Information processing method, device, computing equipment and medium

Through Granger causal analysis and attention mechanism, combined with personalized and common causal relationships, variable association weights are calculated, which solves the problem of inaccurate calculation of the correlation degree of object variables with fewer sample data, and improves the accuracy of analysis and prediction effect. .

CN113379436BActive Publication Date: 2024-11-01ALIBABA GROUP HOLDING LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010155687.1
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-03-09
Publication Date
2024-11-01
Estimated Expiration
2040-03-09

AI Technical Summary

Technical Problem

In the prior art, when analyzing the degree of variable correlation between multiple objects, objects with less sample data cannot accurately determine the causal relationship between variables, resulting in inaccurate calculation of the degree of correlation.

Method used

By obtaining the time series of each variable of the target object, using the Granger causal analysis method and attention mechanism, combining personalized causal weights and common causal weights, the association weights of each variable and the target object are calculated, and then the correlation weights of each variable and the target object are determined Target correlation weights between variables.

Benefits of technology

The accuracy of calculating the correlation degree between variables is improved, and the correlation weight between variables of target objects with less sample data can be accurately determined, which enhances the accuracy of the predicted value of the target variable at the target time point.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113379436B_ABST
    Figure CN113379436B_ABST
Patent Text Reader

Abstract

The present invention discloses an information processing method, device, computing equipment and medium. The method comprises: obtaining a time series of each first variable among a plurality of first variables of a target object; determining a first association weight of each first variable with a second variable of the target object based on each time series; obtaining a target association weight of each first variable with the second variable based on each first association weight and preset association weight information corresponding to the target object; wherein the preset association weight information corresponding to the target object comprises a second association weight of each first variable with the second variable, and the second association weight is the association degree between the first variable of each object among the plurality of objects and the second variable of the object. According to an embodiment of the present invention, the accuracy of calculating the association degree between variables is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing, and in particular to an information processing method, device, computing equipment and medium. Background Art

[0002] In practical applications, an object often corresponds to multiple variables of different dimensions. For example, a product has variables of different dimensions such as sales volume, click-through rate, and pageviews. For a target variable of an object, other variables of the object may have an association relationship with the target variable, but the degree of association may vary. In order to analyze the degree of association between the target variable of an object and other variables, it is necessary to collect the time series of other variables, that is, multivariate time series. The time series of a variable refers to a sequence of values ​​of the variable arranged in the order of their occurrence.

[0003] Different objects may have the same variables. For example, different products may have variables such as sales volume, click-through rate, and pageviews. However, the correlation between the same variables may be different in different objects. For example, the correlation between pageviews and sales volume may be different for different products.

[0004] Therefore, the current analysis of the correlation between variables is based on the object, that is, the variables of each object are analyzed separately. However, the sample data of some objects are relatively small, and it is impossible to accurately obtain the correlation between the variables of the object. Summary of the invention

[0005] Embodiments of the present invention provide an information processing method, apparatus, computing device, and medium, which can improve the accuracy of calculating the degree of association between variables.

[0006] In a first aspect, an embodiment of the present invention provides an information processing method, including:

[0007] Obtain a time series of each first variable among a plurality of first variables of a target object;

[0008] Based on each of the time series, determining a first association weight of each of the first variables with a second variable of the target object;

[0009] Based on each of the first association weights and the preset association weight information corresponding to the target object, a target association weight between each of the first variables and the second variable is obtained; wherein the preset association weight information corresponding to the target object includes a second association weight between each of the first variables and the second variable, and the second association weight is a degree of association between the first variable of each object in a plurality of objects and the second variable of the object.

[0010] In a second aspect, an embodiment of the present invention provides an information processing method, including:

[0011] Acquire a time series of each first variable among a plurality of first variables of target multimedia data;

[0012] Based on each of the time series, determining a first association weight of each of the first variables with a second variable of the target multimedia data;

[0013] Based on each of the first association weights and the preset association weight information corresponding to the target multimedia data, a target association weight between each of the first variables and the second variable is obtained; wherein the preset association weight information corresponding to the target multimedia data includes a second association weight between each of the first variables and the second variable, and the second association weight is a degree of association between the first variable of each multimedia data in a plurality of multimedia data and the second variable of the multimedia data.

[0014] In a third aspect, an embodiment of the present invention provides an information processing method, including:

[0015] Obtaining a time series of each user behavior variable among multiple user behavior variables of a target object;

[0016] Based on each time series, determining a first association weight between each user behavior variable and a transaction variable of a target object;

[0017] Based on each first association weight and the preset association weight information corresponding to the target object, a target association weight between each user behavior variable and the transaction variable is obtained; wherein the preset association weight information corresponding to the target object includes a second association weight between each user behavior variable and the transaction variable, and the second association weight is a degree of association between the user behavior variable of each object in the multiple objects and the transaction variable of the object;

[0018] Determine the prompt information associated with the target object based on the target association weight of each user behavior variable and the transaction variable;

[0019] Displays a prompt message.

[0020] In a fourth aspect, an embodiment of the present invention provides an information processing device, including:

[0021] A time series acquisition module, used to acquire the time series of each first variable among a plurality of first variables of a target object;

[0022] A first determination module, for determining, based on each time series, a first association weight of each first variable with a second variable of the target object;

[0023] A second determination module, configured to obtain a target association weight between each first variable and a second variable based on each first association weight and preset association weight information corresponding to the target object;

[0024] The preset association weight information corresponding to the target object includes a second association weight between each first variable and a second variable, and the second association weight is a degree of association between the first variable of each object in the plurality of objects and the second variable of the object.

[0025] In a fifth aspect, an embodiment of the present invention provides an information processing device, including:

[0026] A time series acquisition module, used to acquire a time series of each first variable among a plurality of first variables of target multimedia data;

[0027] A first determination module, configured to determine, based on each time series, a first association weight of each first variable with a second variable of the target multimedia data;

[0028] A second determination module, configured to obtain a target association weight between each first variable and a second variable based on each first association weight and preset association weight information corresponding to the target multimedia data;

[0029] The preset association weight information corresponding to the target multimedia data includes a second association weight between each first variable and a second variable, and the second association weight is a degree of association between the first variable of each multimedia data in the plurality of multimedia data and the second variable of the multimedia data.

[0030] In a sixth aspect, an embodiment of the present invention provides an information processing device, including:

[0031] A time series acquisition module, used to acquire the time series of each user behavior variable among multiple user behavior variables of the target object;

[0032] A first determination module, used to determine a first association weight between each user behavior variable and a transaction variable of a target object based on each time series;

[0033] A second determination module is used to obtain a target association weight between each user behavior variable and a transaction variable based on each first association weight and preset association weight information corresponding to the target object; wherein the preset association weight information corresponding to the target object includes a second association weight between each user behavior variable and the transaction variable, and the second association weight is a degree of association between the user behavior variable of each object in the plurality of objects and the transaction variable of the object;

[0034] A prompt information determination module, used to determine prompt information associated with a target object based on a target association weight of each user behavior variable and a transaction variable;

[0035] The display module is used to display prompt information.

[0036] In a seventh aspect, an embodiment of the present invention provides a computing device, comprising: a processor and a memory storing computer program instructions;

[0037] When the processor executes the computer program instructions, the information processing method provided by the first aspect, the second aspect or the third aspect is implemented.

[0038] In an eighth aspect, an embodiment of the present invention provides a computer storage medium having computer program instructions stored thereon, and when the computer program instructions are executed by a processor, the information processing method provided in the first aspect, the second aspect or the third aspect described above is implemented.

