ordered data set; the second ordered data set is input into a pre-trained neural network model to obtain target data.

[0046] The embodiments of the present application are described in detail below.

[0047] In order to further improve the conversion rate prediction efficiency, the present application provides a method for optimizing feature data extraction, which is applied to electronic devices, specifically, Figure 2A As shown, the method may include but is not limited to the following steps:

[0048] S201, the electronic device acquires a plurality of first feature data;

[0049] Among them, the multiple first feature data are the original model features of multiple users, and the first feature data include structured feature data such as the user's age, gender, education level, user behavior characteristics, and user value characteristics. Multiple first feature data can form a wide table. The first feature data can be obtained through a database, and can be obtained by obtaining user consumption data, user browsing data, user information data, etc. on a web page. The user information data includes the user's age, gender, education level, and name, and the user consumption data includes the user's daily consumption data, insurance consumption data, and financial management consumption data. Daily consumption data can be consumption data on food, clothing, housing, and transportation.

[0050] S202, the electronic device groups the plurality of first feature data to obtain a plurality of first feature data sets;

[0051] In a possible example, the electronic device groups multiple first feature data to obtain multiple first feature data sets, including but not limited to: the electronic device groups multiple first feature data according to a cluster analysis algorithm to obtain multiple first feature data sets.

[0052] The clustering analysis algorithm may be a hierarchical clustering analysis algorithm or a non-hierarchical clustering analysis algorithm, wherein the hierarchical clustering analysis algorithm includes a merging method, a decomposition method, and a dendrogram method. The non-hierarchical clustering analysis algorithm includes a partitioning clustering analysis algorithm, a spectral clustering analysis algorithm, and the like.

[0053] In a specific implementation, the electronic device groups multiple first feature data according to a clustering analysis algorithm to obtain multiple first feature data sets, including but not limited to: the electronic device calculates the similarity between the features of multiple first feature data according to a kmeans clustering method; the electronic device groups multiple first feature data according to the similarity between the features to obtain multiple first feature data sets.

[0054] The electronic device calculating the similarity between the features of the plurality of first feature data according to the kmeans clustering method may include the following specific steps:

[0055] Step 1: First, the electronic device inputs a value of k, where k is the number of groups obtained by clustering a plurality of first feature data using a kmeans algorithm;

[0056] Step 2: The electronic device randomly selects k first feature data from the first data as initial data, namely, initial centroids;

[0057] Step 3, the electronic device calculates the distance between each feature data except the initial data among the multiple first feature data in each of the k groups and the initial data. The first feature data except the initial data among the multiple first feature data in each of the k groups is closest to which initial centroid among the k initial centroids, and the k group data is obtained;

[0058] Step 4, the electronic device selects a new centroid for each of the k grouped data using the kmeans algorithm;

[0059] Step 5: If the distance between the new centroid of each of the k grouped data and the initial centroid is less than the first threshold, it means that the position of the new centroid does not change much and tends to be stable, or converges, that is, the clustering has achieved the desired result, the algorithm terminates, and the first analysis result is obtained;

[0060] Step 6: If the distance between the new centroid and the initial centroid in each of the k grouped data is greater than the first threshold, steps 3-5 need to be iterated.

[0061] The method for calculating the distance between each feature data except the initial data in the multiple first feature data in each group of the k groups in step 3 and the initial data may be the cosine similarity algorithm or the Euclidean distance algorithm.

[0062] S203, the electronic device calculates the correlation between every two first feature data in each first feature data set in the plurality of first feature data sets to obtain a plurality of correlations;

[0063] Among them, the algorithms for calculating the correlation between each two first feature data in each of the multiple first feature data sets include but are not limited to: cosine similarity algorithm, Pearson correlation coefficient algorithm, Jaccard similarity coefficient algorithm, Euclidean distance algorithm, Manhattan distance algorithm, Chebyshev distance algorithm, and Minkowski distance algorithm.

[0064] In a specific implementation, the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations, including but not limited to: the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets according to the cosine similarity algorithm to obtain multiple correlations.

