Data prediction method, device and equipment based on factor decomposition machine and partial order relationship

Abstract

The present invention relates to artificial intelligence, and provides a data prediction method, device, equipment and medium based on a factor decomposition machine and a partial order relationship. The method first obtains the data to be predicted and the corresponding first-order feature set and second-order feature set, then obtains the second-order feature partial order relationship extraction strategy, obtains the second-order feature partial order relationship extraction result corresponding to the second-order feature set according to the strategy, then performs weighted summation of each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain the first-order feature result, and finally sums the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result to obtain the prediction result corresponding to the data to be predicted, and sends the prediction result to the user end. It is realized that on the basis of the traditional factor decomposition machine, the feature cross term of the partial order relationship is added, the partial order relationship between the features can be automatically obtained, the model expression ability is enhanced, and the obtained prediction result is more accurate.

Description

Technical Field

[0001] The present invention relates to the field of intelligent decision-making of artificial intelligence, and in particular to a data prediction method, device, computer equipment and storage medium based on a factorization machine and a partial order relationship. Background Art

[0002] Factorization Machine is a classic machine learning algorithm. It is widely used in business scenarios such as data mining and advertising recommendation because it can express the intersection relationship of second-order features. Traditional factorization machines regard feature intersection as disordered, that is, the intersection of features (A, B) and the intersection of features (B, A) obtain the same result. In some specific scenarios (such as insurance purchase scenarios), the intersection between features is ordered. For example, A buys a product for B and B buys a product for A, which actually represent different meanings. These two behaviors will be regarded as one behavior in traditional factorization machines, but obviously these two behaviors are not equivalent. This results in the inability of existing factorization machines to distinguish the intersection order of second-order intersection features, which reduces the accuracy of the prediction results obtained based on the factorization machine. Summary of the invention

[0003] The embodiments of the present invention provide a data prediction method, apparatus, computer device and storage medium based on a factor decomposition machine and a partial order relationship, aiming to solve the problem in the prior art that the factor decomposition machine cannot distinguish the cross order of second-order cross features, resulting in reduced accuracy of the prediction results obtained based on the factor decomposition machine.

[0004] In a first aspect, an embodiment of the present invention provides a data prediction method based on a factorization machine and a partial order relationship, which includes:

[0005] In response to a data upload instruction, acquiring data to be predicted corresponding to the data upload instruction;

[0006] Obtaining a first-order feature set and a second-order feature set of the data to be predicted;

[0007] Obtaining a preset second-order feature partial order relationship extraction strategy, and obtaining a second-order feature partial order relationship extraction result corresponding to the second-order feature set according to the second-order feature partial order relationship extraction strategy;

[0008] Obtaining a first-order feature weight value of each first-order feature in the first-order feature set, and performing weighted summation on each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain a first-order feature result; and

[0009] A preset global bias value is obtained, the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result are summed to obtain a prediction result corresponding to the data to be predicted, and the prediction result is sent to the user end.

[0010] In a second aspect, an embodiment of the present invention provides a data prediction device based on a factorization machine and a partial order relationship, which includes:

[0011] a to-be-predicted data acquisition unit, configured to, in response to a data upload instruction, acquire the to-be-predicted data corresponding to the data upload instruction;

[0012] A feature set acquisition unit, used to acquire a first-order feature set and a second-order feature set of the data to be predicted;

[0013] A partial order relation extraction unit, used to obtain a preset second-order feature partial order relation extraction strategy, and obtain a second-order feature partial order relation extraction result corresponding to the second-order feature set according to the second-order feature partial order relation extraction strategy;

[0014] a first-order feature result obtaining unit, configured to obtain a first-order feature weight value of each first-order feature in the first-order feature set, and perform weighted summation of each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain a first-order feature result; and

[0015] The prediction result acquisition unit is used to obtain a preset global bias value, sum the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result to obtain a prediction result corresponding to the data to be predicted, and send the prediction result to the user end.

[0016] In the third aspect, an embodiment of the present invention further provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the data prediction method based on factor decomposition machine and partial order relationship described in the first aspect is implemented.

[0017] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the data prediction method based on factor decomposition machine and partial order relationship described in the first aspect above.

[0018] The embodiment of the present invention provides a data prediction method, device, computer equipment and storage medium based on factor decomposition machine and partial order relationship, firstly obtain the data to be predicted and its corresponding first-order feature set and second-order feature set, then obtain the preset second-order feature partial order relationship extraction strategy, obtain the second-order feature partial order relationship extraction result corresponding to the second-order feature set according to the second-order feature partial order relationship extraction strategy, then obtain the first-order feature weight value of each first-order feature in the first-order feature set, perform weighted summation on each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain the first-order feature result, finally obtain the pre-set global bias value, sum the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result to obtain the prediction result corresponding to the data to be predicted, and send the prediction result to the user end. It is realized that on the basis of the traditional factor decomposition machine, the feature cross term of the partial order relationship is added, the partial order relationship between the features can be automatically obtained, the model expression ability is enhanced, and the prediction result obtained is more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0020] Figure 1 A schematic diagram of an application scenario of a data prediction method based on a factorization machine and a partial order relationship provided in an embodiment of the present invention;

[0021] Figure 2 A schematic diagram of a flow chart of a data prediction method based on a factorization machine and a partial order relationship provided in an embodiment of the present invention;

[0022] Figure 3 A schematic block diagram of a data prediction device based on a factorization machine and a partial order relationship provided by an embodiment of the present invention;

[0023] Figure 4 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0024] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0025] It should be understood that when used in this specification and the appended claims, the terms "include" and "comprises" indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or combinations thereof.