[0039] In an embodiment of the present invention, the first association weight of each first variable and the second variable of the target object is determined by using the time series of each first variable of the target object, that is, by using the personalized data of the target object, that is, the personalized causal relationship weight of each first variable and the second variable is determined. For the second association weight of the first variable and the second variable, it refers to the degree of association between the first variable and the second variable shared by the target object and multiple objects, that is, the common causal relationship weight of the first variable and the second variable. By using the personalized causal relationship weight (first association weight) of each first variable and the second variable and the common causal relationship weight (second association weight) of each first variable and the second variable, the target association weight of each first variable and the second variable of the target object is determined, that is, with the help of the causal relationship shared by other objects related to the target object, the association weight between the variables of the target object with less sample data is accurately determined. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solution of the embodiment of the present invention, the following briefly introduces the drawings required for use in the embodiment of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1 A schematic diagram of a flow chart of an information processing method provided by an embodiment of the present invention;

[0042] Figure 2 A schematic diagram of the structure of a weight determination model provided in an embodiment of the present invention;

[0043] Figure 3 A schematic diagram of the structure of a prediction module of a weight determination model provided in an embodiment of the present invention;

[0044] Figure 4 A schematic diagram of a system architecture provided by an embodiment of the present invention;

[0045] Figure 5 A flowchart of an information processing method provided by another embodiment of the present invention;

[0046] Figure 6 A flowchart of an information processing method provided by yet another embodiment of the present invention;

[0047] Figure 7 is a structural block diagram of an information processing device according to an embodiment of the present invention;

[0048] Figure 8 is a structural block diagram of an information processing device according to another embodiment of the present invention;

[0049] Fig. 9 A structural block diagram of an information processing device according to another embodiment of the present invention;

[0050] Fig.10 A schematic diagram of the structure of a computing device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The features and exemplary embodiments of various aspects of the present invention will be described in detail below. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings and Examples. It should be understood that the specific embodiments described herein are only configured to explain the present invention and are not configured to limit the present invention. For those skilled in the art, the present invention can be implemented without the need for some of these specific details. The following description of the embodiments is only to provide a better understanding of the present invention by illustrating examples of the present invention.

[0052] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0053] In the era of big data, a large amount of data can be obtained through various data collection channels. By analyzing and mining these data, a lot of useful information can be obtained. For example, in the field of e-commerce, based on the data of multiple variables of a product, the causality between these variables can be analyzed, which facilitates the promotion and management of the product.

[0054] Causality can refer to the relationship between multiple variables (cause) and a target variable (effect). The degree of association between each variable and the target variable may be different. In the field of causal relationship mining, Granger causality analysis is a commonly used method. Granger causality analysis is based on the linear regression model and uses the time series of variables to analyze the causality between variables.

[0055] In practical applications, for different objects, the multivariate sequence of each object is different. Multivariate sequences from different objects can be called heterogeneous multivariate time series. The Granger causality analysis method can only establish regression models for different objects separately. When the sample data of some objects is small, it is impossible to accurately find the degree of association between multiple variables of the object, that is, it is impossible to accurately determine the causal relationship between the variables.

[0056] Based on this, the embodiments of the present invention provide an information processing method, apparatus, computing device and medium, which can accurately determine the degree of association between variables, that is, the causal relationship between variables.

[0057] Figure 1 FIG. 1 is a flow chart showing an information processing method provided by an embodiment of the present invention. Figure 1 As shown, the information processing method includes the following steps:

[0058] S110, obtaining a time series of each first variable among a plurality of first variables of a target object.

[0059] S120: Determine, based on each time series, a first association weight between each first variable and a second variable of the target object.

[0060] S130, based on each first association weight and preset association weight information corresponding to the target object, obtain a target association weight between each first variable and the second variable.

[0061] Among them, the preset association weight information corresponding to the target object includes the second association weight of each first variable and the second variable. For each first variable, the second association weight of the first variable and the second variable is: the degree of association between the first variable of each object in the multiple objects and the second variable of the object, and the degree of association between the first variable of each object and the second variable of the object is the same. In other words, the second association weight of the first variable and the second variable refers to the degree of association between the first variable and the second variable shared by the target object and the above-mentioned multiple objects, that is, the common causal relationship weight of the first variable and the second variable.

[0062] In an embodiment of the present invention, the first association weight of each first variable and the second variable is determined by utilizing the time series of each first variable of the target object, i.e. utilizing the personalized data of the target object, i.e. determining the personalized causal relationship weight of each first variable and the second variable.

[0063] By utilizing the personalized causal relationship weight (first association weight) of each first variable and the second variable and the common causal relationship weight (second association weight) of each first variable and the second variable, the target association weight of each first variable and the second variable of the target object is determined. That is, with the help of the causal relationship shared by other objects related to the target object, the association weights between the variables of the target object with less sample data can be accurately determined.

[0064] In some embodiments of the present invention, in S110, the time series of the first variable of the target object is a sequence consisting of values ​​of the first variable at multiple preset time points within a historical preset time period.

[0065] As an example, the number of first variables x of the target object is S, where S is a positive integer. k , the corresponding time series is a sequence of values ​​of preset T historical moments, that is, Among them, k = 1, 2, ..., S.

[0066] As an example, the target object may be a selected target product, and the two first variables of the target product may be click-through rate and pageviews. The time series corresponding to the click-through rate of the target product may be the value of the click-through rate at midnight every day of the previous month. The time series corresponding to the pageviews of the target product may be the value of the pageviews at midnight every day of the previous month.

[0067] In some embodiments of the present invention, in S120, the first variable of the target object is the cause, and the second variable of the target object is the effect. For each first variable, the first association weight between the first variable and the second variable is the causal relationship weight between the two variables, that is, the influence degree of the first variable on the second variable.

[0068] As an example, the second variable of the target product is the sales volume of the target product. That is, the click rate of the target product and the page views of the target product will both affect the sales volume of the target product, but the degree of influence may be different.

[0069] In some embodiments of the present invention, S120 includes S121 to S124. S121. For each time series, encode the time series using an encoder to obtain a first vector representation corresponding to the time series. S122. Input the second vector representation into the first neural network corresponding thereto to obtain a first prediction value of the second variable at a target time point, and the second vector representation includes each first vector representation. S123. For each target first vector representation, concatenate each first vector representation except the target first vector representation, and input the concatenated vector representation into the second neural network corresponding to the target first vector representation to obtain a second prediction value of the second variable at the target time point. S124. Based on the first prediction value and each second prediction value, the third vector representation includes a first association weight between each first variable and the second variable.

[0070] The specific implementation method of each step in S121 to S124 is introduced below.

[0071] In some embodiments of the present invention, in S121, each time series of the first variable of the target object has its corresponding encoder. In some embodiments, the encoder can be a long short-term memory network (Long Short-Term Memory, LSTM), a gated recurrent unit (Gated Recurrent Unit, GRU), a temporal convolutional network (Temporalconvolutional network, TCN) or a convolutional neural network (Convolutional Neural Network, CNN). In other words, the encoder is a network that can vectorize the time series.

[0072] In some embodiments, the first variable x of the target object is k The corresponding time series After inputting the corresponding encoder, the first vector representation corresponding to the time series can be obtained.

[0073] In S122, each first vector is first represented Splice and get the second vector representation h all That is, the second vector represents h allThe first vector representation corresponding to the time series of each first variable of the target object is included. Then, the second vector representation h all Input the first neural network corresponding to it to obtain the first predicted value of the second variable of the target object at the target time point It should be noted that the target time point is a future time point, that is, a time that has not yet occurred. There is no limitation on the vector concatenation method.

[0074] In S123, the target first vector representation is any first vector representation, wherein each first vector representation is a target first vector representation.

[0075] As an example, if the target first vector is represented as Except Each first vector other than Then, Splice to get the spliced ​​vector representation h all\1 . The vector representation h all\1 Input the second neural network corresponding to the target first vector representation to obtain the second predicted value of the second variable at the target time point The second predicted value is correspond.

[0076] Similarly, if the target first vector is represented as Then remove Each of the first vector representations other than In the corresponding neural network, the second predicted value of the second variable at the target time point is obtained The second predicted value is correspond.

[0077] Since each first vector representation is taken as the target first vector representation, a second prediction value corresponding to each first vector representation will be obtained, that is, S second prediction values ​​will be obtained.

[0078] In some embodiments of the present invention, S124 includes steps A to D. Step A, obtaining a first error value based on the first prediction value and the label value of the second variable. Step B, for each second prediction value, obtaining a second error value based on the second prediction value and the label value. Step C, for each second error value, subtracting the first error value from the second error value to obtain a target error value. Step D, obtaining a first association weight of each first variable and the second variable based on each target error value.

[0079] In step A, the first predicted value The mean square error value of the label value of the second variable is used as the first error value ε allSimilarly, in step B, for the second predicted value The mean square error between the second predicted value and the label value of the second variable is taken as the second error value ε all\k .