[0065] In a specific implementation, the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations, including but not limited to: the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets according to the Pearson correlation coefficient algorithm to obtain multiple correlations.

[0066] In a specific implementation, the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations, including but not limited to: the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets according to the Jaccard similarity coefficient algorithm to obtain multiple correlations.

[0067] In a specific implementation, the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations, including but not limited to: the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets according to the Euclidean distance algorithm to obtain multiple correlations.

[0068] In a specific implementation, the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations, including but not limited to: the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets according to the Manhattan distance algorithm to obtain multiple correlations.

[0069] In a specific implementation, the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations, including but not limited to: the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets according to the Chebyshev distance algorithm to obtain multiple correlations.

[0070] In a specific implementation, the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations, including but not limited to: the electronic device calculates the correlation between every two first feature data in each of the multiple first feature data sets according to the Minkowski distance algorithm to obtain multiple correlations.

[0071] S204, the electronic device sorts the first feature data in each first feature data set according to a plurality of correlations to obtain a first ordered data set;

[0072] In one possible example, the electronic device sorts the first feature data in each first feature data set according to multiple relevances to obtain a first ordered data set, including but not limited to: the electronic device sorts the first feature data in each first feature data set in descending order according to multiple relevances to obtain a first ordered data set; or, the electronic device sorts the first feature data in each first feature data set in ascending order according to multiple relevances to obtain a first ordered data set.

[0073] It needs to be explained that the electronic device sorts the first feature data in each first feature data set in descending order according to multiple relevances, which can be: the electronic device sorts the first feature data in each first feature data set from top to descending order according to multiple relevances; or, the electronic device sorts the first feature data in each first feature data set from left to right in descending order according to multiple relevances; or, the electronic device sorts the first feature data in each first feature data set from left to right in descending order according to multiple relevances; or, the electronic device sorts the first feature data in each first feature data set from right to left in descending order according to multiple relevances.

[0074] It needs to be explained that the electronic device sorts the first feature data in each first feature data set in ascending order according to multiple relevances, which can be: the electronic device sorts the first feature data in each first feature data set in ascending order from top to bottom according to multiple relevances; or, the electronic device sorts the first feature data in each first feature data set in ascending order from bottom to ascending according to multiple relevances; or, the electronic device sorts the first feature data in each first feature data set in ascending order from left to right to bottom according to multiple relevances; or, the electronic device sorts the first feature data in each first feature data set in ascending order from left to right to bottom according to multiple relevances; or, the electronic device sorts the first feature data in each first feature data set in ascending order from right to left to bottom according to multiple relevances.

[0075] S205. The electronic device determines a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model;

[0076] The second ordered data set corresponds to the first ordered data set.

[0077] Among them, Figure 2B As shown, Figure 2B Schematic diagram of a feature graph pyramid network (FPN) model. The pre-trained first feature extraction model is a feature graph pyramid network FPN model.

[0078] In one possible example, the electronic device determines a second ordered data set based on the first ordered data set and the trained first feature extraction model, including but not limited to: converting the first feature data in the first ordered data set into a preprocessing matrix; inputting the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set, wherein the feature pyramid network model is the pretrained first feature extraction model.

[0079] In a specific implementation, the electronic device inputs the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set, including but not limited to: the electronic device inputs the preprocessing matrix into a first convolution device and performs a maximum pooling operation to obtain a first feature matrix; the electronic device inputs the first feature matrix into a first transformation module and performs a maximum pooling operation to obtain a first target feature matrix; the electronic device inputs the first feature matrix into a second convolution device and performs a maximum pooling operation to obtain a second feature matrix; the electronic device inputs the second feature matrix into a second transformation module and performs a maximum pooling operation to obtain a second target feature matrix; the electronic device concatenates and fuses the preprocessing matrix, the first target feature matrix, and the second target feature matrix to obtain a second ordered data set.

[0080] The preprocessing matrix includes n first features, and the dimension of the features of the preprocessing matrix is ​​k.