[0026] It should also be understood that the terms used in this specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. As used in the specification of the present invention and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0027] It should be further understood that the term "and / or" used in the present description and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0028] See also Figure 1 and Figure 2 , Figure 1 A schematic diagram of an application scenario of a data prediction method based on a factorization machine and a partial order relationship provided in an embodiment of the present invention; Figure 2 A flow chart of a data prediction method based on a factor decomposition machine and a partial order relationship provided in an embodiment of the present invention. The data prediction method based on a factor decomposition machine and a partial order relationship is applied in a server, and the method is executed by application software installed in the server.

[0029] like Figure 2 As shown, the method includes steps S101 to S105.

[0030] S101. In response to a data upload instruction, obtain data to be predicted corresponding to the data upload instruction.

[0031] In this embodiment, the technical solution is described with the server as the execution subject. Under the premise of obtaining user authorization and legal acquisition of data, the user behavior data obtained by the user when using the application installed on the user terminal is uploaded to the server. Among them, the data upload instruction is triggered when the application in the user terminal uploads data regularly, or it is triggered by the user operating the user terminal and manually clicking the data upload button. When the data upload instruction is generated in the user terminal, the data upload instruction is sent to the server. When the server receives the data upload instruction, it indicates that the server needs to predict the data. Among them, when the server detects the data upload instruction, the server receives the corresponding data to be predicted while receiving the data upload instruction. In this application, an example is used to illustrate that a user terminal uploads a piece of data to be predicted, and multiple user terminals upload the data to be predicted separately to predict in the server. Its essence is still that the server predicts each uploaded data to be predicted separately, so the processing method when the server receives multiple data to be predicted can refer to the processing method for one data to be predicted.

[0032] Here, a piece of data to be predicted includes N features as an example, wherein the N features include features such as gender and age, and also include features such as consumption information of product 1 (consumption information is more specific such as purchaser information) and consumption information of product 2. The data to be predicted corresponding to the data upload instruction uploaded by the user end is first stored in a corresponding data table in the server for preservation, and the data table also includes N+1 field names, wherein the first N field names of the above N+1 field names correspond one-to-one to the N features included in the data to be predicted, and the remaining 1 field name represents the prediction result and the initial value of the prediction result is a null value (to indicate that the data to be predicted has not yet been predicted).

[0033] S102: Obtain a first-order feature set and a second-order feature set of the data to be predicted.

[0034] In this embodiment, after obtaining the data to be predicted, the first-order features whose individual values ​​can affect the final predicted value are extracted, and the second-order features whose values ​​affect each other and can affect the final predicted value at the same time are also extracted. Among them, each pair of features in the general second-order features are often frequently marked to occur together, and the inner product between the features is large.

[0035] In one embodiment, step S102 includes:

[0036] A pre-trained factor decomposition machine is obtained, and the to-be-predicted data is factorized by the factor decomposition machine to obtain a first-order feature set and a second-order feature set.

[0037] In this embodiment, for any symmetric positive semidefinite matrix W, as long as k is large enough, there must exist a matrix V such that W = V*VT As long as the decomposition dimension k is large enough (such as k is 100), the factorization machine can basically fit any second-order cross feature by decomposition, and the factorization machine can be trained using the gradient descent algorithm.

[0038] For example, the mathematical form of a factorization machine is:

[0039]

[0040] in, is the predicted value of the sample, ω i is the weight of the first-order feature, v i 、v j is feature x i 、x j The hidden vector of the inner product is the second-order cross feature x i x j From the mathematical form, we can see that when the factorization machine generates the second-order cross features, it does not consider the order of the second-order features and converts x i x j and x j x i It can be seen that by inputting any multi-dimensional data to be predicted into the factor decomposition machine, the first-order feature set and the second-order feature set can be quickly obtained.

[0041] In one embodiment, factoring the data to be predicted by the factorization machine to obtain a first-order feature set and a second-order feature set includes:

[0042] The feature value set included in the data to be predicted is obtained, each feature value in the feature value set is converted into a corresponding embedding vector according to a preset conversion strategy, and the input vector composed of the embedding vectors corresponding to each feature value is input into the factor decomposition machine for calculation to obtain a first-order feature set and a second-order feature set.

[0043] In this embodiment, in order to obtain the first-order feature set and the second-order feature set more quickly, as well as the weight corresponding to each first-order feature and the weight corresponding to the second-order feature, the data to be predicted needs to be preprocessed first. Generally, there are features of the types of numerical type (such as the value of features such as age and income), category type (such as gender, etc.), and text type (such as user brief description, etc.) in the data to be predicted. It is impossible to complete the factor decomposition by directly inputting the original data to be predicted into the factor decomposition machine, so it is necessary to convert these types of features into vectors according to the corresponding conversion strategy before factor decomposition can be performed. The conversion strategy includes a numerical type feature conversion sub-strategy, a category type feature conversion sub-strategy, and a text type feature conversion sub-strategy, wherein the numerical type feature conversion sub-strategy is used to convert the numerical type feature into a feature vector, the category type feature conversion sub-strategy is used to convert the category type feature into a feature vector, and the text type feature conversion sub-strategy is used to convert the text type feature into a feature vector. Specifically, when the numerical type features in the data to be predicted are converted according to the numerical type feature conversion sub-strategy, each numerical value is converted into a numerical value with an average value of 0 and a standard deviation of 1 according to numerical standardization as a feature vector; when the category type features in the data to be predicted are converted according to the category type feature conversion sub-strategy, it is converted into a dummy variable with a value range between 0 and 1 as a feature vector; when the text type features in the data to be predicted are converted according to the text type feature conversion sub-strategy, it is first segmented and then keyword extracted to obtain a keyword set, and then each keyword in the above keyword set can be finally converted into a word vector concatenated to form a vector as a feature vector according to the word2vec model. Finally, after obtaining the embedding vector (i.e., feature vector) corresponding to each eigenvalue, the input vector composed of the embedding vector corresponding to each eigenvalue is input into the factor decomposition machine for calculation to obtain a first-order feature set and a second-order feature set. It can be seen that through this conversion method, each eigenvalue can be quantized and converted, which is convenient for input into the factor decomposition machine for calculation.

