Business Processing Method, Apparatus, and Device Based on Causal Knowledge

Abstract

The embodiments of this specification disclose a service processing method, apparatus, and device based on causal knowledge. The solution includes: determining a plurality of features related to a service object; determining the causal knowledge of each of the features according to the to-be-predicted index of the service object, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted index; generating a causal graph including nodes for representing each of the features according to the causal knowledge; generating an embedding vector of the nodes according to the values of the plurality of features and the causal graph; predicting the value of the to-be-predicted index of the service object according to the embedding vector; and performing corresponding service processing according to the prediction result.

Description

Technical Field

[0001] This specification relates to the field of machine learning technology, and in particular, to a business processing method, apparatus, and device based on causal knowledge. Background Art

[0002] With the rapid popularization of the Internet and mobile terminals, a large number of applications have emerged. Users can conveniently install the clients of various applications on mobile terminals, and the corresponding servers provide online services for the clients, so that users can obtain corresponding services.

[0003] In the process of serving users, there are often multiple alternative solutions for the same purpose. These purposes can be either improvements to the underlying support system, such as optimizing the distributed architecture of the server, optimizing the transaction processing flow, etc., or improvements at the business level for users, such as optimizing the interaction method with users, increasing the business traffic of users, etc. And these purposes are ultimately to help the business run better and improve the user experience. Currently, the method of controlled experiments is often used to test multiple alternative solutions to help select a better solution and then apply it to actual business.

[0004] However, in actual applications, there are often situations where it is not suitable or even impossible to conduct controlled experiments due to reasons such as too high experimental costs. Based on this, in the scenario where multiple alternative solutions exist, a more applicable and reliable business processing solution is needed. Summary of the Invention

[0005] One or more embodiments of this specification provide a business processing method, apparatus, device, and storage medium based on causal knowledge to solve the following technical problem: In the scenario where multiple alternative solutions exist, a more applicable and reliable business processing solution is needed.

[0006] To solve the above technical problem, one or more embodiments of this specification are implemented as follows:

[0007] A business processing method based on causal knowledge provided by one or more embodiments of this specification includes:

[0008] Determine multiple features related to a business object;

[0009] According to the prediction index to be predicted of the business object, determine the causal knowledge of each feature, where the causal knowledge reflects the causal relationship between the corresponding feature and the prediction index to be predicted;

[0010] Generate a causal graph including nodes representing each feature according to the causal knowledge;

[0011] Generate an embedding vector of the node according to the values of the multiple features and the causal graph;

[0012] Predict the value of the to-be-predicted indicator of the business object according to the embedding vector;

[0013] Perform corresponding business processing according to the prediction result.

[0014] A business processing apparatus based on causal knowledge provided by one or more embodiments of this specification includes:

[0015] A business feature determination module that determines a plurality of features related to a business object;

[0016] A causal knowledge determination module that determines the causal knowledge of each feature according to the to-be-predicted indicator of the business object, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted indicator;

[0017] A causal graph generation module that generates a causal graph including nodes for representing each feature according to the causal knowledge;

[0018] An embedding vector generation module that generates embedding vectors of the nodes according to the values of the plurality of features and the causal graph;

[0019] A business indicator prediction module that predicts the value of the to-be-predicted indicator of the business object according to the embedding vector;

[0020] A corresponding business processing module that performs corresponding business processing according to the prediction result.

[0021] A business processing device based on causal knowledge provided by one or more embodiments of this specification includes:

[0022] At least one processor; and,

[0023] A memory communicatively connected to the at least one processor; wherein,

[0024] The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to:

[0025] Determine a plurality of features related to a business object;

[0026] Determine the causal knowledge of each feature according to the to-be-predicted indicator of the business object, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted indicator;

[0027] Generate a causal graph including nodes for representing each feature according to the causal knowledge;

[0028] Generate an embedding vector for the node according to the values of the multiple features and the causal graph;

[0029] Predict the value of the to-be-predicted indicator of the business object according to the embedding vector;

[0030] Perform corresponding business processing according to the prediction result.

[0031] A non-volatile computer storage medium provided by one or more embodiments of this specification stores computer-executable instructions, and the computer-executable instructions are set to:

[0032] Determine multiple features related to the business object;

[0033] According to the to-be-predicted indicator of the business object, determine the causal knowledge of each feature, and the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted indicator;

[0034] Generate a causal graph containing nodes for representing each feature according to the causal knowledge;

[0035] Generate an embedding vector for the node according to the values of the multiple features and the causal graph;

[0036] Predict the value of the to-be-predicted indicator of the business object according to the embedding vector;

[0037] Perform corresponding business processing according to the prediction result.

[0038] One or more of the above technical solutions adopted by the embodiments of this specification can achieve the following beneficial effects: By embedding causal knowledge and constructing a causal graph structure, it is possible to freely and flexibly predict the impact of changes in some feature values on business effects without relying on experiments, and then be able to guide the selection decision of the alternative solutions to which these changes belong to help the business proceed better. This method not only has better applicability and higher reliability, but also has good interpretability. Brief Description of the Drawings

[0039] In order to more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0040] Figure 1 It is a schematic flowchart of a business processing method based on causal knowledge provided by one or more embodiments of this specification;

[0041] Figure 2 Schematic diagram of a solution architecture for estimating causal knowledge provided for one or more embodiments of this specification;

[0042] Figure 3 Schematic diagram of a causal sub - graph provided for one or more embodiments of this specification;

[0043] Figure 4 Schematic diagram of a global causal graph provided for one or more embodiments of this specification;

[0044] Figure 5 For one or more embodiments of this specification, in an application scenario, Figure 1 Schematic diagram of a specific implementation of the method in;

[0045] Figures 6(a) and 6(b) are comparison diagrams of the effects of some prediction models including the model in on the test data set provided for one or more embodiments of this specification; Figure 5 in the model provided for one or more embodiments of this specification;

[0046] Figure 7 Schematic diagram of the structure of a business processing device based on causal knowledge provided for one or more embodiments of this specification;

[0047] Figure 8 Schematic diagram of the structure of a business processing device based on causal knowledge provided for one or more embodiments of this specification. Specific implementation manners

[0048] Embodiments of this specification provide a business processing method, device, equipment, and storage medium based on causal knowledge.