[0081] For example, the electronic device inputting the preprocessing matrix into the FPN model may be the following specific steps:

[0082] 1. The electronic device input is n features in the preprocessing matrix, and the dimension of the feature is k. Assume that here is (n, k) = (16, 32),

[0083] 2. The electronic device obtains (n / 2,k,c)=(16,32,4) through c=4 one-dimensional convolutions conv(c,1)=(4,1) and padding=1, and then obtains the first characteristic matrix of (8,32,4) through one-dimensional max polling=(2,1).

[0084] 3. The electronic device inputs the first feature matrix into the reshaped matrix to obtain (8*4,32), and then undergoes the maximum pooling operation (max pooling) = (2,1) to obtain the first target feature matrix of (16,32) (obtaining a new feature matrix new feature); the first feature matrix is ​​input into c = 4 one-dimensional convolutions conv(c,1) = (4,1) and padding = 1, one-dimensional max polling = (2,1), reshaped to obtain (8*4,32), and then undergoes max pooling = (2,1) to obtain the second target feature matrix (feature map with a larger receptive field).

[0085] 4. The electronic device combines and fuses the preprocessing matrix, the first target feature matrix, and the second target feature matrix into fourth feature data, that is, synthesizes more features to obtain a second ordered data set.

[0086] In a specific implementation, the electronic device inputs the preprocessing matrix into the feature pyramid network model FPN to obtain the second ordered data set, including but not limited to: the electronic device inputs the preprocessing matrix into the first convolution device and performs a maximum pooling operation to obtain a first feature matrix; the electronic device inputs the first feature matrix into the first transformation module and performs a maximum pooling operation to obtain a first target feature matrix; the electronic device inputs the first feature matrix into the second convolution device and performs a maximum pooling operation to obtain a second feature matrix; the electronic device inputs the second feature matrix into the second transformation module and performs a maximum pooling operation to obtain a second target feature matrix; the electronic device inputs the second feature matrix into the third convolution device and performs a maximum pooling operation to obtain a third feature matrix; the electronic device inputs the third feature matrix into the third transformation module and performs a maximum pooling operation to obtain a third target feature matrix; the electronic device splices and fuses the preprocessing matrix, the first target feature matrix, the second target feature matrix and the third target feature matrix to obtain a second ordered data set.

[0087] For example, the electronic device inputting the preprocessing matrix into the FPN model may be the following specific steps:

[0088] 1. The electronic device input is n features in the preprocessing matrix, and the dimension of the feature is k. Assume that here is (n, k) = (16, 32);

[0089] 2. The electronic device obtains (n / 2,k,c)=(16,32,4) through c=4 one-dimensional convolutions conv(c,1)=(4,1) and padding=1, and then obtains the first characteristic matrix of (8,32,4) through one-dimensional max polling=(2,1);

[0090] 3. The electronic device inputs the first feature matrix into the reshape to obtain a matrix of (8*4,32), and then performs a maximum pooling operation (max pooling) = (2,1) to obtain a first target feature matrix of (16,32) (obtaining a new feature matrix new feature);

[0091] 4. The electronic device inputs the first feature matrix into c = 4 one-dimensional convolutions conv(c, 1) = (4, 1) and padding = 1 to obtain a second feature matrix;

[0092] 5. The electronic device inputs the second feature matrix into one-dimensional max polling = (2, 1), reshapes it to (8*4, 32), and then passes through max pooling = (2, 1) to obtain the second target feature matrix (feature map with a larger receptive field);

[0093] 6. The electronic device inputs the second feature matrix into c = 4 one-dimensional convolutions conv(c, 1) = (4, 1) and padding = 1 to obtain a third feature matrix;

[0094] 7. The electronic device inputs the third feature matrix into one-dimensional max polling = (2, 1), reshapes to (8*4, 32), and then passes max pooling = (2, 1) to obtain the third target feature matrix;

[0095] 8. The electronic device combines and fuses the preprocessing matrix, the first target feature matrix, the second target feature matrix and the third target feature matrix into fourth feature data, that is, synthesizes more features to obtain a second ordered data set.