[0044] S103: Obtain a preset second-order feature partial order relationship extraction strategy, and obtain a second-order feature partial order relationship extraction result corresponding to the second-order feature set according to the second-order feature partial order relationship extraction strategy.

[0045] In this embodiment, when the factor decomposition machine generates the second-order cross features, the order of the second-order features is not considered. i x j and x j x i The second-order features are regarded as the same feature, and in order to obtain the prediction value more accurately in this application, the order of the second-order features needs to be considered. Specifically, the second-order feature partial order relationship extraction strategy can be pre-stored in the server to calculate the second-order feature partial order relationship extraction result based on the sequence of the second-order features.

[0046] In one embodiment, step S103 includes:

[0047] Obtaining the i-th pair of second-order features in the second-order feature set; wherein the initial value of i is 1 and the value range of i is [1, M], where M represents the total number of second-order features in the second-order feature set;

[0048] Obtaining the first feature of the i-th pair of second-order features as the i-th pair of first main features, and obtaining the second feature of the i-th pair of second-order features as the i-th pair of first subordinate features;

[0049] Obtaining the inner product between the i-th pair of first main features and the i-th pair of first subordinate features as the i-th pair of first latent vectors;

[0050] The i-th pair of first extraction results of the i-th pair of second-order features are obtained according to the i-th pair of first main features, the i-th pair of first subordinate features, the i-th pair of first latent vectors and the second-order feature partial order relationship extraction strategy; wherein the i-th pair of first extraction results are expressed as Indicates that the i-th pair of second-order features is represented by x i1 As the main feature and x j1 is a dependent feature, i1 ,v j1 > indicates The inner product of , and j1=i1+1;

[0051] Obtaining the second feature of the i-th pair of second-order features as the i-th pair of second main features, and obtaining the first feature of the i-th pair of second-order features as the i-th pair of second subordinate features;

[0052] Obtaining the inner product between the i-th pair of second main features and the i-th pair of second subordinate features as the i-th pair of second latent vectors;

[0053] The i-th pair of second main features, the i-th pair of second subordinate features, the i-th pair of second latent vectors and the second-order feature partial order relationship extraction strategy are used to obtain the i-th pair of second extraction results; wherein the i-th pair of second extraction results are expressed as Table ith pair of second-order features with x j2 As the main feature and x i2 is a dependent feature, j2 ,v i2 > indicates The inner product of , and i2=j2+1, x j2 With x j1 represents the same feature in the i-th pair of second-order features, x i1 With x i2 ​​represents the same feature in the i-th pair of second-order features;

[0054] The i-th pair of first extraction results and the i-th pair of second extraction results form the i-th pair of second-order feature partial order relationship extraction results;

[0055] Increment the value of i by 1 to update the value of i;

[0056] If it is determined that i does not exceed M, return to the step of obtaining the i-th pair of second-order features in the second-order feature set;

[0057] If it is determined that i exceeds M, the first pair of second-order feature partial order relationship extraction results to the Mth pair of second-order feature partial order relationship extraction results are obtained to form the second-order feature partial order relationship extraction results.

[0058] In this embodiment, in order to express the partial order relationship, in the feature x i 、x j When doing feature crossover, we need to consider two situations, namely x i As the subject, and x j As the subject:

[0059] 1. x i As the subject:

[0060] At this time, the feature cross pair is recorded as The corresponding latent vector is denoted as u i1 , and x j The corresponding latent vector is denoted as v j1 , the corresponding mathematical form is

[0061] 2. x j As the subject:

[0062] At this time, the feature cross pair is recorded as The corresponding latent vector is denoted as u j2 , and x i The corresponding latent vector is denoted as v i2 , the corresponding mathematical form is

[0063] Combining the two cases, the mathematical form of the factorization machine expressing the partial order relationship is:

[0064]

[0065] The factor decomposition machine expressing the partial order relationship is then trained using the gradient descent algorithm. After the training converges, the model can automatically learn the partial order relationship between features.

[0066] In one embodiment, obtaining the inner product between the i-th pair of first main features and the i-th pair of first subordinate features as the i-th pair of first latent vectors includes:

[0067] Obtaining an i-th pair of quadratic term parameters of an i-th pair of second-order features in the second-order feature set, and multiplying the i-th pair of quadratic term parameters by a preset first coefficient to update the i-th pair of quadratic term parameters;

[0068] The i-th symmetric matrix corresponding to the i-th pair of quadratic term parameters;

[0069] Decomposing the i-th symmetric matrix to obtain an i-th decomposition matrix and an i-th transposed matrix of the i-th decomposition matrix; wherein the product of the i-th transposed matrix and the i-th decomposition matrix is ​​equal to the i-th symmetric matrix;

[0070] The i-th column in the i-th decomposition matrix is ​​obtained as the inner product between the i-th pair of first main features and the i-th pair of first subordinate features, and as the i-th pair of first latent vectors.

[0071] In this embodiment, the i-th pair of quadratic term parameters obtained by the initial factor decomposition machine decomposition can be used as the initial data for obtaining the i-th pair of first latent vectors, because the i-th pair of quadratic term parameters obtained by the initial factor decomposition machine decomposition are overall coefficients that do not consider the second-order feature partial order relationship. At this time, the i-th pair of quadratic term parameters can be multiplied by the preset first coefficient (such as setting the first coefficient to 0.6) to update the i-th pair of quadratic term parameters. Moreover, the quadratic term parameters obtained by the initial factor decomposition machine decomposition can form a symmetric matrix. At this time, the symmetric matrix can be decomposed into the i-th decomposition matrix and the i-th transposed matrix of the i-th decomposition matrix. Finally, the i-th pair of first latent vectors can be obtained based on the i-th column in the i-th decomposition matrix. It can be seen that the latent vector can be quickly obtained by this method of decomposing coefficients.