[0080] In step D, each target error value is first mapped using an activation function to obtain a mapping result corresponding to each target error value.

[0081] As an example, the activation function is a Rectified Linear Unit (Relu) activation function. The Relu activation function is a nonlinear activation function that converts all negative values ​​to 0, while leaving positive values ​​unchanged. That is, after each target error value is mapped using the Relu activation function, the mapping result corresponding to the target error value less than 0 is 0, and the mapping result corresponding to the target error value greater than or equal to 0 is itself.

[0082] Next, each mapping result is normalized to obtain the first association weight of each first variable and second variable, which is q 1 ,q 2 ,……,q S Among them, the third vector representation q is an S-dimensional vector, that is, (q 1 ,q 2 ,……,q S ).

[0083] That is to say, after normalizing the S target error values, S normalized values ​​are obtained, i.e., S first association weights. It should be noted that the normalized result of the target error value corresponding to the target first vector representation is the first association weight between the first variable and the second variable corresponding to the vector representation.

[0084] In the case of time series, the Granger causality between two variables X and Y is defined as: if the prediction effect of variable Y is better than the prediction effect of Y based on the past information of Y alone, that is, variable X helps to explain the future changes of variable Y, then variable X is considered to be the Granger cause of variable Y.

[0085] In S120, by using the Granger causality analysis method, a third vector representation can be obtained, and the third vector representation is the Granger causal attention vector.

[0086] In S130 , for each first variable, a first association weight between the first variable and the second variable and a second association weight between the first variable and the second variable are weighted summed to obtain a target association weight between the first variable and the second variable.

[0087] That is, for each first variable x of the target object k , the personalized causal relationship weight (first association weight q k ) and the commonality causal relationship weight (second association weight r) between the first variable and the second variable k ) is weighted and summed to obtain the target association weight a of the first variable and the second variable. k .

[0088] As an example, r k ×(1-α)+q k ×α=a k , α is greater than 0 and less than 1.

[0089] That is to say, the Granger causal attention vector (i.e., the third vector representation) and the vector corresponding to the preset association weight information corresponding to the target object are weighted and mixed through the attention mechanism to obtain the attention factor corresponding to each first variable of the target object, i.e., the target association weight of each first variable with the second variable. By using the attention mechanism, it is ensured that both the individualized causal relationship of the individual and the common causal relationship can be captured.

[0090] In some embodiments of the present invention, before S130, the information processing method provided by the embodiments of the present invention also includes: respectively calculating the similarity between the third vector representation and each preset association weight information in multiple preset association weight information; and taking the preset association weight information with the greatest similarity to the third vector representation as the preset association weight information corresponding to the target object.

[0091] It should be noted that each preset association weight information is common causal weight information. For each preset association weight information, the preset association weight information includes the preset association weight of each first variable and the second variable. In some embodiments, the preset association weight information p j is an S-dimensional vector, which includes the preset association weights of each first variable and the second variable in the S first variables. That is, j is an integer greater than or equal to 1 and less than or equal to n, n is a positive integer, and n is the total number of preset association weight information.

[0092] For a certain category of objects, the degree of association between each first variable and the second variable has commonality. That is, for each first variable, the degree of association between the first variable and the second variable of each object in a certain category of objects is the same, and the association weight of each first variable and the second variable of the class of objects can be used as a preset association weight information.

[0093] That is, objects with the same association weights of the first variable and the second variable can be classified as objects of one category. For objects of different categories, the association degree between the first variable and the second variable of each object in different categories may be different, so multiple preset association weight information can be obtained.

[0094] Therefore, the preset association weight information with the greatest similarity to the third vector representation may be used as the preset association weight information corresponding to the target object, that is, the weight information between the variables shared by the target object and other objects.

[0095] It should be noted that, in some examples, the similarity measurement method between the third vector representation and the preset association weight information includes methods such as Euclidean distance or cosine similarity. After the similarity parameter between the third vector representation and each preset association weight information is calculated using the similarity measurement method, the similarity parameter is processed using the Gumbel Softmax function to obtain the similarity between the third vector representation and each preset association weight information.

[0096] In some embodiments of the present invention, the target association weight of each first variable and second variable of the target object can be used to accurately predict the value of the second variable at the target time point. As an embodiment, after S110, the information processing method provided by the embodiment of the present invention also includes S115 and S116. S115, for each of the first vector representations, the first vector representation is input into the third neural network corresponding thereto to obtain the third predicted value of the second variable at the target time point. S116, based on each third predicted value and each target association weight, the target predicted value of the second variable at the target time point is obtained.

[0097] In some embodiments of the present invention, in S121, the time series is encoded using an encoder to obtain a first vector representation corresponding to the time series. For each first vector representation, the first vector representation is input into a third neural network corresponding thereto to obtain a third prediction value.

[0098] That is to say, each first vector representation has its corresponding third neural network, so S third prediction values ​​can be obtained, that is, each first vector representation has its corresponding third prediction value. That is, using the first vector representation corresponding to the time series of each first variable and the third neural network corresponding to the vector representation, the third prediction value of the second variable corresponding to each first variable at the target time point is predicted respectively. It should be noted that the first neural network, the second neural network and the third neural network can all be multi-layer perceptrons.

[0099] In S116, the third predicted value corresponding to each first variable is The target association weights of the first variable and the second variable are weighted summed to obtain the target predicted value y of the second variable at the target time point T+1 .

[0100] In the embodiment of the present invention, the contribution of each first variable of the target object to the prediction of the second variable is reflected by using the attention mechanism, thereby reflecting the target association weight of each first variable and the second variable. Since the association weight of each first variable and the second variable of the target object is accurately determined by means of the causal relationship shared by other objects related to the target object, the accuracy of the prediction of the target predicted value of the second variable of the target object at the target time point is also improved.

[0101] The information processing method provided by the embodiment of the present invention can be implemented by using a pre-trained weight determination model. Figure 2 The structural diagram of the weight determination model is shown in FIG. Figure 2 As shown, the weight determination model includes a first association weight determination module and a target association weight determination module.

[0102] First, the acquired time series of each first variable of the target object is input into a first association weight determination module to obtain a first association weight of each first variable and second variable of the target object.

[0103] See also Figure 2 , the first variable x 1 The time series is The first variable x 1 The association weight with the second variable is q 1 The first variable x 2 The time series is The first variable x 2 The association weight with the second variable is q 2 . Similarly, the first variable x S The time series is The first variable x S The association weight with the second variable is q S That is, the first variable x k The time series is The first variable x k The association weight with the second variable is q k .

[0104] Then, the first association weight of each first variable and second variable of the target object and the second association weight of each first variable and second variable are input into the target association weight determination module to obtain the target association weight of each first variable and second variable.

[0105] Continue to see Figure 2, for the first variable x k , the first association weight q between the first variable and the second variable k and the second association weight r of the first variable and the second variable k By performing weighted summation, the target association weight a of the first variable and the second variable can be obtained. k .like Figure 2 As shown, Figure 2 The circle with a plus sign in the center represents the addition operation, i.e. r k ×(1-α)+q k ×α=a k , α is greater than 0 and less than 1.

[0106] Continue to see Figure 2 The first association weight determination module includes a plurality of encoders, the number of which is the same as the number of the first variables of the target object. Each time series of the first variable of the target object has its corresponding encoder.

[0107] See also Figure 2 , the first variable x of the target object k Time Series Input the corresponding encoder to get the first vector representation of the time series. For details, please refer to the description of S121 above, which will not be repeated here. The network architecture of the encoder corresponding to each time series is basically the same, but the parameters of each encoder are different.

[0108] Continue to see Figure 2 , when the first association weight determination module determines the first vector representation corresponding to the time series of each first variable of the target object Afterwards, each first vector is represented Splice and get the second vector representation h all Then the second vector represents h all Input the first neural network corresponding to it to obtain the first predicted value of the second variable of the target object at the target time point For details, please refer to the description of S122, which will not be repeated here.

[0109] Then, for each first vector, it is represented as Then remove Each of the first vector representations other than In the corresponding neural network, the second predicted value of the second variable at the target time point is obtained

[0110] Then, the first association weight determination module converts the first prediction value The mean square error value of the label value of the second variable is used as the first error value ε all For the second predicted value The mean square error between the second predicted value and the label value of the second variable is taken as the second error value ε all\k , thereby obtaining the second error value corresponding to each first variable.