[0096] In a specific implementation, the electronic device inputs the preprocessing matrix into the feature pyramid network model FPN to obtain a second ordered data set, including but not limited to:

[0097] Step 1: The electronic device inputs the preprocessing matrix into the first convolution device and performs a maximum pooling operation to obtain a first feature matrix;

[0098] Step 2: The electronic device inputs the first feature matrix into the first transformation module and performs a maximum pooling operation to obtain a first target feature matrix;

[0099] Step 3: The electronic device inputs the first feature matrix into the second convolution device and performs a maximum pooling operation to obtain a second feature matrix;

[0100] Step 4: The electronic device inputs the second feature matrix into the second transformation module and performs a maximum pooling operation to obtain a second target feature matrix;

[0101] Step 5: The electronic device inputs the second feature matrix into the third convolution device and performs a maximum pooling operation to obtain a third feature matrix;

[0102] Step 6: The electronic device inputs the third feature matrix into the third transformation module and performs a maximum pooling operation to obtain a third target feature matrix;

[0103] Repeat steps 5 and 6 continuously to obtain N-3 target feature matrices, namely the 3+1th target feature matrix and ... the Nth target feature matrix, where N is a positive integer greater than three.

[0104] The electronic device concatenates and fuses the preprocessing matrix, the first target feature matrix, the second target feature matrix, the third target feature matrix, ... and the Nth target feature matrix to obtain a second ordered data set.

[0105] S206: The electronic device inputs the second ordered data set into a pre-trained neural network model to obtain target data.

[0106] The neural network model can be any one of a feedforward neural network, a recurrent network, and a symmetrically connected network. The neural network model can also be other neural network models, which are not restricted here. For example, convolutional neural network CNN, recurrent neural network (recursive neural network) RNN, deep belief network DBN, generative adversarial network GAN

[0107] It can be seen that the embodiment of the present application obtains multiple first feature data; groups the multiple first feature data to obtain multiple first feature data sets; calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations; sorts the first feature data in each first feature data set according to the multiple correlations to obtain a first ordered data set; inputs the first ordered data set into a pre-selected and trained first feature extraction model to obtain a second ordered data set; inputs the second ordered data set into a pre-trained neural network model to obtain target data. The embodiment of the present application groups and sorts multiple feature data, and then inputs the sorted feature data into a feature extraction model and a neural network model in sequence to obtain target data, thereby reducing the processing time of the feature extraction model on the feature data and improving the efficiency of conversion rate prediction.

[0108] The following is a detailed description of the embodiments of the present application through a specific example.

[0109] With the above Figure 2A The embodiment shown is consistent with that shown in FIG. Figure 3 , Figure 3 1 is a flow chart of a method for optimizing feature data extraction provided in an embodiment of the present application, which is applied to electronic equipment, and the method includes:

[0110] S301, the electronic device acquires a plurality of first feature data;

[0111] S302, the electronic device calculates similarities between features of a plurality of first feature data according to a kmeans clustering algorithm;

[0112] S303, the electronic device groups the plurality of first feature data according to similarities between the features to obtain a plurality of first feature data sets;

[0113] S304, the electronic device calculates the correlation between every two first feature data in each first feature data set in the plurality of first feature data sets to obtain a plurality of correlations;

[0114] S305. The electronic device sorts the first feature data in each first feature data set in descending order according to multiple correlations to obtain a first ordered data set; or sorts the first feature data in each first feature data set in ascending order according to multiple correlations to obtain a first ordered data set;

[0115] S306, the electronic device inputs the first ordered data set into the FPN model to obtain a second ordered data set;

[0116] S307: The electronic device inputs the second ordered data set into a pre-trained neural network model to obtain target data.