[0072] S104: Obtain a first-order feature weight value of each first-order feature in the first-order feature set, and perform weighted summation on each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain a first-order feature result.

[0073] In this embodiment, when the first-order feature set is obtained based on the initial factor decomposition machine decomposition, the first-order feature weight values ​​corresponding to each first-order feature in the first-order feature set are also obtained, and the first-order feature weight values ​​corresponding to each first-order feature obtained can be used in subsequent calculations. Specifically, the weighted sum of each first-order feature and the corresponding first-order feature weight value in the first-order feature set is performed to obtain the first-order feature result.

[0074] S105, obtaining a preset global bias value, summing the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result to obtain a prediction result corresponding to the data to be predicted, and sending the prediction result to the user end.

[0075] In this embodiment, in order to obtain the final prediction result, it is necessary to obtain the preset global bias value ω 0 , and the first-order characteristic results Then the global offset value, the first-order feature result and the second-order feature partial order relationship extraction result are summed to obtain the prediction result corresponding to the data to be predicted. That is:

[0076] in, It indicates the prediction result. It can be seen that by adding the feature cross term of partial order relationship on the basis of the traditional factor decomposition machine, the partial order relationship between features can be automatically obtained, the model expression ability is enhanced, and the prediction result obtained is more accurate. Finally, the prediction result is sent to the user end so that the prediction result can be viewed intuitively.

[0077] In one embodiment, after step S105, the method further includes:

[0078] If a second-order feature partial order relationship acquisition instruction is received, the second-order feature partial order relationship extraction result is sent to the user end.

[0079] In this embodiment, when the user terminal needs to view the detailed corresponding second-order feature partial order relationship extraction results in addition to the prediction results, the server can send the second-order feature partial order relationship extraction results to the user terminal after detecting that the user terminal has received the second-order feature partial order relationship acquisition instruction, so that the user can view it intuitively.

[0080] The embodiment of the present application can acquire and process relevant data in the server based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0081] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0082] This method adds feature cross terms of partial order relationships on the basis of traditional factor decomposition machines, which can automatically obtain partial order relationships between features, enhance the model's expressiveness, and make the resulting predictions more accurate.

[0083] The embodiment of the present invention further provides a data prediction device based on a factor decomposition machine and a partial order relationship, and the data prediction device based on a factor decomposition machine and a partial order relationship is used to execute any embodiment of the data prediction method based on a factor decomposition machine and a partial order relationship. Figure 3 , Figure 3 It is a schematic block diagram of a data prediction device 100 based on a factorization machine and a partial order relationship provided by an embodiment of the present invention.

[0084] Among them, Figure 3 As shown, the data prediction device 100 based on factor decomposition machine and partial order relationship includes a to-be-predicted data acquisition unit 101, a feature set acquisition unit 102, a partial order relationship extraction unit 103, a first-order feature result acquisition unit 104 and a prediction result acquisition unit 105.

[0085] The to-be-predicted data acquisition unit 101 is configured to, in response to a data upload instruction, acquire the to-be-predicted data corresponding to the data upload instruction.

[0086] In this embodiment, the technical solution is described with the server as the execution subject. Under the premise of obtaining user authorization and legal acquisition of data, the user behavior data obtained by the user when using the application installed on the user terminal is uploaded to the server. Among them, the data upload instruction is triggered when the application in the user terminal uploads data regularly, or it is triggered by the user operating the user terminal and manually clicking the data upload button. When the data upload instruction is generated in the user terminal, the data upload instruction is sent to the server. When the server receives the data upload instruction, it indicates that the server needs to predict the data. Among them, when the server detects the data upload instruction, the server receives the corresponding data to be predicted while receiving the data upload instruction. In this application, an example is used to illustrate that a user terminal uploads a piece of data to be predicted, and multiple user terminals upload the data to be predicted separately to predict in the server. Its essence is still that the server predicts each uploaded data to be predicted separately, so the processing method when the server receives multiple data to be predicted can refer to the processing method for one data to be predicted.

[0087] Here, a piece of data to be predicted includes N features as an example, wherein the N features include features such as gender and age, and also include features such as consumption information of product 1 (consumption information is more specific such as purchaser information) and consumption information of product 2. The data to be predicted corresponding to the data upload instruction uploaded by the user end is first stored in a corresponding data table in the server for preservation, and the data table also includes N+1 field names, wherein the first N field names of the above N+1 field names correspond one-to-one to the N features included in the data to be predicted, and the remaining 1 field name represents the prediction result and the initial value of the prediction result is a null value (to indicate that the data to be predicted has not yet been predicted).

[0088] The feature set acquisition unit 102 is used to acquire a first-order feature set and a second-order feature set of the data to be predicted.

[0089] In this embodiment, after obtaining the data to be predicted, the first-order features whose individual values ​​can affect the final predicted value are extracted, and the second-order features whose values ​​affect each other and can affect the final predicted value at the same time are also extracted. Among them, each pair of features in the general second-order features are often frequently marked to occur together, and the inner product between the features is large.

[0090] In one embodiment, the feature set acquisition unit 102 is specifically used to:

[0091] A pre-trained factor decomposition machine is obtained, and the to-be-predicted data is factorized by the factor decomposition machine to obtain a first-order feature set and a second-order feature set.

[0092] In this embodiment, for any symmetric positive semidefinite matrix W, as long as k is large enough, there must exist a matrix V such that W = V*V T As long as the decomposition dimension k is large enough (such as k is 100), the factorization machine can basically fit any second-order cross feature by decomposition, and the factorization machine can be trained using the gradient descent algorithm.