[0049] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this application, rather than all the embodiments. Based on the embodiments of this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this application.

[0050] In the background art, it is mentioned that the applicability of controlled experiments in practical applications is prone to problems. High cost is one of the typical problems. Moreover, the limitations of the experiments are relatively large and the flexibility is poor, resulting in a reduction in the reliability of the experimental results, and thus the reference value for scheme selection is also reduced. This solution has considered estimating the improvement of business effects brought by different schemes specifically within the framework of the difference-in-differences idea. However, the disadvantages of this solution are also obvious. Its interpretability is relatively low, it is difficult to interpret the model and transfer knowledge according to actual needs, and its persuasiveness in practical business applications is low. Based on this, this solution has proposed a business effect prediction scheme based on embedding causal knowledge and constructing a graph structure, which improves the reliability and interpretability of the model. The following will continue to elaborate on this scheme in detail.

[0051] Figure 1 The flowchart of a causal knowledge-based business processing method for a client provided by one or more embodiments of this specification. This method can be applied to different business fields, such as: electronic payment business field, e-commerce business field, instant messaging business field, game business field, official business field, etc. This process can be executed by devices in the corresponding fields, such as the risk control strategy server in the electronic payment field, the marketing strategy server and intelligent customer service server in the e-commerce business field, and so on. Some input parameters or intermediate results in the process allow manual intervention and adjustment to help improve accuracy.

[0052] Figure 1 The process in [the figure] may include the following steps:

[0053] S102: Determine multiple features related to the business object.

[0054] In one or more embodiments of this specification, the business object generally can be a user, a commodity or a merchant, or can also be the software and hardware that support the business object, such as a server, a distributed system, public facilities, etc.

[0055] The features related to the business object can include some attributes that it itself has, and can also include some attributes imposed on it by the outside world subsequently. Taking a commodity as an example, the attributes it itself has, such as raw materials, quality, appearance, etc., and some attributes imposed on it by the outside world subsequently, such as price, selling discount, selling area, inventory, etc. Many of these features can be flexibly intervened. Interventions for different features, or different degrees of intervention for the same feature, can correspondingly constitute different optional schemes, and the expected purpose of the intervention is to make an index of a certain aspect of the business object improve in a good direction.

[0056] S104: Determine the causal knowledge of each of the features according to the to-be-predicted metric of the service object, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted metric.

[0057] The to-be-predicted metrics include the metrics that are desired to be improved by intervening in the features. Taking a commodity as an example, the to-be-predicted metrics may include metrics representing sales conditions or customer conversion conditions, such as the sales volume of the commodity, the number of users who click on the commodity, and so on.

[0058] In one or more embodiments of this specification, the corresponding feature is the cause that affects the change of the to-be-predicted metric, and the to-be-predicted metric correspondingly belongs to the result corresponding to this cause. There may be multiple different features that can affect the to-be-predicted metric. The causal knowledge can also reflect to what extent the corresponding feature can affect the to-be-predicted metric. The causal knowledge that measures this extent can be called causal weight. The causal knowledge can also contain other forms of content. For example, if a certain feature that a service object already has can affect the to-be-predicted metric of the service object, and this feature can be affected by another feature (such as a certain new type of marketing strategy), and the service object does not currently have this other feature, but can be actively intervened to add this other feature to the service object, then the causal knowledge of this feature can also contain this other feature. Based on such causal knowledge, it can help predict whether it is necessary to add this other feature to the service object.

[0059] In one or more embodiments of this specification, the causal knowledge can also reflect whether the to-be-predicted metric can in turn affect the corresponding feature, and to what extent it can reversely affect the to-be-predicted metric. This has relatively high reference value for some features that can adaptively change. The deeper utilization of the causal knowledge in this regard will be specifically described later.

[0060] S106: Generate a causal graph including nodes used to represent each of the features according to the causal knowledge.

[0061] In one or more embodiments of this specification, the causal knowledge of each feature is relatively discrete and may not be able to show the causal relationship between different features. Moreover, the causal knowledge between the feature and the to-be-predicted metric can also change due to the influence of other features. Based on this, learn a causal graph representing the global causal relationship according to the causal knowledge of each feature. The nodes of the causal graph at least include nodes used to represent each of the features (for example, each feature is represented as a node), and may also include nodes used to represent the to-be-predicted metric.

[0062] In the case of a large number of features, the causal relationships between the features are relatively complex. The causal relationship between the index to be predicted and the features may be indirectly reflected by the causal relationships between the features. In this case, it is also possible to consider not adding a node representing the index to be predicted to the causal graph. In this way, it helps to decouple the features and the index to be predicted to a certain extent, so as to facilitate reasoning about other indices based on the knowledge contained in the causal graph, and helps to improve the applicability and flexibility of the solution.

[0063] S108: Generate an embedding vector of the node according to the values of the multiple features and the causal graph.

[0064] Embed the values of multiple features into the representation of the node, so that the embedding vector can basically represent the corresponding node. On this basis, further represent the corresponding node based on the causal graph.