[0117] It can be seen that in the embodiment of the present application, the electronic device obtains multiple first feature data; calculates the similarity between the features of the multiple first feature data according to the kmeans clustering method; groups the multiple first feature data according to the similarity between the features to obtain multiple first feature data sets; calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations; the electronic device sorts the first feature data in each first feature data set in descending order according to the multiple correlations to obtain a first ordered data set; or, sorts the first feature data in each first feature data set in ascending order according to the multiple correlations to obtain a first ordered data set; the electronic device inputs the first ordered data set into the FPN model to obtain a second ordered data set; the electronic device inputs the second ordered data set into a pre-trained neural network model to obtain target data. The embodiment of the present application groups and sorts multiple feature data, and then inputs the sorted feature data into the FPN model and the neural network model in sequence to obtain target data, thereby reducing the processing time of the feature extraction model for the feature data and improving the efficiency of conversion rate prediction.

[0118] See also Figure 4 , Figure 44 is a structural diagram of an electronic device 400 provided in an embodiment of the present application. As shown in the figure, the electronic device 400 includes an application processor 410, a memory 420, a communication interface 430, and one or more programs 421, wherein the one or more programs 421 are stored in the above-mentioned memory 420 and are configured to be executed by the above-mentioned application processor 410, and the one or more programs 421 include for performing the following steps:

[0119] Acquire a plurality of first feature data;

[0120] Grouping the plurality of first feature data to obtain a plurality of first feature data sets;

[0121] Calculating the correlation between every two first feature data in each of the plurality of first feature data sets to obtain a plurality of correlations;

[0122] Sort the first feature data in each of the first feature data sets according to the multiple correlations to obtain a first ordered data set;

[0123] Determine a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model, where the second ordered data set corresponds to the first ordered data set;

[0124] The second ordered data set is input into a pre-trained neural network model to obtain target data.

[0125] It can be seen that the embodiment of the present application obtains multiple first feature data; groups the multiple first feature data to obtain multiple first feature data sets; calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations; sorts the first feature data in each first feature data set according to the multiple correlations to obtain a first ordered data set; determines a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model, the second ordered data set corresponding to the first ordered data set; inputs the second ordered data set into a pre-trained neural network model to obtain target data. The embodiment of the present application groups and sorts multiple feature data, and then inputs the sorted feature data into a feature extraction model and a neural network model in sequence to obtain target data, thereby reducing the processing time of the feature extraction model on the feature data and improving the efficiency of conversion rate prediction.

[0126] In one possible example, in terms of grouping the multiple first feature data to obtain multiple first feature data sets, the one or more programs 421 include methods for performing the following steps: grouping the multiple first feature data according to a clustering analysis algorithm to obtain the multiple first feature data sets.

[0127] In one possible example, the clustering analysis algorithm includes a K-means clustering (kmeans) clustering algorithm; in terms of grouping the multiple first feature data according to the clustering analysis algorithm to obtain the multiple first feature data sets, the one or more programs 421 include methods for performing the following steps: calculating the similarity between the features of the multiple first feature data according to the kmeans clustering method; grouping the multiple first feature data according to the similarity between the features to obtain the multiple first feature data sets.

[0128] In one possible example, in terms of sorting the first feature data in each first feature data set according to the multiple relevances to obtain the first ordered data set, the one or more programs 421 include the steps of: sorting the first feature data in each first feature data set in descending order according to the multiple relevances to obtain the first ordered data set.

[0129] In one possible example, in determining the second ordered data set based on the first ordered data set and the trained first feature extraction model, the one or more programs 421 include the steps of: converting the first feature data in the first ordered data set into a preprocessing matrix; inputting the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set, wherein the feature pyramid network model is the pretrained first feature extraction model.

[0130] In one possible example, in terms of inputting the preprocessing matrix into the feature pyramid network model to obtain the second ordered data set, the one or more programs 421 include the steps of: inputting the preprocessing matrix into a first convolution device and performing a maximum pooling operation to obtain a first feature matrix; inputting the first feature matrix into a first transformation module and performing a maximum pooling operation to obtain a first target feature matrix; inputting the first feature matrix into a second convolution device and performing a maximum pooling operation to obtain a second feature matrix; inputting the second feature matrix into a second transformation module and performing a maximum pooling operation to obtain a second target feature matrix; and concatenating and fusing the preprocessing matrix, the first target feature matrix, and the second target feature matrix to obtain the second ordered data set.