[0093] For example, the mathematical form of a factorization machine is:

[0094]

[0095] in, is the predicted value of the sample, ω i is the weight of the first-order feature, v i 、v j is feature x i 、x j The hidden vector of the inner product is the second-order cross feature x i x jFrom the mathematical form, we can see that when the factorization machine generates the second-order cross features, it does not consider the order of the second-order features and converts x i x j and x j x i It can be seen that by inputting any multi-dimensional data to be predicted into the factor decomposition machine, the first-order feature set and the second-order feature set can be quickly obtained.

[0096] In one embodiment, factoring the data to be predicted by the factorization machine to obtain a first-order feature set and a second-order feature set includes:

[0097] The feature value set included in the data to be predicted is obtained, each feature value in the feature value set is converted into a corresponding embedding vector according to a preset conversion strategy, and the input vector composed of the embedding vectors corresponding to each feature value is input into the factor decomposition machine for calculation to obtain a first-order feature set and a second-order feature set.

[0098] In this embodiment, in order to obtain the first-order feature set and the second-order feature set more quickly, as well as the weight corresponding to each first-order feature and the weight corresponding to the second-order feature, the data to be predicted needs to be preprocessed first. Generally, there are features of the types of numerical type (such as the value of features such as age and income), category type (such as gender, etc.), and text type (such as user brief description, etc.) in the data to be predicted. It is impossible to complete the factor decomposition by directly inputting the original data to be predicted into the factor decomposition machine, so it is necessary to convert these types of features into vectors according to the corresponding conversion strategy before factor decomposition can be performed. The conversion strategy includes a numerical type feature conversion sub-strategy, a category type feature conversion sub-strategy, and a text type feature conversion sub-strategy, wherein the numerical type feature conversion sub-strategy is used to convert the numerical type feature into a feature vector, the category type feature conversion sub-strategy is used to convert the category type feature into a feature vector, and the text type feature conversion sub-strategy is used to convert the text type feature into a feature vector. Specifically, when the numerical type features in the data to be predicted are converted according to the numerical type feature conversion sub-strategy, each numerical value is converted into a numerical value with an average value of 0 and a standard deviation of 1 according to numerical standardization as a feature vector; when the category type features in the data to be predicted are converted according to the category type feature conversion sub-strategy, it is converted into a dummy variable with a value range between 0 and 1 as a feature vector; when the text type features in the data to be predicted are converted according to the text type feature conversion sub-strategy, it is first segmented and then keyword extracted to obtain a keyword set, and then each keyword in the above keyword set can be finally converted into a word vector concatenated to form a vector as a feature vector according to the word2vec model. Finally, after obtaining the embedding vector (i.e., feature vector) corresponding to each eigenvalue, the input vector composed of the embedding vector corresponding to each eigenvalue is input into the factor decomposition machine for calculation to obtain a first-order feature set and a second-order feature set. It can be seen that through this conversion method, each eigenvalue can be quantized and converted, which is convenient for input into the factor decomposition machine for calculation.

[0099] The partial order relation extraction unit 103 is used to obtain a preset second-order feature partial order relation extraction strategy, and obtain a second-order feature partial order relation extraction result corresponding to the second-order feature set according to the second-order feature partial order relation extraction strategy.

[0100] In this embodiment, when the factor decomposition machine generates the second-order cross features, the order of the second-order features is not considered. i x j and x j x iThe second-order features are regarded as the same feature, and in order to obtain the prediction value more accurately in this application, the order of the second-order features needs to be considered. Specifically, the second-order feature partial order relationship extraction strategy can be pre-stored in the server to calculate the second-order feature partial order relationship extraction result based on the sequence of the second-order features.

[0101] In one embodiment, the partial order relation extraction unit 103 is specifically used for:

[0102] Obtaining the i-th pair of second-order features in the second-order feature set; wherein the initial value of i is 1 and the value range of i is [1, M], where M represents the total number of second-order features in the second-order feature set;

[0103] Obtaining the first feature of the i-th pair of second-order features as the i-th pair of first main features, and obtaining the second feature of the i-th pair of second-order features as the i-th pair of first subordinate features;

[0104] Obtaining the inner product between the i-th pair of first main features and the i-th pair of first subordinate features as the i-th pair of first latent vectors;

[0105] The i-th pair of first extraction results of the i-th pair of second-order features are obtained according to the i-th pair of first main features, the i-th pair of first subordinate features, the i-th pair of first latent vectors and the second-order feature partial order relationship extraction strategy; wherein the i-th pair of first extraction results are expressed as Indicates that the i-th pair of second-order features is represented by x i1 As the main feature and x j1 is a dependent feature, i1 ,v j1 > indicates The inner product of , and j1=i1+1;

[0106] Obtaining the second feature of the i-th pair of second-order features as the i-th pair of second main features, and obtaining the first feature of the i-th pair of second-order features as the i-th pair of second subordinate features;

[0107] Obtaining the inner product between the i-th pair of second main features and the i-th pair of second subordinate features as the i-th pair of second latent vectors;

[0108] The i-th pair of second main features, the i-th pair of second subordinate features, the i-th pair of second latent vectors and the second-order feature partial order relationship extraction strategy are used to obtain the i-th pair of second extraction results; wherein the i-th pair of second extraction results are expressed as Table ith pair of second-order features with x j2 As the main feature and x i2 is a dependent feature, j2 ,v i2 > indicates​​ The inner product of , and i2=j2+1, x j2 With x j1 represents the same feature in the i-th pair of second-order features, x i1 With x i2 represents the same feature in the i-th pair of second-order features;

[0109] The i-th pair of first extraction results and the i-th pair of second extraction results form the i-th pair of second-order feature partial order relationship extraction results;

[0110] Increment the value of i by 1 to update the value of i;

[0111] If it is determined that i does not exceed M, return to the step of obtaining the i-th pair of second-order features in the second-order feature set;

[0112] If it is determined that i exceeds M, the first pair of second-order feature partial order relationship extraction results to the Mth pair of second-order feature partial order relationship extraction results are obtained to form the second-order feature partial order relationship extraction results.