[0111] Figure 2 The circle with a minus sign in the center of represents a subtraction operation. For each second error value, the first error value is subtracted from the second error value to obtain a target error value. Figure 2 The circle with a wavy mark in the center represents the Relu activation function. Referring to the description of step D above, after each target error value is mapped using the activation function, each mapping result is normalized to obtain the first association weight of each first variable and second variable, which is q 1 ,q 2 ,……,q S .

[0112] It should be noted that the pre-trained weight determination model also includes a preset associated weight information determination module. Figure 2 , p1, p2, ..., p n is n preset association weight information. Among them, the preset association weight information p j is an S-dimensional vector, which includes the preset association weights of each first variable and the second variable in the S first variables.

[0113] Then, the preset association weight information determination module measures the similarity between the third vector representation q and each preset association weight information, and then uses the Gumbel Softmax function to process the similarity parameters to obtain the similarity between the third vector representation and each preset association weight information. Finally, the preset association weight information with the highest similarity to the third vector representation q is used as the preset association weight information corresponding to the target object. In other words, the preset association weight information corresponding to the target object includes the preset association weights of each first variable and the second variable, then r k is the first variable x in the preset association weight information k The preset association weight with the second variable. For details, please refer to the above description, which will not be repeated here.

[0114] See also Figure 3 The weight determination model provided in the embodiment of the present invention also includes a prediction module. The prediction module uses Figure 2 The third neural network corresponding to each first vector representation in predicts the first vector representation to obtain the third predicted value of the second variable corresponding to each first variable at the target time point. kThe corresponding third predicted value of the second variable at the target time point is

[0115] See also Figure 3 , and the circle with dots represents the multiplication operation. The prediction module is used to convert Associated weight a with the corresponding target k Then, the prediction module adds the products corresponding to each third prediction value to obtain the target prediction value of the second variable of the target object at the target time point. For details, please refer to the description of S116, which will not be repeated here.

[0116] The parameters of the encoder in the weight determination model provided by the embodiment of the present invention, multiple preset association weight information, etc. can be shared among multiple individuals, and inductive learning of causal relationships can be achieved. The embodiment of the present invention transforms the causal relationship discovery problem into a time series prediction problem based on the attention mechanism, and by comprehensively considering Granger causal attribution and learning common association weight information, it can capture the individual's personalized causal structure and the common causal structure between individuals.

[0117] The following describes a training method for a weight determination model. First, a training sample set is obtained. The sample set includes multiple batches of sample data. Each batch of sample data includes a time series of each first variable of each sample object in multiple sample objects.

[0118] For each sample object in a batch of sample data, the time series of each first variable of the sample object is input into the weight determination model to be trained to obtain the target prediction value of the second variable of the sample object at the target time point, as well as the first error value corresponding to the sample object and the second error value corresponding to each first variable of the sample object.

[0119] Then, the loss function is used to adjust the weights to be trained to determine each parameter in the model. Among them, the loss function L = L pred +λ1L aux +λ2L div , through the back propagation algorithm, the loss function L is minimized to adjust the parameters of the model determined by the weights to be trained. Among them, λ1 and λ2 are preset coefficients.

[0120] in:

[0121]

[0122]

[0123]

[0124] Among them, y i,T+1 is the target predicted value of the second variable of sample object i at the target time point. is the label value of the second variable of sample object i. is the first error value corresponding to sample object i, is the second error value corresponding to the S first variables of sample object i. N is the total number of sample objects in a batch. j Determine the jth preset association weight information in the model for the weight to be trained. γ is the preset similarity threshold. max() represents the sum of γ and The maximum value in .

[0125] After adjusting the parameters of the weight determination model to be trained using the loss function, the adjusted weight determination model to be trained can be obtained. Then, it is determined whether the preset training stop condition is met. If it is met, the trained weight determination model is obtained. If it is not met, a new batch of sample data is obtained, and the adjusted weight determination model to be trained is continued to be trained until a weight determination model that meets the preset training stop condition is obtained.

[0126] In one embodiment of the present invention, the preset training condition includes the number of iterations reaching a preset threshold.

[0127] In another embodiment of the present invention, the preset training condition includes that the difference between the loss function of the weight determination model to be trained and the loss function determined in the last iteration process does not exceed the preset difference threshold. That is to say, if the loss function of the weight determination model to be trained is close to the loss function calculated in the last iteration process, it can be considered that the performance of the trained weight determination model has reached the best.

[0128] It should be noted that the weight determination model to be trained with the first batch of sample data input is the initial weight determination model. The initial value of the initial weight determination model parameter can be a random parameter or a parameter value set by the user based on experience. In the subsequent iteration process, the weight determination model to be trained is the weight determination model obtained after the previous iteration process ends.

[0129] It is worth mentioning that the n preset association weight information is obtained based on the initial weight determination model for the third vector representation corresponding to each sample object in the first batch of sample data. By clustering the third vector representation corresponding to each sample object in the first batch of sample data using a clustering algorithm, the initial values ​​of the n preset association weight information can be obtained.

[0130] The training method of the weight determination model provided by the embodiment of the present invention combines the Granger causal analysis method and the attention mechanism, and weightedly mixes the third vector representation of the sample object and the preset association weight information corresponding to the sample object to obtain the Granger causal attention vector, that is, a vector composed of the target association weights of each first variable and second variable of the sample object. Through the attention mechanism, the target association weight of each first variable and second variable not only reflects the individual personalized causal relationship, but also captures the common causal relationship, so it can improve the accuracy of determining the association weights (i.e., causal weights) between variables, and can also achieve the accuracy of predicting the predicted value of the target variable at the target time point.

[0131] Figure 4 A schematic diagram of a scenario showing an information processing method provided by an embodiment of the present invention.

[0132] like Figure 4 As shown, the target object is an image of a shoe. For example, the second variable is the sales volume of the shoe in the image. As an example, Figure 4 The three first variables corresponding to the image of the shoes are the number of clicks, forwardings and views of the image.

[0133] The client 410 can obtain the time series of each first variable corresponding to the image by collecting the operation data of different users on the image. For example, the time series corresponding to each first variable is a sequence of the values ​​of the variable every day in the last month arranged in chronological order. For example, the time series corresponding to the number of clicks on the image of the shoe is The time series corresponding to the forwarding volume of the shoe image is The time series corresponding to the number of views of the shoe image is

[0134] like Figure 4 If a merchant of shoes in the image wants to count the associated weights of the number of clicks, forwardings, and views corresponding to the image of the shoes and the sales volume of the shoes, the merchant can send the time series corresponding to each first variable of the image to the server 420 through the client 410. That is, the client 410 sends the time series of the number of clicks, the time series of forwardings, and the time series of views of the image of the shoes to the server 420.

[0135] The server 420 receives the time series corresponding to each first variable of the shoe image sent by the client 410. Then, based on the obtained time series of each first variable, the server 420 determines the first association weight of each first variable with the sales volume of the shoe. As an example, the first association weights of the click volume, forwarding volume, and pageview volume of the shoe image with the sales volume of the shoe are respectively: q 1 ,q 2and q 3 .

[0136] Then, the server 420 obtains the target association weight of each first variable and the second variable based on the first association weight of each first variable and the second variable and the preset association weight information corresponding to the image of the shoe.

[0137] Among them, the preset association weight information corresponding to the image of the shoe includes the second association weight of each first variable and the second variable. For each first variable, the second association weight of the first variable and the second variable is: the degree of association between the first variable of each multimedia data in the multiple multimedia data and the second variable of the multimedia data, and the degree of association between the first variable of each multimedia data and the second variable of the object is the same. In other words, the second association weight for the first variable and the second variable refers to the degree of association between the image of the shoe and the first variable and the second variable shared by the above-mentioned multiple multimedia data, that is, the common causal relationship weight of the first variable and the second variable. In other words, for the above-mentioned multiple multimedia data, these data and Figure 4 The images of shoes in have commonalities in the degree of association between each of their first variables and the second variable.