[0131] In one possible example, in terms of inputting the preprocessing matrix into the feature pyramid network model to obtain the second ordered data set, the one or more programs 421 include the steps of: inputting the preprocessing matrix into the first convolution device and performing a maximum pooling operation to obtain a first feature matrix; inputting the first feature matrix into the first transformation module and performing a maximum pooling operation to obtain a first target feature matrix; inputting the first feature matrix into the second convolution device and performing a maximum pooling operation to obtain a second feature matrix; inputting the second feature matrix into the second transformation module and performing a maximum pooling operation to obtain a second target feature matrix; inputting the second feature matrix into the third convolution device and performing a maximum pooling operation to obtain a third feature matrix; inputting the third feature matrix into the third transformation module and performing a maximum pooling operation to obtain a third target feature matrix; splicing and fusing the preprocessing matrix, the first target feature matrix, the second target feature matrix and the third target feature matrix to obtain the second ordered data set.

[0132] The above mainly introduces the scheme of the embodiment of the present application from the perspective of the execution process on the method side. It is understandable that, in order to realize the above functions, the electronic device includes a hardware structure and / or software module corresponding to the execution of each function. It should be easily appreciated by those skilled in the art that, in combination with the units and algorithm steps of each example described in the embodiments provided herein, the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present application.

[0133] The embodiment of the present application can divide the electronic device into functional units according to the above method example. For example, each functional unit can be divided according to each function, or two or more functions can be integrated into one processing unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. It should be noted that the division of units in the embodiment of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation.

[0134] Figure 5 : is a functional unit composition block diagram of the device 500 for optimizing feature data extraction involved in the embodiment of the present application. The device 500 for optimizing feature data extraction includes:

[0135] An acquisition unit 501 is used to acquire a plurality of first feature data;

[0136] A grouping unit 502 is used to group the plurality of first feature data to obtain a plurality of first feature data sets;

[0137] A calculation unit 503 is used to calculate the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations;

[0138] A sorting unit 504, configured to sort the first feature data in each of the first feature data sets according to the multiple correlations to obtain a first ordered data set;

[0139] A determination unit 505 is used to input the first ordered data set into a pre-trained first feature extraction model to obtain a second ordered data set;

[0140] The input unit 506 is used to input the second ordered data set into a pre-trained neural network model to obtain target data.

[0141] The device 500 for optimizing feature data extraction may further include a storage unit 507 for storing program codes and data of the electronic device. The storage unit 507 may be a memory.

[0142] It can be seen that the embodiment of the present application obtains multiple first feature data; groups the multiple first feature data to obtain multiple first feature data sets; calculates the correlation between every two first feature data in each of the multiple first feature data sets to obtain multiple correlations; sorts the first feature data in each first feature data set according to the multiple correlations to obtain a first ordered data set; inputs the first ordered data set into a pre-selected and trained first feature extraction model to obtain a second ordered data set; inputs the second ordered data set into a pre-trained neural network model to obtain target data. The embodiment of the present application groups and sorts multiple feature data, and then inputs the sorted feature data into a feature extraction model and a neural network model in sequence to obtain target data, thereby reducing the processing time of the feature extraction model on the feature data and improving the efficiency of conversion rate prediction.

[0143] In a possible example, in terms of grouping the multiple first feature data to obtain multiple first feature data sets, the grouping unit 502 is specifically used to: group the multiple first feature data according to a clustering analysis algorithm to obtain the multiple first feature data sets.

[0144] In one possible example, the clustering analysis algorithm includes a K-means clustering (kmeans) clustering algorithm; in terms of grouping the multiple first feature data according to the clustering analysis algorithm to obtain the multiple first feature data sets, the grouping unit 502 is specifically used to: calculate the similarity between the features of the multiple first feature data according to the kmeans clustering method; group the multiple first feature data according to the similarity between the features to obtain the multiple first feature data sets.