[0113] In this embodiment, in order to express the partial order relationship, in the feature x i 、x j When doing feature crossover, we need to consider two situations, namely x i As the subject, and x j As the subject:

[0114] 1.x i As the subject:

[0115] At this time, the feature cross pair is recorded as The corresponding latent vector is denoted as u i1 , and x j The corresponding latent vector is denoted as v j1 , the corresponding mathematical form is

[0116] 2. x j As the subject:

[0117] At this time, the feature cross pair is recorded as The corresponding latent vector is denoted as u j2 , and x i The corresponding latent vector is denoted as v i2 , the corresponding mathematical form is

[0118] Combining the two cases, the mathematical form of the factorization machine expressing the partial order relationship is:

[0119]

[0120] The factor decomposition machine expressing the partial order relationship is then trained using the gradient descent algorithm. After the training converges, the model can automatically learn the partial order relationship between features.

[0121] In one embodiment, obtaining the inner product between the i-th pair of first main features and the i-th pair of first subordinate features as the i-th pair of first latent vectors includes:

[0122] Obtaining an i-th pair of quadratic term parameters of an i-th pair of second-order features in the second-order feature set, and multiplying the i-th pair of quadratic term parameters by a preset first coefficient to update the i-th pair of quadratic term parameters;

[0123] The i-th symmetric matrix corresponding to the i-th pair of quadratic term parameters;

[0124] Decomposing the i-th symmetric matrix to obtain an i-th decomposition matrix and an i-th transposed matrix of the i-th decomposition matrix; wherein the product of the i-th transposed matrix and the i-th decomposition matrix is ​​equal to the i-th symmetric matrix;

[0125] The i-th column in the i-th decomposition matrix is ​​obtained as the inner product between the i-th pair of first main features and the i-th pair of first subordinate features, and as the i-th pair of first latent vectors.

[0126] In this embodiment, the i-th pair of quadratic term parameters obtained by the initial factor decomposition machine decomposition can be used as the initial data for obtaining the i-th pair of first latent vectors, because the i-th pair of quadratic term parameters obtained by the initial factor decomposition machine decomposition are overall coefficients that do not consider the second-order feature partial order relationship. At this time, the i-th pair of quadratic term parameters can be multiplied by the preset first coefficient (such as setting the first coefficient to 0.6) to update the i-th pair of quadratic term parameters. Moreover, the quadratic term parameters obtained by the initial factor decomposition machine decomposition can form a symmetric matrix. At this time, the symmetric matrix can be decomposed into the i-th decomposition matrix and the i-th transposed matrix of the i-th decomposition matrix. Finally, the i-th pair of first latent vectors can be obtained based on the i-th column in the i-th decomposition matrix. It can be seen that the latent vector can be quickly obtained by this method of decomposing coefficients.

[0127] The first-order feature result acquisition unit 104 is used to obtain the first-order feature weight value of each first-order feature in the first-order feature set, and perform weighted summation on each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain the first-order feature result.

[0128] In this embodiment, when the first-order feature set is obtained based on the initial factor decomposition machine decomposition, the first-order feature weight values ​​corresponding to each first-order feature in the first-order feature set are also obtained, and the first-order feature weight values ​​corresponding to each first-order feature obtained can be used in subsequent calculations. Specifically, the weighted sum of each first-order feature and the corresponding first-order feature weight value in the first-order feature set is performed to obtain the first-order feature result.

[0129] The prediction result acquisition unit 105 is used to obtain a preset global bias value, sum the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result to obtain a prediction result corresponding to the data to be predicted, and send the prediction result to the user end.

[0130] In this embodiment, in order to obtain the final prediction result, it is necessary to obtain the preset global bias value ω 0 , and the first-order characteristic results Then the global offset value, the first-order feature result and the second-order feature partial order relationship extraction result are summed to obtain the prediction result corresponding to the data to be predicted. That is:

[0131] in, It indicates the prediction result. It can be seen that by adding the feature cross term of partial order relationship on the basis of the traditional factor decomposition machine, the partial order relationship between features can be automatically obtained, the model expression ability is enhanced, and the prediction result obtained is more accurate. Finally, the prediction result is sent to the user end so that the prediction result can be viewed intuitively.

[0132] In one embodiment, the data prediction device 100 based on the factorization machine and the partial order relationship further includes:

[0133] The partial order result sending unit is used to send the second-order feature partial order relationship extraction result to the user end if a second-order feature partial order relationship acquisition instruction is received.

[0134] In this embodiment, when the user terminal needs to view the detailed corresponding second-order feature partial order relationship extraction results in addition to the prediction results, the server can send the second-order feature partial order relationship extraction results to the user terminal after detecting that the user terminal has received the second-order feature partial order relationship acquisition instruction, so that the user can view it intuitively.

[0135] The device adds feature cross terms of partial order relations on the basis of traditional factor decomposition machines, which can automatically obtain partial order relations between features, enhance the expression ability of the model, and make the obtained prediction results more accurate.

[0136] The above-mentioned data prediction device based on factor decomposition machine and partial order relation can be implemented in the form of a computer program. The computer program can be used in Figure 4 Runs on the computer device shown.

[0137] See also Figure 4 , Figure 4 5 is a schematic block diagram of a computer device provided in an embodiment of the present invention. The computer device 500 is a server or a server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0138] See also Figure 4 The computer device 500 includes a processor 502 , a memory and a network interface 505 connected via a device bus 501 , wherein the memory may include a storage medium 503 and an internal memory 504 .

[0139] The storage medium 503 may store an operating device 5031 and a computer program 5032. When the computer program 5032 is executed, the processor 502 may execute a data prediction method based on a factorization machine and a partial order relationship.

[0140] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500 .

[0141] The internal memory 504 provides an environment for the operation of the computer program 5032 in the storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute a data prediction method based on a factorization machine and a partial order relationship.