[0138] In one embodiment, assuming that the second association weights of the number of clicks, forwardings, and views of the shoe image and the sales volume of the shoe are: 1 、r 2 and r 3 For each first variable, the first association weight and the second association weight of the first variable and the second variable are weighted and summed to obtain the target association weight of the first variable and the second variable.

[0139] As a specific example, the target association weights of the number of clicks, forwardings, and views of a shoe image with the sales volume of the shoe are: a 1 、a 2 and a 3 Among them, r k ×(1-α)+q k ×α=a k .

[0140] The information processing method provided by the embodiment of the present invention can be applied to the scenario of determining the degree of association between different variables of multimedia data, which is described in detail below. Figure 5As shown, the information processing method applied to multimedia data provided by the embodiment of the present invention includes S510 to S530. S510, obtaining the time series of each first variable among multiple first variables of the target multimedia data. S520, based on each time series, determining the first association weight of each first variable with the second variable of the target multimedia data. S530, based on each first association weight and the preset association weight information corresponding to the target multimedia data, obtaining the target association weight of each first variable with the second variable.

[0141] Among them, the preset association weight information corresponding to the target multimedia data includes the second association weight of each first variable and the second variable. For each first variable, the second association weight of the first variable and the second variable is: the degree of association between the first variable of each multimedia data in the multiple multimedia data and the second variable of the multimedia data, and the degree of association between the first variable of each multimedia data and the second variable of the object is the same. In other words, the second association weight of the first variable and the second variable refers to the degree of association between the first variable and the second variable shared by the target multimedia data and the above-mentioned multiple multimedia data, that is, the common causal relationship weight of the first variable and the second variable.

[0142] As an example, the target multimedia data may be an image or a video. For example, the target multimedia data may be an image corresponding to a certain product on an e-commerce platform, or a video corresponding to a certain product. The second variable may be the sales volume of the product corresponding to the target multimedia data. The first variable may be a parameter such as the number of clicks, views, or forwardings of the target multimedia data.

[0143] In other embodiments, the target multimedia data may be a static image or a dynamic image.

[0144] In an embodiment of the present invention, the first association weight of each first variable and the second variable is determined by utilizing the time series of each first variable of the target multimedia data, i.e. utilizing the personalized data of the target multimedia data, i.e. determining the personalized causal relationship weight of each first variable and the second variable.

[0145] By utilizing the personalized causal relationship weight (first association weight) of each first variable and the second variable and the common causal relationship weight (second association weight) of each first variable and the second variable, the target association weight of each first variable and the second variable of the target multimedia data is determined. That is, with the help of the causal relationship shared by other multimedia data related to the target multimedia data, the association weights between the variables of the target multimedia data with less sample data can be accurately determined.

[0146] In some embodiments of the present invention, the specific implementation method of S510 is similar to the specific implementation method of S110, the specific implementation method of S520 is similar to the specific implementation method of S120, and the specific implementation method of S530 is similar to the specific implementation method of S130, which will not be repeated here.

[0147] In some embodiments, in S520, first, for each time series of the target multimedia data, the time series is encoded using an encoder to obtain a first vector representation corresponding to the time series. Next, the second vector representation is input into the first neural network corresponding thereto to obtain a first prediction value of the second variable of the target multimedia data at the target time point, and the second vector representation includes each first vector representation. Next, for each target first vector representation, each other first vector representation except the target first vector representation is spliced, and the spliced ​​vector representation is input into the second neural network corresponding to the target first vector representation to obtain a second prediction value of the second variable of the target multimedia data at the target time point. Finally, based on the first prediction value and each second prediction value, a third vector representation is obtained, and the third vector representation includes a first association weight between each first variable and the second variable of the target multimedia data.

[0148] In an embodiment of the present invention, by using the Granger causality analysis method to obtain the first association weight of each first variable and second variable of the target multimedia data, the association relationship between the first variable and the second variable can be obtained more accurately.

[0149] In some embodiments of the present invention, after S510, the information processing method further includes S515 and S516. S515, for each first vector representation, input the first vector representation into the third neural network corresponding thereto, and obtain a third predicted value of the second variable at the target time point. S516, according to each third predicted value and each target association weight, obtain a target predicted value of the second variable at the target time point.

[0150] In the embodiment of the present invention, the specific implementation of S115 is similar to the specific implementation of S515, and the specific implementation of S516 is similar to the specific implementation of S116, which will not be repeated here.

[0151] As an example, see Figure 4 ,like Figure 4The merchant of shoes wants to predict the sales volume of the shoes at time T1. Then, the encoder can be used to encode the time series of clicks, forwardings and views of the shoe images respectively, and obtain the first vector representation corresponding to each time series. Then, the first vector representation corresponding to each time series is input into the corresponding third neural network to obtain the three third prediction values ​​of the shoe at time T1. Then, the three third prediction values ​​are weighted and summed with the target association weight corresponding to each prediction value, and the target prediction value y of the shoe at time T1 can be obtained. T1 .

[0152] In an embodiment of the present invention, the contribution of each first variable of the target multimedia data to the prediction of the second variable is reflected by using the attention mechanism, thereby reflecting the target association weight of each first variable and the second variable. Since the association weight of each first variable and the second variable of the target multimedia data is accurately determined by means of the causal relationship shared by other multimedia data related to the target multimedia data, the accuracy of the prediction of the target predicted value of the second variable of the target multimedia data at the target time point is also improved.

[0153] It should be noted that the target object can be not only a commodity, a multimedia data, but also a merchant and other objects with different variables.

[0154] Figure 6 FIG. 1 is a flow chart showing an information processing method provided by an embodiment of the present invention. Figure 6 As shown, the information processing method provided by the embodiment of the present invention includes S610 to S650.

[0155] S610: Obtain a time series of each user behavior variable among a plurality of user behavior variables of a target object.

[0156] S620: Determine a first association weight between each user behavior variable and a transaction variable of the target object based on each time series.

[0157] S630, based on each first association weight and the preset association weight information corresponding to the target object, obtain a target association weight between each user behavior variable and the transaction variable. The preset association weight information corresponding to the target object includes a second association weight between each user behavior variable and the transaction variable, and the second association weight is the degree of association between the user behavior variable of each object in the multiple objects and the transaction variable of the object.

[0158] S640: Determine prompt information associated with the target object based on the target association weight of each user behavior variable and the transaction variable.

[0159] S650: Display a prompt message.

[0160] In an embodiment of the present invention, the target association weight of each user behavior variable and transaction variable of the target object is determined by utilizing the personalized causal relationship weight (first association weight) of each user behavior variable and transaction variable and the common causal relationship weight (second association weight) of each user behavior variable and transaction variable. That is, with the help of the causal relationship shared by other objects related to the target object, the association weights between the variables of the target object with less sample data can be accurately determined, thereby generating more accurate prompt information so that users can make more accurate business plans for the target object.

[0161] In some embodiments of the present invention, the specific implementation method of S610 is similar to the specific implementation method of S110, the specific implementation method of S620 is similar to the specific implementation method of S120, and the specific implementation method of S530 is similar to the specific implementation method of S630, which will not be repeated here.

[0162] In S620, each time series can be analyzed by Granger causality analysis method to obtain the first correlation weight between each user behavior variable and the transaction variable of the target object. For the detailed description of S620, reference can be made to the relevant description of S120 above, which will not be repeated here.

[0163] In some embodiments, in S640 , the prompt information associated with the target object includes a target association weight of each user behavior variable and transaction variable of the target object.

[0164] In some embodiments of the present invention, the multiple user behavior variables include at least two of the following variables: click volume, forwarding volume, pageview volume, like volume, and comment volume.

[0165] In some embodiments of the present invention, the transaction variable is any one of the following variables: sales quantity, price, and sales amount.

[0166] That is to say, the prompt information may be the association between the user behavior data and the business data.

[0167] As an example, the target object is a product, and the prompt information includes the target association weight between the sales volume of the product and the number of clicks on the product by users, and the target association weight between the sales volume of the product and the number of forwardings of the product by users. By displaying this prompt information to the merchant, the merchant can compare the target association weights of the clicks and forwardings with the sales volume, respectively, so that the merchant can know which factor, clicks or forwardings, has a greater impact on the sales volume of the product, so that the merchant can pay more attention to the user behavior data of the product in a targeted manner to promote the sales volume of the product.