[0145] In one possible example, in terms of sorting the first feature data in each first feature data set according to the multiple relevances to obtain the first ordered data set, the sorting unit 504 is specifically used to: sort the first feature data in each first feature data set in descending order according to the multiple relevances to obtain the first ordered data set.

[0146] In a possible example, in determining the second ordered data set based on the first ordered data set and the trained first feature extraction model, the determination unit 505 is specifically used to: convert the first feature data in the first ordered data set into a preprocessing matrix; input the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set, wherein the feature pyramid network model is the pretrained first feature extraction model.

[0147] In a possible example, in terms of inputting the preprocessing matrix into the feature pyramid network model to obtain the second ordered data set, the determination unit 505 is specifically used to: input the preprocessing matrix into the first convolution device and perform a maximum pooling operation to obtain a first feature matrix; input the first feature matrix into the first transformation module and perform a maximum pooling operation to obtain a first target feature matrix; input the first feature matrix into the second convolution device and perform a maximum pooling operation to obtain a second feature matrix; input the second feature matrix into the second transformation module and perform a maximum pooling operation to obtain a second target feature matrix; and concatenate and fuse the preprocessing matrix, the first target feature matrix, and the second target feature matrix to obtain the second ordered data set.

[0148] In one possible example, in terms of inputting the first ordered data set into the FPN model to obtain the second ordered data set, the determination unit is specifically used to: convert the first feature data in the first ordered data set into a preprocessing matrix; input the preprocessing matrix into the first convolution device and perform a maximum pooling operation to obtain a first feature matrix; input the first feature matrix into the first transformation module and perform a maximum pooling operation to obtain a first target feature matrix; input the first feature matrix into the second convolution device and perform a maximum pooling operation to obtain a second feature matrix; input the second feature matrix into the second transformation module and perform a maximum pooling operation to obtain a second target feature matrix; input the second feature matrix into the third convolution device and perform a maximum pooling operation to obtain a third feature matrix; input the third feature matrix into the third transformation module and perform a maximum pooling operation to obtain a third target feature matrix; splice and fuse the preprocessing matrix, the first target feature matrix, the second target feature matrix and the third target feature matrix to obtain the second ordered data set.

[0149] An embodiment of the present application also provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps of any method described in the above method embodiments, and the above computer includes an electronic device.

[0150] The embodiment of the present application also provides a computer program product, the computer program product includes a non-transitory computer-readable storage medium storing a computer program, the computer program is operable to cause a computer to execute some or all of the steps of any method described in the method embodiment. The computer program product may be a software installation package, and the computer includes an electronic device.

[0151] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

[0152] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0153] In the several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the above-mentioned units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0154] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0155] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0156] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a memory, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the above-mentioned methods of each embodiment of the present application. The aforementioned memory includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, disk or CD-ROM and other media that can store program codes.

[0157] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable memory, and the memory can include: a flash drive, a read-only memory (English: Read-Only Memory, abbreviated as: ROM), a random access memory (English: Random Access Memory, abbreviated as: RAM), a magnetic disk or an optical disk, etc.

[0158] The embodiments of the present application are introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method and core idea of ​​the present application. At the same time, for general technical personnel in this field, according to the idea of ​​the present application, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for optimizing feature data extraction, characterized in that: Applied to electronic equipment, the method comprises: When estimating the probability of a user purchasing a given item, obtaining a plurality of first feature data includes: obtaining the plurality of first feature data by obtaining user consumption data, user browsing data and user information data on a webpage, wherein the user information data includes the user's age, gender, education level and name, the user consumption data includes the user's daily consumption data, insurance consumption data and financial management consumption data, the first feature data is the user's original model feature, and the first feature data includes the user's age, gender, education level, user behavior feature and user value feature; Grouping the plurality of first feature data to obtain a plurality of first feature data sets; Calculating the correlation between every two first feature data in each of the plurality of first feature data sets to obtain a plurality of correlations; Sort the first feature data in each of the first feature data sets according to the multiple correlations to obtain a first ordered data set; Determine a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model, where the second ordered data set corresponds to the first ordered data set; The second ordered data set is input into a pre-trained neural network model to obtain target data.