[0142] The network interface 505 is used for network communication, such as providing data information transmission, etc. Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present invention, and does not constitute a limitation on the computer device 500 to which the solution of the present invention is applied. The specific computer device 500 may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0143] The processor 502 is used to run the computer program 5032 stored in the memory to implement the data prediction method based on factor decomposition machine and partial order relationship disclosed in the embodiment of the present invention.

[0144] Those skilled in the art will understand that Figure 4 The embodiments of the computer device shown in the figure do not constitute a limitation on the specific composition of the computer device. In other embodiments, the computer device may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently. For example, in some embodiments, the computer device may only include a memory and a processor. In such embodiments, the structure and function of the memory and the processor are the same as those of the embodiment of the present invention. Figure 4 The embodiments shown are consistent and will not be described again here.

[0145] It should be understood that in the embodiment of the present invention, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0146] In another embodiment of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the data prediction method based on a factor decomposition machine and a partial order relationship disclosed in an embodiment of the present invention is implemented.

[0147] Those skilled in the art can clearly understand that, for the convenience and simplicity of description, the specific working process of the above-described equipment, devices and units can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here. Those of ordinary skill in the art can appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented with electronic hardware, computer software or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to the function. Whether these functions are executed in hardware or software 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 exceed the scope of the present invention.

[0148] In the several embodiments provided by the present invention, it should be understood that the disclosed equipment, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. Units with the same function may also be combined into one unit. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices or units, or may be electrical, mechanical or other forms of connection.

[0149] The units described 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 embodiments of the present invention.

[0150] In addition, each functional unit in each embodiment of the present invention 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.

[0151] If the 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 storage medium. Based on this understanding, the technical solution of the present invention 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. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a background server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), disk or CD-ROM and other media that can store program codes.

[0152] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A data prediction method based on factor decomposition machine and partial order relation, characterized in that: include: In response to a data upload instruction, acquiring data to be predicted corresponding to the data upload instruction; Obtaining a first-order feature set and a second-order feature set of the data to be predicted; Obtaining a preset second-order feature partial order relationship extraction strategy, and obtaining a second-order feature partial order relationship extraction result corresponding to the second-order feature set according to the second-order feature partial order relationship extraction strategy; Obtaining a first-order feature weight value of each first-order feature in the first-order feature set, and performing weighted summation on each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain a first-order feature result; and Obtaining a preset global bias value, summing the global bias value, the first-order feature result, and the second-order feature partial order relationship extraction result to obtain a prediction result corresponding to the data to be predicted, and sending the prediction result to a user terminal; The obtaining of the first-order feature set and the second-order feature set of the data to be predicted includes: Obtain a pre-trained factor decomposition machine, and factorize the data to be predicted by the factor decomposition machine to obtain a first-order feature set and a second-order feature set; The factor decomposition machine is used to factorize the data to be predicted to obtain a first-order feature set and a second-order feature set, including: Acquire a feature value set included in the data to be predicted, convert each feature value in the feature value set into a corresponding embedding vector according to a preset conversion strategy, and input an input vector composed of the embedding vectors corresponding to each feature value into the factor decomposition machine for operation to obtain a first-order feature set and a second-order feature set; The step of obtaining the second-order feature partial order relationship extraction result corresponding to the second-order feature set according to the second-order feature partial order relationship extraction strategy includes: Obtaining the i-th pair of second-order features in the second-order feature set; wherein the initial value of i is 1 and the value range of i is [1, M], where M represents the total number of second-order features in the second-order feature set; Obtaining the first feature of the i-th pair of second-order features as the i-th pair of first main features, and obtaining the second feature of the i-th pair of second-order features as the i-th pair of first subordinate features; Obtaining the inner product between the i-th pair of first main features and the i-th pair of first subordinate features as the i-th pair of first latent vectors; The i-th pair of first extraction results of the i-th pair of second-order features are obtained according to the i-th pair of first main features, the i-th pair of first subordinate features, the i-th pair of first latent vectors and the second-order feature partial order relationship extraction strategy; wherein the i-th pair of first extraction results are expressed as Indicates that the i-th pair of second-order features is represented by x i1 As the main feature and x j1 is a dependent feature, i1 ,v j1 > indicates The inner product of , and j1=i1+1;​ Obtaining the second feature of the i-th pair of second-order features as the i-th pair of second main features, and obtaining the first feature of the i-th pair of second-order features as the i-th pair of second subordinate features; Obtaining the inner product between the i-th pair of second main features and the i-th pair of second subordinate features as the i-th pair of second latent vectors; The i-th pair of second main features, the i-th pair of second subordinate features, the i-th pair of second latent vectors and the second-order feature partial order relationship extraction strategy are used to obtain the i-th pair of second extraction results; wherein the i-th pair of second extraction results are expressed as Table ith pair of second-order features with x j2 As the main feature and x i2 is a dependent feature, <v j2 ,v i2 > indicates The inner product of , and i2=j2+1, x j2 With x j1 represents the same feature in the i-th pair of second-order features, x i1 With x i2 represents the same feature in the i-th pair of second-order features; The i-th pair of first extraction results and the i-th pair of second extraction results form the i-th pair of second-order feature partial order relationship extraction results; Increment the value of i by 1 to update the value of i; If it is determined that i does not exceed M, return to the step of obtaining the i-th pair of second-order features in the second-order feature set; If it is determined that i exceeds M, obtain the first pair of second-order feature partial order relationship extraction results to the Mth pair of second-order feature partial order relationship extraction results to form the second-order feature partial order relationship extraction results; The conversion strategy includes a numerical type feature conversion sub-strategy, a category type feature conversion sub-strategy and a text type feature conversion sub-strategy; wherein the numerical type feature conversion sub-strategy is used to convert numerical type features into feature vectors; the category type feature conversion sub-strategy is used to convert category type features into feature vectors; and the text type feature conversion sub-strategy is used to convert text type features into feature vectors.