[0065] In one or more embodiments of this specification, the causal knowledge of the features is directly or indirectly reflected in the causal graph. It should be noted that since the causal graph considers the global situation, the causal knowledge of the features may change after being incorporated into the causal graph, rather than remaining unchanged. Through the causal graph, the causal knowledge of the features can be embedded into the representation of the node to a certain extent, so as to obtain an embedding vector with richer information and more targeted for the index to be predicted.

[0066] S110: Predict the value of the index to be predicted of the business object according to the embedding vector.

[0067] In one or more embodiments of this specification, after assigning values to the features, the obtained embedding vector can be input into a trained machine learning model to map the embedding vector to predict the value of the index to be predicted. The assignment of values to the features can be adjusted according to different alternative solutions, so as to obtain the values of the index to be predicted corresponding to these alternative solutions respectively.

[0068] In one or more embodiments of this specification, when comparing an alternative solution (referred to as an intervention solution) with another alternative solution or an implemented solution, the other alternative solution or an implemented solution can be used as a reference solution, obtain the values of multiple features in the reference solution, and obtain the value of the index to be predicted in the reference solution as a reference value. Then, based on the obtained values, make adjustments according to the intervention solution. For example, select at least one feature from these multiple features as an intervention feature and adjust its value, and the values of other features can remain unchanged. Then, generate an embedding vector of the node according to the adjusted values of the multiple features, so as to determine the influence of the adjustment of the value of the intervention feature on the reference value of the index to be predicted.

[0069] When measuring the impact, the gain value brought by the adjustment of the value of the intervention feature to the reference value of the to-be-predicted indicator can be determined according to the predicted result and the reference value of the to-be-predicted indicator (for example, subtracting the two). The gain may be positive or negative. Then, according to the gain value, it is determined whether to make corresponding adjustments to the service corresponding to the business object. For example, if the gain is positive, it can be decided to adopt the intervention plan, and corresponding resources can be additionally allocated for the implementation of the intervention plan. If the gain is negative, the intervention plan can be abandoned, or the intervention plan can be further modified and then re-predicted.

[0070] S112: Perform corresponding service processing according to the predicted result.

[0071] In one or more embodiments of this specification, the predicted result can reflect whether the corresponding alternative can improve the performance of the to-be-predicted indicator, and reflect how much the improvement degree is. Furthermore, it can help to make a decision on which alternative to choose, and whether the alternative can be further adjusted in a better direction. The service processing in S112 can include determining the gain brought by the alternative (for example, the plan waiting to be decided whether to newly implement) to the to-be-predicted indicator compared with the reference plan (for example, the currently actually adopted plan).

[0072] In addition, the service processing can also include more operations. For example, implementing the alternative, or allocating corresponding resources for the better implementation effect of the reference plan or the alternative. Such resources can be: computing resources of the server, cloud storage resources, etc., front-end resources exposed to users, traffic resources guided for merchants, and so on. Through such service processing, the service can be carried out better, serve users better, and improve the user experience.

[0073] By Figure 1 the method, by embedding causal knowledge and constructing a causal graph structure, it is possible to freely and flexibly predict the impact of changes in some feature values on the business effect through a graph model without relying on experiments. Furthermore, it can guide the selection decision of the alternative to which these changes belong to help the business run better. This method not only has better applicability and higher reliability, but also has good interpretability.

[0074] Based on Figure 1 the method, this specification also provides some specific implementation schemes and extended schemes of the method, which will be further described below.

[0075] In one or more embodiments of this specification, when determining the causal knowledge of each feature, it can also be learned and implemented based on a graph. Local causal graphs can be constructed for each feature respectively to reduce the computational complexity and focus more on that feature. This causal graph is called a causal sub-graph to facilitate differentiation from the causal graph reflecting the global situation in S106. Subsequently, the causal graph reflecting the global situation can also be obtained through the fusion and learning of each causal sub-graph. Based on this idea, an example of determining causal knowledge is given.

[0076] One or more embodiments of this specification provide a schematic diagram of a scheme architecture for estimating causal knowledge, as Figure 2 shown. Under the architecture of Figure 2 , the causal knowledge specifically includes causal weights. Correspondingly, the above-mentioned causal sub-graph is called a causal weight sub-graph.

[0077] This architecture mainly includes a data set and two large modules. The data set provides training samples for these two modules. Module 1 is a knowledge distillation and representation module, which contains one or more teacher models and can be used to share knowledge labels with Module 2, transfer knowledge to Module 2, and help the models in Module 2 train quickly. Module 2 is a multi-head (for example, multiple prediction heads or multiple prediction models, each corresponding to a feature respectively) causal weight calculation module. In Module 2, causal weight sub-graphs are constructed for each feature (denoted as X1, X2,..., Xn) respectively and input into the corresponding causal weight estimators for processing to estimate the causal weights (denoted as W1, W2,..., Wn) corresponding to each feature respectively.

[0078] Furthermore, one or more embodiments of this specification also provide a schematic diagram of a causal sub-graph, as Figure 3 shown. The causal weight sub-graph in Figure 2 can be constructed according to the idea of this graph. Under this idea, for each feature respectively, the following can be executed: taking this feature as the cause node, taking the predicted index of the business object or its corresponding effect as the result node, constructing an edge pointing from the cause node to the result node, selecting one or more confounding features for this feature as confounding nodes, constructing edges respectively pointing from the confounding nodes to the cause node and the result node, and based on the causal sub-graph composed of the edges between the cause node, the result node, and each confounding node, the effect estimation can be carried out accordingly, and the data reflecting the effect of this feature on the predicted index of the business object can be determined as the causal knowledge of this feature.