[0168] In other words, merchants can obtain the degree of influence of each user behavior variable on the transaction variable through the target association weight between the user behavior variable and the transaction variable, so that merchants can selectively use user behavior data to formulate relevant product promotion strategies and product sales strategies.

[0169] In some embodiments, the prompt information includes a target predicted value of the transaction variable of the target object at the target time point. Referring to the description of S115 and S116 above, the first vector representation corresponding to the time series of each user behavior variable is input into the third neural network corresponding to the vector representation, and the third predicted value of the transaction variable at the target time point can be obtained. Then, the third predicted value corresponding to each user behavior variable is weighted and summed with the target association weight of the user behavior variable and the transaction variable, so as to obtain the target predicted value of the transaction variable at the target time point.

[0170] In an embodiment of the present invention, by comprehensively considering the associated weight information of Granger causal attribution and learning commonality, the personalized causal structure of individuals and the common causal structure between individuals can be captured, so the prediction of transaction variables is more accurate.

[0171] In the embodiments of the present invention, by predicting the sales volume or sales amount of the target object at the target time point, the merchant can formulate a sales strategy for the target object in advance, which is convenient for the merchant to sell.

[0172] In some embodiments, by predicting the price value of the target object at a target time point, consumers can know the price change trend of the target object in advance, which is convenient for consumers to formulate purchase strategies.

[0173] In some embodiments, the user behavior variable may also be a user's experience variable for a product. For example, the experience variable may be a variable such as the number of times or duration of use of a function of a product by a user. Among them, the transaction variable may be purchase intention. By utilizing the correlation between the user experience variable and purchase intention, product designers can understand consumers' purchase intention for the product, so as to formulate product improvement strategies to better meet user needs.

[0174] For example, in the current customer-to-manufacturer (C2M) business model, factories produce personalized products based on consumer demand. Designers can obtain consumer product demand based on consumer experience data of the new products they design, so as to design products that better meet consumer needs.

[0175] In some embodiments of the present invention, Figure 6 The information processing method in the present invention can be applied to a user's terminal.

[0176] In some embodiments, in order to meet the personalized needs of users, if the prompt information includes the target association weights of each user behavior variable and the transaction variable, after S650, the information processing method provided by the embodiment of the present invention also includes: receiving a user's selection instruction for the prompt information, and determining a first target association weight; receiving a user's change instruction for the first target association weight; in response to the change instruction, changing the first target association weight to obtain a changed first target association weight; and displaying the changed first target association weight.

[0177] As an example, the selection instruction may be a selection operation such as a click operation or a slide operation by the user on the first target association weight to select the first target association weight that needs to be changed.

[0178] Wherein, the modification instruction includes the modified first target association weight, and in response to the modification instruction, the first target association weight is modified, and then the modified first target association weight is displayed to the user, so that the user can use the modified first target association weight for business forecasting or business planning.

[0179] As an example, if some special dates are encountered when obtaining the time series of user behavior variables, and the user believes that they may affect the target association weights between the user behavior variables and the transaction variables, the user can modify the target association weights to meet the user's personalized needs.

[0180] Figure 7 FIG. 2 shows a schematic diagram of the structure of an information processing device provided according to an embodiment of the present invention. Figure 7 As shown, the information processing device 700 includes:

[0181] The time series acquisition module 710 is used to acquire the time series of each first variable among multiple first variables of the target object.

[0182] The first determination module 720 is used to determine a first association weight of each first variable with a second variable of the target object based on each time series.

[0183] The second determination module 730 is used to obtain a target association weight between each first variable and the second variable based on each first association weight and preset association weight information corresponding to the target object.

[0184] The preset association weight information corresponding to the target object includes a second association weight between each first variable and a second variable, and the second association weight is a degree of association between the first variable of each object in the plurality of objects and the second variable of the object.

[0185] By utilizing the personalized causal relationship weight (first association weight) of each first variable and the second variable and the common causal relationship weight (second association weight) of each first variable and the second variable, the target association weight of each first variable and the second variable of the target object is determined. That is, with the help of the causal relationship shared by other objects related to the target object, the association weights between the variables of the target object with less sample data can be accurately determined.

[0186] In an embodiment of the present invention, the first determining module 720 includes:

[0187] The first encoding unit is used to encode each time series by using an encoder to obtain a first vector representation corresponding to the time series.

[0188] The first prediction unit is used to input the second vector representation into the first neural network corresponding thereto to obtain a first predicted value of the second variable at a target time point, the second vector representation including each first vector representation.

[0189] The second prediction unit is used to splice each first vector representation except the target first vector representation for each target first vector representation, and input the spliced ​​vector representation to the second neural network corresponding to the target first vector representation to obtain a second predicted value of the second variable at the target time point.

[0190] The first association weight determination unit is used to obtain a third vector representation based on the first prediction value and each second prediction value, where the third vector representation includes a first association weight between each first variable and the second variable.

[0191] In an embodiment of the present invention, the first association weight determination unit includes:

[0192] The first error value determining subunit is used to obtain a first error value based on the first prediction value and the label value of the second variable.

[0193] The second error value determination subunit is used to obtain a second error value for each second prediction value based on the second prediction value and the label value.

[0194] The target error value determination subunit is used to, for each second error value, subtract the first error value from the second error value to obtain a target error value.

[0195] The association weight determination subunit is used to obtain a first association weight between each first variable and the second variable based on each target error value.

[0196] In an embodiment of the present invention, the association weight determination subunit is used to:

[0197] Use the activation function to map each target error value to obtain the mapping result corresponding to each target error value;

[0198] Each mapping result is normalized to obtain a first association weight between each first variable and the second variable.

[0199] In the embodiment of the present invention, the first determining module 720 is used to:

[0200] For each first variable, a first association weight between the first variable and the second variable and a second association weight between the first variable and the second variable are weighted summed to obtain a target association weight between the first variable and the second variable.

[0201] In an embodiment of the present invention, the information processing device 700 further includes:

[0202] The similarity calculation module is used to respectively calculate the similarity between the third vector representation and each of the preset association weight information in the plurality of preset association weight information.

[0203] The association weight information determination module is used to use the preset association weight information with the greatest similarity to the third vector representation as the preset association weight information corresponding to the target object.

[0204] In an embodiment of the present invention, the information processing device 700 further includes a first prediction module, which is used to:

[0205] For each first vector representation, input the first vector representation into a third neural network corresponding thereto to obtain a third predicted value of the second variable at a target time point;

[0206] According to each third prediction value and each target association weight, a target prediction value of the second variable at the target time point is obtained.

[0207] In an embodiment of the present invention, the encoder is LSTM, GRU, TCN or CNN.

[0208] Other details of the information processing device 700 according to the embodiment of the present invention are combined with the above Figures 1 to 3 The information processing method according to the embodiment of the present invention described is similar and will not be described in detail here.

[0209] Figure 8 FIG. 2 shows a schematic diagram of the structure of an information processing device provided according to an embodiment of the present invention. Figure 8 As shown, the information processing device 800 includes:

[0210] The time series acquisition module 810 is used to acquire the time series of each first variable among multiple first variables of the target multimedia data.

[0211] The first determination module 820 is configured to determine, based on each time series, a first association weight between each first variable and a second variable of the target multimedia data.

[0212] The second determination module 830 is configured to obtain a target association weight between each first variable and a second variable based on each first association weight and preset association weight information corresponding to the target multimedia data.

[0213] The preset association weight information corresponding to the target multimedia data includes a second association weight between each first variable and a second variable, and the second association weight is a degree of association between the first variable of each multimedia data in the plurality of multimedia data and the second variable of the multimedia data.

[0214] In an embodiment of the present invention, the target association weight of each first variable and second variable of the target multimedia data is determined by utilizing the personalized causal relationship weight (first association weight) of each first variable and second variable and the common causal relationship weight (second association weight) of each first variable and second variable. That is, with the help of the causal relationship shared by other multimedia data related to the target multimedia data, the association weights between the variables of the target multimedia data with less sample data can be accurately determined.

[0215] Optionally, in some embodiments of the present invention, the first determining module 820 includes:

[0216] The second encoding unit is used to encode each time series by using an encoder to obtain a first vector representation corresponding to the time series.