2. The data prediction method based on factor decomposition machine and partial order relation according to claim 1, characterized in that: The obtaining the inner product between the i-th pair of first main features and the i-th pair of first subordinate features as the i-th pair of first latent vectors includes: Obtaining an i-th pair of quadratic term parameters of an i-th pair of second-order features in the second-order feature set, and multiplying the i-th pair of quadratic term parameters by a preset first coefficient to update the i-th pair of quadratic term parameters; The i-th symmetric matrix corresponding to the i-th pair of quadratic term parameters; Decomposing the i-th symmetric matrix to obtain an i-th decomposition matrix and an i-th transposed matrix of the i-th decomposition matrix; wherein the product of the i-th transposed matrix and the i-th decomposition matrix is ​​equal to the i-th symmetric matrix; The i-th column in the i-th decomposition matrix is ​​obtained as the inner product between the i-th pair of first main features and the i-th pair of first subordinate features, and as the i-th pair of first latent vectors.

3. The data prediction method based on factor decomposition machine and partial order relation according to claim 1, characterized in that: The method further comprises: obtaining a preset global bias value, summing the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result to obtain a prediction result corresponding to the data to be predicted, and sending the prediction result to the user end; If a second-order feature partial order relationship acquisition instruction is received, the second-order feature partial order relationship extraction result is sent to the user end.

4. A data prediction device based on factor decomposition machine and partial order relation, characterized in that: include: a to-be-predicted data acquisition unit, configured to, in response to a data upload instruction, acquire the to-be-predicted data corresponding to the data upload instruction; A feature set acquisition unit, used to acquire a first-order feature set and a second-order feature set of the data to be predicted; A partial order relation extraction unit, used to obtain a preset second-order feature partial order relation extraction strategy, and obtain a second-order feature partial order relation extraction result corresponding to the second-order feature set according to the second-order feature partial order relation extraction strategy; a first-order feature result obtaining unit, configured to obtain a first-order feature weight value of each first-order feature in the first-order feature set, and perform weighted summation of each first-order feature in the first-order feature set and the corresponding first-order feature weight value to obtain a first-order feature result; and A prediction result acquisition unit, used to acquire a preset global bias value, sum the global bias value, the first-order feature result and the second-order feature partial order relationship extraction result to obtain a prediction result corresponding to the data to be predicted, and send the prediction result to a user terminal; The feature set acquisition unit is specifically used for: Obtain a pre-trained factor decomposition machine, and factorize the data to be predicted by the factor decomposition machine to obtain a first-order feature set and a second-order feature set; The factor decomposition machine is used to factorize the data to be predicted to obtain a first-order feature set and a second-order feature set, including: Acquire a feature value set included in the data to be predicted, convert each feature value in the feature value set into a corresponding embedding vector according to a preset conversion strategy, and input an input vector composed of the embedding vectors corresponding to each feature value into the factor decomposition machine for operation to obtain a first-order feature set and a second-order feature set; The partial order relation extraction unit is specifically used for: Obtaining the i-th pair of second-order features in the second-order feature set; wherein the initial value of i is 1 and the value range of i is [1, M], where M represents the total number of second-order features in the second-order feature set; Obtaining the first feature of the i-th pair of second-order features as the i-th pair of first main features, and obtaining the second feature of the i-th pair of second-order features as the i-th pair of first subordinate features; Obtaining the inner product between the i-th pair of first main features and the i-th pair of first subordinate features as the i-th pair of first latent vectors; The i-th pair of first extraction results of the i-th pair of second-order features are obtained according to the i-th pair of first main features, the i-th pair of first subordinate features, the i-th pair of first latent vectors and the second-order feature partial order relationship extraction strategy; wherein the i-th pair of first extraction results are expressed as Indicates that the i-th pair of second-order features is represented by x i1 As the main feature and x j1 is a dependent feature, i1 ,x j1 > indicates The inner product of , and j1=i1+1;​ Obtaining the second feature of the i-th pair of second-order features as the i-th pair of second main features, and obtaining the first feature of the i-th pair of second-order features as the i-th pair of second subordinate features; Obtaining the inner product between the i-th pair of second main features and the i-th pair of second subordinate features as the i-th pair of second latent vectors; The i-th pair of second main features, the i-th pair of second subordinate features, the i-th pair of second latent vectors and the second-order feature partial order relationship extraction strategy are used to obtain the i-th pair of second extraction results; wherein the i-th pair of second extraction results are expressed as Table ith pair of second-order features with x j2 As the main feature and x i2 is a dependent feature, j2 ,v i2 > indicates The inner product of , and i2=j2+1, x j2 With x j1 represents the same feature in the i-th pair of second-order features, x i1 With x i2 represents the same feature in the i-th pair of second-order features;​ The i-th pair of first extraction results and the i-th pair of second extraction results form the i-th pair of second-order feature partial order relationship extraction results; Increment the value of i by 1 to update the value of i; If it is determined that i does not exceed M, return to the step of obtaining the i-th pair of second-order features in the second-order feature set; If it is determined that i exceeds M, obtain the first pair of second-order feature partial order relationship extraction results to the Mth pair of second-order feature partial order relationship extraction results to form the second-order feature partial order relationship extraction results; The conversion strategy includes a numerical type feature conversion sub-strategy, a category type feature conversion sub-strategy and a text type feature conversion sub-strategy; wherein the numerical type feature conversion sub-strategy is used to convert numerical type features into feature vectors; the category type feature conversion sub-strategy is used to convert category type features into feature vectors; and the text type feature conversion sub-strategy is used to convert text type features into feature vectors.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the data prediction method based on a factorization machine and a partial order relationship is implemented as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed by a processor, causes the processor to perform the data prediction method based on a factorization machine and a partial order relationship as described in any one of claims 1 to 3.

Images