[0079] In Figure 2Among them, feature X1 is an intervention feature and serves as a cause node. Features X2 to Xm are corresponding confounding features and serve as confounding nodes. The pointing relationship can be visually seen from the arrow connection line. The effect corresponding to the prediction target serves as an effect node. Here, a teacher model can be introduced to provide labels for supervised learning, so as to more accurately determine the causal weight through effect estimation. Based on the above architecture, the causal weight of each feature can be calculated relatively independently, enabling efficient parallel computing and improving the overall implementation efficiency.

[0080] Based on the causal knowledge corresponding to each feature, a global causal graph can be established that at least includes nodes for representing these features. The specific method is, for example, directly fusing each causal subgraph or learning based on causal knowledge. Taking learning as an example, according to causal knowledge, the structural topology between the nodes for each feature can be learned using a Bayesian network (for example, using the PC algorithm based on causal analysis, etc.), and then the structural topology can be locally adjusted using expert knowledge, thus generating a causal graph that at least includes these nodes.

[0081] When there are many features and the causal relationships are relatively complex, the global causal graph is correspondingly more complex, with a much higher complexity compared to the causal subgraph, and can also more correctly represent the true and complete global causal relationship. Intuitively, one or more embodiments of this specification provide a schematic diagram of a global causal graph, as Figure 4 shown. In Figure 4 it, the small oval blocks represent nodes, and the arrow connection lines represent the causal relationships between the nodes.

[0082] In one or more embodiments of this specification, after obtaining the causal graph, the representation of the nodes (which can represent features) and business objects (which can be represented by multiple features) can be embedded according to the values of the features and causal knowledge, thereby generating corresponding embedding vectors.

[0083] For causal knowledge, it can be embedded by directly combining the values of the features, or it can also be embedded through the structural topology in the causal graph. The former method is particularly applicable when the causal knowledge is causal weight, because the embedding can be conveniently achieved by weighting the values of the features.

[0084] For example, the value of each feature is used as the value of the corresponding dimension in the vector to form a feature value vector. According to the causal weight, the value of each dimension in the feature value vector is weighted accordingly to obtain a feature weighted value vector. Then, according to the causal graph and the feature weighted value vector, the embedding vector of the node is generated. In this example, if the structural topology in the causal graph is used, it is equivalent to combining the two methods in the previous paragraph to achieve embedding. In this case, the graph structure vector of the node can be generated according to the topological structure in the causal graph, and the feature weighted value vector and the graph structure vector are fused to generate the embedding vector of the node. The graph structure vector can be obtained by using an adjacency matrix that reflects the topological structure.

[0085] It should be noted that the utilization value of the causal graph is not limited to this. Based on the causal graph, predictions are made by adjusting the value of the feature multiple times (that is, adjusting the cause) to try to achieve the target value of the indicator to be predicted. This efficiency still has room for improvement. This solution takes into account that in practical applications, the relationship between cause and effect is not necessarily an absolute one-way relationship (from cause to effect). There may also be situations where the effect reversely or even strongly affects the cause. For example, the inventory of goods can affect sales, and sales can in turn strongly affect inventory. There may be a situation of mutual game between the two. This solution is based on the causal graph to explore this game relationship and use it in the prediction process.

[0086] For example, a part of the topological structure contained in the causal graph (which can be randomly selected, or a part with a relatively small causal weight) can be obtained, including cause nodes and result nodes, wherein the cause nodes represent the characteristics of the business object, and the result nodes represent the indicators to be predicted of the business object or their corresponding effects. The direction of the edge between the obtained cause nodes and result nodes is adjusted to the reverse direction to represent the reverse influence relationship, a local reverse topological structure, and a local reverse structure vector is generated for the local reverse topological structure. The local reverse structure vector and the feature weight value vector are fused to generate an embedding vector of the node, which can then be used to predict the indicators to be predicted.

[0087] It should be noted that although the local reverse structure vector fusion was performed in the previous paragraph, errors were also introduced. The errors can be controlled to an acceptable level based on the comparison of differentiated prediction results. For example, the value of the indicator to be predicted is predicted based on the embedding vector of the node obtained by fusing the local reverse structure vector and the embedding vector of the node obtained without fusing the local reverse structure vector, and the difference between the prediction results is compared. If the difference is less than the set threshold, the local reverse topological structure can be retained, otherwise, it can be corrected.

[0088] In one or more embodiments of the present specification, since a reverse structure is generated and the end point of the reverse structure is a feature (originally a starting point or an intermediate point), it is easier to further extend (find the next possible result) compared to the index to be predicted. Therefore, it is considered to supplement the causal graph to achieve a wider prediction.

[0089] Specifically, for example, taking the cause node as the starting point, determining features other than multiple features or indexes other than the index to be predicted, and adding them as supplementary nodes to the locally reversed topological structure. If the above gap is less than the set threshold, then the embedding vector of the node obtained according to the un-fused locally reversed structure vector can be used to predict the values of features or indexes other than the above as the optimization reference value.