[0217] The third prediction unit is used to input the second vector representation into the first neural network corresponding to it to obtain a first predicted value of the second variable at a target time point, and the second vector representation includes each first vector representation.

[0218] The fourth prediction unit is used to splice each first vector representation except the target first vector representation for each target first vector representation, and input the spliced ​​vector representation to the second neural network corresponding to the target first vector representation to obtain a second predicted value of the second variable at the target time point.

[0219] The second association weight determination unit is used to obtain a third vector representation based on the first prediction value and each second prediction value, where the third vector representation includes a first association weight between each first variable and the second variable.

[0220] Optionally, in some embodiments of the present invention, the information processing device 800 further includes a second prediction module, which is used to:

[0221] For each first vector representation, the first vector representation is input into a third neural network corresponding thereto to obtain a third predicted value of a second variable of the target multimedia data at a target time point.

[0222] According to each third prediction value and each target association weight, a target prediction value of the second variable at the target time point is obtained.

[0223] Other details of the information processing device 800 according to the embodiment of the present invention are combined with the above Figure 4 and Figure 5 The information processing method according to the embodiment of the present invention described is similar and will not be described in detail here.

[0224] Fig. 9 FIG. 9 is a schematic diagram showing the structure of an information processing device 900 provided according to an embodiment of the present invention. Fig. 9 As shown, the information processing device 900 includes:

[0225] The time series acquisition module 910 is used to acquire the time series of each user behavior variable among multiple user behavior variables of the target object.

[0226] The first determination module 920 is used to determine a first association weight between each user behavior variable and a transaction variable of a target object based on each time series.

[0227] The second determination module 930 is used to obtain a target association weight between each user behavior variable and a transaction variable based on each first association weight and preset association weight information corresponding to the target object. The preset association weight information corresponding to the target object includes a second association weight between each user behavior variable and a transaction variable, and the second association weight is a degree of association between the user behavior variable of each object in the plurality of objects and the transaction variable of the object.

[0228] The prompt information determination module 940 is used to determine the prompt information associated with the target object based on the target association weight of each user behavior variable and the transaction variable.

[0229] The display module 950 is used to display prompt information.

[0230] In an embodiment of the present invention, the target association weight of each user behavior variable and transaction variable of the target object is determined by utilizing the personalized causal relationship weight (first association weight) of each user behavior variable and transaction variable and the common causal relationship weight (second association weight) of each user behavior variable and transaction variable. That is, with the help of the causal relationship shared by other objects related to the target object, the association weights between the variables of the target object with less sample data can be accurately determined, thereby generating more accurate prompt information so that users can make more accurate business plans for the target object.

[0231] In some embodiments, the plurality of user behavior variables include at least two of the following variables: click volume, forwarding volume, pageview volume, like volume, and comment volume.

[0232] The transaction variable is any one of the following variables: sales quantity, price, and sales amount.

[0233] In some embodiments, the first determination module 920 is used to:

[0234] Each time series is analyzed by the Granger causality analysis method to obtain the first correlation weight between each user behavior variable and the transaction variable of the target object.

[0235] In some embodiments, the prompt information includes a target predicted value of a transaction variable of the target object at a target time point.

[0236] In some embodiments, the prompt information includes a target association weight of each user behavior variable and a transaction variable, and the information processing device 900 further includes:

[0237] The third determination module is used to receive a user's selection instruction for the prompt information and determine the first target association weight.

[0238] The change instruction receiving module is used to receive a user's change instruction on the first target association weight.

[0239] The changing module is used to change the first target association weight in response to the changing instruction to obtain the changed first target association weight.

[0240] The display module 950 is further used to display the modified first target association weight.

[0241] Other details of the information processing device 900 according to the embodiment of the present invention are combined with the above Figure 6 The information processing method according to the embodiment of the present invention described is similar and will not be described in detail here.

[0242] Combination Figure 1 and Fig. 9 The information processing method and apparatus described in the embodiments of the present invention may be implemented by a computing device. Fig.10 FIG. 1 is a schematic diagram showing a hardware structure 1000 of a computing device according to an embodiment of the invention.

[0243] like Fig.10As shown, the computing device 1000 includes an input device 1001, an input interface 1002, a central processing unit 1003, a memory 1004, an output interface 1005, and an output device 1006. The input interface 1002, the central processing unit 1003, the memory 1004, and the output interface 1005 are connected to each other via a bus 1010, and the input device 1001 and the output device 1006 are connected to the bus 1010 via the input interface 1002 and the output interface 1005, respectively, and are further connected to other components of the computing device 1000.

[0244] Specifically, the input device 1001 receives input information from the outside and transmits the input information to the central processing unit 1003 through the input interface 1002; the central processing unit 1003 processes the input information based on the computer executable instructions stored in the memory 1004 to generate output information, temporarily or permanently stores the output information in the memory 1004, and then transmits the output information to the output device 1006 through the output interface 1005; the output device 1006 outputs the output information to the outside of the computing device 1000 for user use.

[0245] That is to say, Fig.10 The computing device shown may also be implemented as including: a memory storing computer executable instructions; and a processor, which can implement the combination of Figures 1 to 9 Described information processing method and device.

[0246] An embodiment of the present invention further provides a computer storage medium, on which computer program instructions are stored; when the computer program instructions are executed by a processor, the information processing method provided by the embodiment of the present invention is implemented.

[0247] The functional blocks shown in the above block diagram can be implemented as hardware, software, firmware or a combination thereof. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments that are used to perform the required tasks. Programs or code segments can be stored in machine-readable media, or transmitted on a transmission medium or a communication link by a data signal carried in a carrier wave. "Machine-readable media" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, optical fiber media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0248] It should also be noted that the exemplary embodiments mentioned in the present invention describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above steps, that is, the steps can be performed in the order mentioned in the embodiments, or in a different order from the embodiments, or several steps can be performed simultaneously.

[0249] The above is only a specific implementation of the present invention. Those skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working processes of the systems, modules and units described above can refer to the corresponding processes in the aforementioned method embodiments, and will not be repeated here. It should be understood that the protection scope of the present invention is not limited to this. 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 covered within the protection scope of the present invention.

Claims

1. An information processing method, wherein: The method comprises: Acquire a time series of each first variable of a plurality of first variables of a target object, wherein the target object includes a target product, and the plurality of first variables include a click rate and a pageview volume; Based on each of the time series, determining a first association weight between each of the first variables and a second variable of the target object, where the second variable is the sales volume of the target commodity; For each of the first variables, a first association weight between the first variable and the second variable and a second association weight between the first variable and the second variable are weightedly summed to obtain a target association weight between the first variable and the second variable, wherein the first association weight is a personalized causal relationship weight between the first variable and the second variable of the target object, the second association weight is a degree of association between the first variable and the second variable shared by the target object and multiple objects, and the second association weight is a degree of association between the first variable of each object in the multiple objects and the second variable of the object, and the degree of association between the first variable of each object and the second variable of the object is the same.

2. The method according to claim 1, wherein: The determining, based on each of the time series, a first association weight of each of the first variables and the second variable of the target object comprises: For each of the time series, encoding the time series using an encoder to obtain a first vector representation corresponding to the time series; Inputting a second vector representation into a first neural network corresponding thereto to obtain a first predicted value of the second variable at a target time point, the second vector representation including each of the first vector representations; For each target first vector representation, concatenate each other first vector representation except the target first vector representation, and input the concatenated vector representation into a second neural network corresponding to the target first vector representation to obtain a second predicted value of the second variable at the target time point; Based on the first predicted value and each of the second predicted values, a third vector representation is obtained, wherein the third vector representation includes a first association weight of each of the first variables and the second variable.

3. The method according to claim 2, wherein: The obtaining a third vector representation based on the first predicted value and each of the second predicted values ​​includes: Obtaining a first error value based on the first prediction value and the label value of the second variable; For each of the second prediction values, obtaining a second error value based on the second prediction value and the label value; For each of the second error values, subtract the first error value from the second error value to obtain a target error value; Based on each of the target error values, a first association weight between each of the first variables and the second variable is obtained.

4. The method according to claim 3, wherein: The obtaining, based on each of the target error values, a first association weight between each of the first variables and the second variable comprises: Mapping each of the target error values ​​using an activation function to obtain a mapping result corresponding to each of the target error values; Each of the mapping results is normalized to obtain a first association weight between each of the first variable and the second variable.