2. The method according to claim 1, characterized in that The step of grouping the plurality of first feature data to obtain a plurality of first feature data sets includes: The plurality of first feature data are grouped according to a cluster analysis algorithm to obtain the plurality of first feature data sets.

3. The method according to claim 2, characterized in that The cluster analysis algorithm includes a K-means clustering (kmeans) clustering algorithm; The step of grouping the plurality of first feature data according to a cluster analysis algorithm to obtain the plurality of first feature data sets includes: Calculating similarities between features of the plurality of first feature data according to a kmeans clustering algorithm; The plurality of first feature data are grouped according to similarities between the features to obtain the plurality of first feature data sets.

4. The method according to claim 1, characterized in that: The step of sorting the first feature data in each first feature data set according to the multiple correlations to obtain a first ordered data set includes: According to the multiple correlations, the first feature data in each first feature data set are sorted in descending order to obtain the first ordered data set; or, According to the multiple correlations, the first feature data in each first feature data set are sorted in ascending order to obtain the first ordered data set.

5. The method according to claim 1, characterized in that The step of determining a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model comprises: Converting the first feature data in the first ordered data set into a preprocessing matrix; The preprocessing matrix is ​​input into a feature pyramid network model to obtain the second ordered data set, wherein the feature pyramid network model is the pre-trained first feature extraction model.

6. The method according to claim 5, characterized in that The step of inputting the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set includes: Input the preprocessing matrix into a first convolution device and perform a maximum pooling operation to obtain a first feature matrix; Inputting the first feature matrix into a first transformation module and performing a maximum pooling operation to obtain a first target feature matrix; Inputting the first feature matrix into a second convolution device and performing a maximum pooling operation to obtain a second feature matrix; Inputting the second feature matrix into a second transformation module and performing a maximum pooling operation to obtain a second target feature matrix; The preprocessing matrix, the first target feature matrix and the second target feature matrix are concatenated and fused to obtain the second ordered data set.

7. The method according to claim 5, characterized in that The step of inputting the preprocessing matrix into a feature pyramid network model to obtain the second ordered data set includes: Input the preprocessing matrix into a first convolution device and perform a maximum pooling operation to obtain a first feature matrix; Inputting the first feature matrix into a first transformation module and performing a maximum pooling operation to obtain a first target feature matrix; Inputting the first feature matrix into a second convolution device and performing a maximum pooling operation to obtain a second feature matrix; Inputting the second feature matrix into a second transformation module and performing a maximum pooling operation to obtain a second target feature matrix; Inputting the second feature matrix into a third convolution device and performing a maximum pooling operation to obtain a third feature matrix; Inputting the third feature matrix into a third transformation module and performing a maximum pooling operation to obtain a third target feature matrix; The preprocessing matrix, the first target feature matrix, the second target feature matrix and the third target feature matrix are concatenated and fused to obtain the second ordered data set.

8. A device for optimizing feature data extraction, characterized in that: The device comprises: An acquisition unit, used for acquiring a plurality of first feature data; A grouping unit, used for grouping the plurality of first feature data to obtain a plurality of first feature data sets; a calculation unit, configured to calculate the correlation between every two first feature data in each of the plurality of first feature data sets to obtain a plurality of correlations; A sorting unit, configured to sort the first feature data in each of the first feature data sets according to the multiple correlations to obtain a first ordered data set; a determining unit, configured to determine a second ordered data set according to the first ordered data set and a pre-trained first feature extraction model, wherein the second ordered data set corresponds to the first ordered data set; An input unit is used to input the second ordered data set into a pre-trained neural network model to obtain target data.

9. An electronic device, characterized in that: The method comprises a processor, a memory, a communication interface, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the processor, and the programs include instructions for executing the steps in the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: It stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute the method according to any one of claims 1 to 7.

Images