[0090] According to the previous description, one or more embodiments of the present specification also provide a schematic diagram of a specific implementation scheme of the method in an application scenario, as Figure 1 shown. Figure 5

[0091] The scheme is divided into three parts. The first part lists the data that needs to be prepared as the basis for subsequent feature fusion and prediction. These data include the values of each feature, which can be represented by a feature value vector, and also include the calculated causal weights of each feature, and also include the adjacency matrix of the causal graph of each feature globally. The adjacency matrix is transformed into a graph structure vector, such as using transformation algorithms such as DeepWalk and Node2Vec. The second part shows the process of feature fusion, including: multiplying each dimension in the feature value vector by the corresponding causal weight to obtain a feature weighted value vector, and then fusing the feature weighted value vector, the feature value vector and the graph structure vector (for example, concatenating them together) to generate the embedding vector of the node. The third part inputs the embedding vector into a pre-trained graph convolutional neural network. Through the processing of the graph convolutional neural network, the value of the index to be predicted for the business object is output, denoted as y_hat, where hat represents the circumflex symbol placed above. The reference value of the index to be predicted is denoted as y 0 , and the gain value brought by the intervention feature to the index to be predicted is obtained based on these two values.

[0092] Figures 6(a) and 6(b) are the effect comparison diagrams of some prediction models including Figure 5 the model in the present specification on the test data set.

[0093] In Figure 6(a), the ordinate represents the absolute error of the prediction of the gain value of the effect, and the abscissa represents the data respectively when different numbers of confounding features are adopted. A total of 6 prediction models are used. The first 5 are existing models, and the 6th is Figure 5 ​The model proposed in [reference] is a graph convolutional neural network (GCN) based on causal weights. The order of the bar charts from left to right in each group of data is the same as the order of the meanings of the bar charts from top to bottom in the upper right corner. It can be seen that the absolute error of the prediction results of the model proposed in this solution is relatively small, and thus the accuracy is relatively high.

[0094] In Figure 6(b), the prediction results of several prediction models are compared with other types of errors, including mean squared error (MSE), KL divergence (KL Divergence), etc. It can be seen that the error performance of the prediction results of the model proposed in this solution is also very good.

[0095] This solution proposes to use the causal weights of features (in order to reduce costs, the causal weights can also be calculated based on CATE values, shapley values, uplift values, qini values, etc.) for subsequent feature weighting, proposes an extensible and parallelizable framework for calculating causal weights, and also proposes a prediction model that embeds causal weights into a graph model, which helps to predict the impacts that various interventions may bring to indicators more reliably at a lower cost, and thus can effectively assist in the decision-making of intervention plan selection, contribute to the better operation of the business, and improve the user experience.

[0096] Based on the same idea, one or more embodiments of this specification also provide devices and equipment corresponding to the above methods, such as Figure 7 、 Figure 8 as shown.

[0097] Figure 7 FIG. [figure number] is a schematic structural diagram of a business processing device based on causal knowledge provided by one or more embodiments of this specification. The device includes:

[0098] A service feature determination module 702 that determines a plurality of features related to a service object;

[0099] A causal knowledge determination module 704 that determines the causal knowledge of each feature according to the to-be-predicted index of the service object, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted index;

[0100] A causal graph generation module 706 that generates a causal graph including nodes for representing each feature according to the causal knowledge;

[0101] An embedding vector generation module 708 that generates an embedding vector of the node according to the values of the plurality of features and the causal graph;

[0102] A service index prediction module 710 that predicts the value of the to-be-predicted index of the service object according to the embedding vector;

[0103] The corresponding service processing module 712 performs corresponding service processing according to the predicted result.

[0104] Optionally, the embedding vector generation module 708 obtains the reference value of the to-be-predicted indicator and the values of the multiple corresponding features;

[0105] Select at least one feature from the multiple features as an intervention feature and adjust its value;

[0106] Generate the embedding vector of the node according to the adjusted values of the multiple features, so as to determine the influence of the adjustment of the value of the intervention feature on the reference value of the to-be-predicted indicator.

[0107] Optionally, the corresponding service processing module 712 determines the gain value brought by the adjustment of the value of the intervention feature to the reference value of the to-be-predicted indicator according to the predicted result and the reference value of the to-be-predicted indicator;

[0108] Determine whether to perform corresponding adjustment on the service corresponding to the service object according to the gain value.

[0109] Optionally, the causal knowledge determination module 704 executes for each of the features:

[0110] Take this feature as a cause node, take the to-be-predicted indicator of the service object as an effect node, and construct an edge pointing from the cause node to the effect node;

[0111] Select one or more confounding features as confounding nodes for this feature, and construct edges respectively pointing from the confounding nodes to the cause node and the effect node;

[0112] Perform effect estimation according to the causal sub-graph composed of the edges between the cause node, the effect node, and each of the confounding nodes, and determine the data reflecting the effect of this feature on the to-be-predicted indicator of the service object as the causal knowledge of this feature.

[0113] Optionally, the causal graph generation module 706 uses the Bayesian network to learn the structural topology between the nodes for each of the features according to the causal knowledge;

[0114] Locally adjust the structural topology using expert knowledge to generate a causal graph including each of the nodes.

[0115] Optionally, the causal knowledge is a causal weight;

[0116] The embedding vector generation module 708 forms a feature value vector by using the values of each of the features as the values of the corresponding dimensions in the vector;

[0117] According to the causal weights, the values of each dimension in the eigenvalue vector are weighted accordingly to obtain a feature-weighted value vector;

[0118] According to the causal graph and the feature-weighted value vector, an embedding vector of the node is generated.

[0119] Optionally, the embedding vector generation module 708 generates a graph structure vector of the node according to the topological structure in the causal graph;

[0120] The feature-weighted value vector and the graph structure vector are fused to generate an embedding vector of the node.

[0121] Optionally, the embedding vector generation module 708 obtains cause nodes and result nodes included in a part of the topological structure in the causal graph, where the cause nodes represent features of the business object, and the result nodes represent the to-be-predicted metrics of the business object or their corresponding effects;

[0122] The direction of the edge between the obtained cause nodes and result nodes is adjusted to be reversed to form a locally reversed topological structure;

[0123] A locally reversed structure vector is generated for the locally reversed topological structure;

[0124] The locally reversed structure vector and the feature-weighted value vector are fused to generate an embedding vector of the node.