5. The method according to claim 2, wherein: Before performing weighted summation of the first association weight between the first variable and the second variable and the second association weight between the first variable and the second variable to obtain the target association weight between the first variable and the second variable, the method further includes: respectively calculating a similarity between the third vector representation and each of the plurality of second association weights; The second association weight having the greatest similarity with the third vector representation is used as the second association weight corresponding to the target object.

6. The method according to claim 2, wherein: The method further comprises: For each of the first vector representations, inputting the first vector representation into a third neural network corresponding thereto to obtain a third predicted value of the second variable at a target time point; According to each of the third prediction values ​​and each of the target association weights, a target prediction value of the second variable at the target time point is obtained.

7. The method according to claim 6, wherein: The encoder is a long short-term memory network LSTM, a gated recurrent unit GRU, a time domain convolutional network TCN or a convolutional neural network CNN.

8. An information processing method, wherein: The method comprises: Acquire a time series of each first variable of a plurality of first variables of target multimedia data, wherein the target multimedia data includes an image or video corresponding to a product, and the plurality of first variables include a click volume, a page view volume, and a forwarding volume of the target multimedia data; Based on each of the time series, determining a first association weight of each of the first variables and a second variable of the target multimedia data, wherein the second variable includes a sales volume of a commodity corresponding to the target multimedia data; For each of the first variables, a first association weight between the first variable and the second variable and a second association weight between the first variable and the second variable are weightedly summed to obtain a target association weight between the first variable and the second variable, wherein the first association weight is a personalized causal relationship weight between the first variable and the second variable of the target multimedia data, the second association weight is a degree of association between the first variable and the second variable shared by the target multimedia data and multiple multimedia data, the second association weight is a degree of association between the first variable of each multimedia data in the multiple multimedia data and the second variable of the multimedia data, and the degree of association between the first variable of each multimedia data and the second variable of each multimedia data is the same.

9. The method according to claim 8, wherein: The determining, based on each of the time series, a first association weight of each of the first variables and a second variable of the target multimedia data comprises: For each of the time series, encoding the time series using an encoder to obtain a first vector representation corresponding to the time series; Inputting a second vector representation into a first neural network corresponding thereto to obtain a first predicted value of the second variable at a target time point, the second vector representation including each of the first vector representations; For each target first vector representation, concatenate each other first vector representation except the target first vector representation, and input the concatenated vector representation into a second neural network corresponding to the target first vector representation to obtain a second predicted value of the second variable at the target time point; Based on the first predicted value and each of the second predicted values, a third vector representation is obtained, wherein the third vector representation includes a first association weight of each of the first variables and the second variable.

10. The method according to claim 9, wherein: The method further comprises: For each of the first vector representations, inputting the first vector representation into a third neural network corresponding thereto to obtain a third predicted value of the second variable at a target time point; According to each of the third prediction values ​​and each of the target association weights, a target prediction value of the second variable at the target time point is obtained.

11. An information processing method, wherein: The method comprises: Obtaining a time series of each user behavior variable among multiple user behavior variables of a target object; Based on each of the time series, determining a first association weight between each of the user behavior variables and the transaction variable of the target object; For each of the user behavior variables, a first association weight between the user behavior variable and the transaction variable and a second association weight between the user behavior variable and the transaction variable are weighted and summed to obtain a target association weight between the user behavior variable and the transaction variable, wherein the first association weight is a personalized causal relationship weight between the user behavior variable and the transaction variable of the target object, the second association weight is a degree of association between the user behavior variable and the transaction variable shared by the target object and multiple objects, the second association weight is a degree of association between the user behavior variable of each object in the multiple objects and the transaction variable of the object, and the degree of association between the user behavior variable of each object and the transaction variable of the object is the same; Determining prompt information associated with the target object based on the target association weight between each of the user behavior variables and the transaction variable; The prompt information is displayed.

12. The method according to claim 11, wherein: The multiple user behavior variables include at least two of the following variables: click volume, forwarding volume, pageview volume, like volume, and comment volume; The transaction variable is any one of the following variables: sales quantity, price and sales amount.

13. The method according to claim 11, wherein: The determining, based on each of the time series, a first association weight between each of the user behavior variables and the transaction variable of the target object includes: Each of the time series is analyzed by a Granger causality analysis method to obtain a first correlation weight between each of the user behavior variables and the transaction variable of the target object.

14. The method according to claim 11, wherein: The prompt information includes a target predicted value of a transaction variable of the target object at a target time point.

15. The method according to claim 11, wherein: The prompt information includes a target association weight between each of the user behavior variables and the transaction variable; After displaying the prompt information, the method further includes: receiving a selection instruction of the user for the prompt information, and determining a first target association weight; receiving a user's instruction to change the first target association weight; In response to the change instruction, the first target association weight is changed to obtain a changed first target association weight; The changed first target association weight is displayed.

16. An information processing device, wherein: The device comprises: A time series acquisition module, used to acquire a time series of each first variable of a plurality of first variables of a target object, wherein the target object includes a target product, and the plurality of first variables include a click rate and a pageview; A first determination module is used to determine, based on each of the time series, a first association weight between each of the first variables and a second variable of the target object, where the second variable is the sales volume of the target commodity; The second determination module is used to, for each of the first variables, perform a weighted sum of a first association weight between the first variable and the second variable and a second association weight between the first variable and the second variable to obtain a target association weight between the first variable and the second variable, wherein the first association weight is a personalized causal relationship weight between the first variable and the second variable of the target object, the second association weight is a degree of association between the first variable and the second variable shared by the target object and multiple objects, the second association weight is a degree of association between the first variable of each object in the multiple objects and the second variable of the object, and the degree of association between the first variable of each object and the second variable of the object is the same.

17. An information processing device, wherein: The device comprises: A time series acquisition module, used to acquire a time series of each first variable of a plurality of first variables of target multimedia data, wherein the target multimedia data includes an image or video corresponding to a product, and the plurality of first variables include a click volume, a page view volume, and a forwarding volume of the target multimedia data; A first determination module, configured to determine, based on each of the time series, a first association weight between each of the first variables and a second variable of the target multimedia data, wherein the second variable includes a sales volume of a commodity corresponding to the target multimedia data; The second determination module is used to perform a weighted summation of the first association weight between the first variable and the second variable and the second association weight between the first variable and the second variable for each of the first variables to obtain a target association weight between the first variable and the second variable, wherein the first association weight is the personalized causal relationship weight between the first variable and the second variable of the target multimedia data, the second association weight is the degree of association between the first variable and the second variable shared by the target multimedia data and multiple multimedia data, the second association weight is the degree of association between the first variable of each multimedia data in the multiple multimedia data and the second variable of the multimedia data, and the degree of association between the first variable of each multimedia data and the second variable of each multimedia data is the same.

18. An information processing device, wherein: The device comprises: A time series acquisition module, used to acquire the time series of each user behavior variable among multiple user behavior variables of the target object; A first determination module, configured to determine, based on each of the time series, a first association weight between each of the user behavior variables and the transaction variable of the target object; a second determination module, configured to perform a weighted summation of a first association weight between the user behavior variable and the transaction variable and a second association weight between the user behavior variable and the transaction variable for each of the user behavior variables, to obtain a target association weight between the user behavior variable and the transaction variable, wherein the first association weight is a personalized causal relationship weight between the user behavior variable and the transaction variable of the target object, the second association weight is a degree of association between the user behavior variable and the transaction variable shared by the target object and multiple objects, the second association weight is a degree of association between the user behavior variable of each object in the multiple objects and the transaction variable of the object, and the degree of association between the user behavior variable of each object and the transaction variable of the object is the same; A prompt information determination module, configured to determine prompt information associated with the target object based on a target association weight between each of the user behavior variables and the transaction variable; The display module is used to display the prompt information.

19. A computing device, wherein: The computing device includes: a processor and a memory storing computer program instructions; When the processor executes the computer program instructions, the information processing method according to any one of claims 1 to 15 is implemented.

20. A computer storage medium, wherein: The computer storage medium stores computer program instructions, and when the computer program instructions are executed by the processor, the information processing method according to any one of claims 1 to 15 is implemented.