[0125] Optionally, the embedding vector generation module 708 uses the cause nodes as starting points to determine features other than the multiple features or metrics other than the to-be-predicted metrics, and adds them as supplementary nodes to the locally reversed topological structure;

[0126] The business metric prediction module 710 predicts the value of the to-be-predicted metric of the business object according to the embedding vector, and predicts the value of the to-be-predicted metric respectively according to the embedding vector of the node obtained by fusing the locally reversed structure vector and the embedding vector of the node without fusing the locally reversed structure vector;

[0127] If the gap between the prediction results is less than a set threshold, then according to the embedding vector of the node without fusing the locally reversed structure vector, the value of the feature other than or the metric other than is predicted as an optimization reference value.

[0128] Optionally, the embedding vector generation module 708 fuses the feature-weighted value vector, the eigenvalue vector, and the graph structure vector to generate an embedding vector of the node.

[0129] Optionally, the service metric prediction module 710 inputs the embedding vector into a pre-trained graph convolutional neural network;

[0130] Through the processing of the graph convolutional neural network, the predicted value of the to-be-predicted metric for the service object is output.

[0131] Optionally, the service object includes a commodity or a merchant, and the to-be-predicted metric includes a metric representing a sales situation or a customer conversion situation.

[0132] Figure 8 The figure is a schematic structural diagram of a service processing device based on causal knowledge provided by one or more embodiments of this specification. The device includes:

[0133] At least one processor; and,

[0134] A memory communicatively connected to the at least one processor; wherein,

[0135] The memory stores instructions executable by the at least one processor. When the instructions are executed by the at least one processor, the at least one processor is enabled to:

[0136] Determine multiple features related to a service object;

[0137] According to the to-be-predicted metric of the service object, determine the causal knowledge of each feature, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted metric;

[0138] Generate a causal graph including nodes for representing each feature according to the causal knowledge;

[0139] Generate an embedding vector of the nodes according to the values of the multiple features and the causal graph;

[0140] Predict the value of the to-be-predicted metric of the service object according to the embedding vector;

[0141] Perform corresponding service processing according to the prediction result.

[0142] The processor and the memory can communicate through a bus, and the device may further include an input / output interface for communicating with other devices.

[0143] Based on the same idea, one or more embodiments of this specification also provide a non-volatile computer storage medium corresponding to the Figure 1 method in the storage, storing computer-executable instructions, and the computer-executable instructions are set to:

[0144] Determine multiple features related to a service object;

[0145] Determine the causal knowledge of each of the features according to the to-be-predicted indicator of the business object, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted indicator;

[0146] Generate a causal graph containing nodes for representing each of the features according to the causal knowledge;

[0147] Generate an embedding vector of the nodes according to the values of the multiple features and the causal graph;

[0148] Predict the value of the to-be-predicted indicator of the business object according to the embedding vector;

[0149] Perform corresponding business processing according to the result of the prediction.

[0150] In the 1990s, improvements to a technology could be clearly distinguished as either hardware improvements (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software improvements (improvements to method flows). However, with the development of technology, many method flow improvements today can be regarded as direct improvements to hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented using a hardware entity module. For example, a Programmable Logic Device (PLD) (such as a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logical function is determined by the user programming the device. Designers can program themselves to "integrate" a digital system onto a single PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a Hardware Description Language (HDL), and there is not just one type of HDL, but many types, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. Currently, the most commonly used are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that by simply performing a little logical programming on the method flow using the above-mentioned several hardware description languages and programming it into an integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0151] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to make the controller implement the same function in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or the structures within the hardware component.

[0152] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0153] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0154] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, the embodiments of this specification can take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Moreover, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0155] This specification is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the specification. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable data processing devices to produce a machine, such that the instructions executed by the processors of the computer or other programmable data processing devices produce means for implementing the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in multiple blocks or multiple blocks specified in the function.

[0156] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory produce a manufactured article including instruction means that implement the functions specified in one flow Figure 1 one flow or multiple flows and / or blocks Figure 1 or in multiple blocks or multiple blocks specified in the function.

[0157] It should also be noted that the term "including", "comprising", or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, commodity, or device including a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity, or device. Without further limitation, an element defined by the statement "including one..." does not exclude the existence of additional identical elements in the process, method, commodity, or device including the said element.

[0158] This specification can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This specification can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0159] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other, and the differences between each embodiment and other embodiments are emphasized. In particular, for the embodiments of devices, equipment, and non-volatile computer storage media, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the relevant parts of the method embodiments for the relevant content.

[0160] The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0161] The above is only one or more embodiments of this specification and is not intended to limit this specification. For those skilled in the art, one or more embodiments of this specification can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of this specification shall be included within the scope of the claims of this specification.

Claims

1. A business processing method based on causal knowledge, including: Determining a plurality of features related to a business object; According to the to-be-predicted index of the business object, determining the causal knowledge of each feature, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted index, and the causal knowledge is a causal weight; Generating a causal graph including nodes representing each feature according to the causal knowledge; Generating an embedding vector of the node according to the values of the plurality of features and the causal graph, specifically including: taking the values of each feature as the values of the corresponding dimensions in the vector to form a feature value vector, weighting the values of each dimension in the feature value vector accordingly according to the causal weight to obtain a feature weighted value vector, and generating the embedding vector of the node according to the causal graph and the feature weighted value vector; obtaining the reference value of the to-be-predicted index and the values of the corresponding plurality of features, selecting at least one feature from the plurality of features as an intervention feature and adjusting its value, and generating the embedding vector of the node according to the adjusted values of the plurality of features to be used to determine the influence of the adjustment of the value of the intervention feature on the reference value of the to-be-predicted index; Predicting the value of the to-be-predicted index of the business object according to the embedding vector; Performing corresponding business processing according to the prediction result, specifically including: determining the gain value brought by the adjustment of the value of the intervention feature to the reference value of the to-be-predicted index according to the prediction result and the reference value of the to-be-predicted index, and determining whether to perform corresponding adjustment on the business corresponding to the business object according to the gain value; allocating corresponding resources, and such resources are at least one of the following: computing resources of the server, cloud storage resources, front-end resources exposed to users, and traffic resources guided for merchants; When the business object is a commodity, the features related to the business object include at least one of the following: raw materials, quality, appearance, price, selling discount, selling area, inventory; the to-be-predicted index includes an index representing the sales situation or the customer conversion situation; The business object includes a commodity, the index representing the sales situation includes the sales amount of the commodity, and the index representing the customer conversion situation includes the number of users who click on the commodity.

2. The method according to claim 1, where determining the causal knowledge of each feature according to the to-be-predicted index of the business object specifically includes: For each feature, respectively perform: Taking this feature as a cause node, taking the to-be-predicted index of the business object or its corresponding effect as an effect node, and constructing an edge pointing from the cause node to the effect node; Selecting one or more confounding features as confounding nodes for this feature, and constructing edges respectively pointing from the confounding nodes to the cause node and the effect node; Performing effect estimation according to the causal sub-graph formed by the edges between the cause node, the effect node, and each confounding node, and determining the data reflecting the effect of this feature on the to-be-predicted index of the business object as the causal knowledge of this feature.

3. The method according to claim 1, wherein generating a causal graph including nodes for representing each of the features according to the causal knowledge specifically includes: According to the causal knowledge, using a Bayesian network to learn the structural topology between the nodes for each of the features; Using expert knowledge to locally adjust the structural topology to generate a causal graph including each of the nodes.

4. The method according to claim 1, wherein generating the embedding vector of the node according to the causal graph and the feature weight value vector specifically includes: Generating a graph structure vector of the node according to the topological structure in the causal graph; Fusing the feature weight value vector and the graph structure vector to generate the embedding vector of the node.

5. The method according to claim 1, wherein generating the embedding vector of the node according to the causal graph and the feature weight value vector specifically includes: Obtaining cause nodes and result nodes included in a part of the topological structure in the causal graph, where the cause nodes represent the features of the business object, and the result nodes represent the to-be-predicted indicators of the business object or their corresponding effects; Adjusting the direction of the edges between the obtained cause nodes and result nodes to be reversed to form a locally reversed topological structure; Generating a locally reversed structure vector for the locally reversed topological structure; Fusing the locally reversed structure vector and the feature weight value vector to generate the embedding vector of the node.

6. The method according to claim 5, wherein adjusting the direction of the edges between the obtained cause nodes and result nodes to be reversed to form a locally reversed topological structure further includes: Taking the cause nodes as starting points, determining features other than the multiple features or indicators other than the to-be-predicted indicators, and adding them as supplementary nodes to the locally reversed topological structure; Predicting the values of the to-be-predicted indicators respectively according to the embedding vector of the node obtained by fusing the locally reversed structure vector and the embedding vector of the node obtained without fusing the locally reversed structure vector; If the gap between the prediction results is less than a set threshold, predicting the values of the features other than or the indicators other than according to the embedding vector of the node obtained without fusing the locally reversed structure vector as the optimization reference value.

7. The method according to claim 4, wherein fusing the feature weight value vector and the graph structure vector to generate the embedding vector of the node specifically includes: Fusing the feature weight value vector, the feature value vector and the graph structure vector to generate the embedding vector of the node.

8. The method according to claim 1, wherein predicting the value of the to-be-predicted indicator of the business object according to the embedding vector specifically includes: Inputting the embedding vector into a pre-trained graph convolutional neural network; Through the processing of the graph convolutional neural network, outputting the predicted value of the to-be-predicted indicator of the business object.

9. The method according to any one of claims 1 to 8, wherein the business object includes a commodity or a merchant, and the to-be-predicted indicator includes an indicator representing a sales situation or a customer conversion situation.

10. A business processing device based on causal knowledge, comprising: A business feature determination module that determines multiple features related to a business object; A causal knowledge determination module that determines the causal knowledge of each feature according to the to-be-predicted index of the business object, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted index, and the causal knowledge is a causal weight; A causal graph generation module that generates a causal graph containing nodes representing each feature according to the causal knowledge; An embedding vector generation module that generates an embedding vector of the node according to the values of the multiple features and the causal graph, specifically including: taking the values of each feature as the values of the corresponding dimensions in the vector to form a feature value vector, weighting the values of each dimension in the feature value vector accordingly according to the causal weight to obtain a feature weighted value vector, and generating the embedding vector of the node according to the causal graph and the feature weighted value vector; obtaining the reference value of the to-be-predicted index and the values of the corresponding multiple features, selecting at least one feature from the multiple features as an intervention feature and adjusting its value, and generating the embedding vector of the node according to the adjusted values of the multiple features to be used to determine the influence of the adjustment of the value of the intervention feature on the reference value of the to-be-predicted index; A business index prediction module that predicts the value of the to-be-predicted index of the business object according to the embedding vector; A corresponding business processing module that performs corresponding business processing according to the prediction result, specifically including: determining the gain value brought by the adjustment of the value of the intervention feature to the reference value of the to-be-predicted index according to the prediction result and the reference value of the to-be-predicted index, and determining whether to perform corresponding adjustment on the business corresponding to the business object according to the gain value; allocating corresponding resources, and such resources are at least one of the following: the computing resources of the server, cloud storage resources, front-end resources exposed to users, and traffic resources guided for merchants; When the business object is a commodity, the features related to the business object include at least one of the following: raw materials, quality, appearance, price, selling discount, selling area, inventory; the to-be-predicted index includes an index representing the sales situation or the customer conversion situation; The business object includes a commodity, the index representing the sales situation includes the sales amount of the commodity, and the index representing the customer conversion situation includes the number of users who click on the commodity.

11. The device according to claim 10, wherein the causal knowledge determination module performs the following for each feature: Taking this feature as a cause node, taking the to-be-predicted index of the business object or its corresponding effect as a result node, and constructing an edge pointing from the cause node to the result node; Selecting one or more confounding features as confounding nodes for this feature, and constructing edges pointing from the confounding nodes to the cause node and the result node respectively; According to the causal sub-graph formed by the edges between the cause nodes, the result nodes, and each of the confounding nodes, perform effect estimation to determine the data reflecting the effect of this feature on the to-be-predicted indicator of the business object, and use it as the causal knowledge of this feature.

12. The device according to claim 10, wherein the causal graph generation module, according to the causal knowledge, uses a Bayesian network to learn the structural topology between the nodes for each of the features; Use expert knowledge to perform local adjustment on the structural topology to generate a causal graph including each of the nodes.

13. The device according to claim 10, wherein the embedding vector generation module, according to the topological structure in the causal graph, generates a graph structure vector of the node; Fuse the feature weighted value vector and the graph structure vector to generate an embedding vector of the node.

14. The device according to claim 10, wherein the embedding vector generation module obtains the cause nodes and result nodes included in a part of the topological structure in the causal graph, the cause nodes represent the features of the business object, and the result nodes represent the to-be-predicted indicator of the business object or its corresponding effect; Adjust the direction of the edges between the obtained cause nodes and result nodes to be reversed to obtain a locally reversed topological structure; Generate a locally reversed structure vector for the locally reversed topological structure; Fuse the locally reversed structure vector and the feature weighted value vector to generate an embedding vector of the node.

15. The device according to claim 14, wherein the embedding vector generation module, with the cause nodes as the starting point, determines features other than the multiple features, or indicators other than the to-be-predicted indicator, and adds them as supplementary nodes to the locally reversed topological structure; The business indicator prediction module predicts the value of the to-be-predicted indicator of the business object according to the embedding vector, and respectively predicts the value of the to-be-predicted indicator according to the embedding vector of the node obtained by fusing the locally reversed structure vector and the embedding vector of the node not fused with the locally reversed structure vector; If the gap between the prediction results is less than the set threshold, then according to the embedding vector of the node not fused with the locally reversed structure vector, predict the value of the feature other than or the indicator other than, as the optimization reference value.

16. The device according to claim 13, wherein the embedding vector generation module fuses the feature weighted value vector, the feature value vector, and the graph structure vector to generate an embedding vector of the node.

17. The device according to claim 10, wherein the business indicator prediction module inputs the embedding vector into a pre-trained graph convolutional neural network; Through the processing of the graph convolutional neural network, output the predicted value of the to-be-predicted indicator of the business object.

18. The device according to any one of claims 10 to 17, wherein the business object includes a commodity or a merchant, and the to-be-predicted indicator includes an indicator representing the sales situation or the customer conversion situation.

19. A business processing device based on causal knowledge, comprising: at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to: Determine a plurality of features related to a business object; According to the to-be-predicted indicator of the business object, determine the causal knowledge of each feature, where the causal knowledge reflects the causal relationship between the corresponding feature and the to-be-predicted indicator, and the causal knowledge is causal weights; Generate a causal graph including nodes for representing each feature according to the causal knowledge; Generate an embedding vector of the nodes according to the values of the plurality of features and the causal graph, specifically including: taking the values of each feature as the values of the corresponding dimensions in the vector to form a feature value vector, weighting the values of each dimension in the feature value vector accordingly according to the causal weights to obtain a feature weighted value vector, and generating the embedding vector of the nodes according to the causal graph and the feature weighted value vector; obtaining a reference value of the to-be-predicted indicator and the values of the corresponding plurality of features, selecting at least one feature from the plurality of features as an intervention feature and adjusting its value, and generating the embedding vector of the nodes according to the adjusted values of the plurality of features to be used to determine the influence of the adjustment of the value of the intervention feature on the reference value of the to-be-predicted indicator; Predict the value of the to-be-predicted indicator of the business object according to the embedding vector; Perform corresponding business processing according to the prediction result, specifically including: determining the gain value brought by the adjustment of the value of the intervention feature to the reference value of the to-be-predicted indicator according to the prediction result and the reference value of the to-be-predicted indicator, and determining whether to perform corresponding adjustment on the business corresponding to the business object according to the gain value; allocating corresponding resources, and such resources are at least one of the following: computing resources of the server, cloud storage resources, front-end resources exposed to users, traffic resources guided for merchants; When the business object is a commodity, the features related to the business object include at least one of the following: raw materials, quality, appearance, price, selling discount, selling area, inventory; the to-be-predicted indicators include indicators representing sales conditions or customer conversion conditions; The business object includes a commodity, the indicator representing the sales condition includes the sales volume of the commodity, and the indicator representing the customer conversion condition includes the number of users who click on the commodity.

Images