Pre-estimation model construction method, pre-estimation model construction device and electronic equipment
By using a causal coefficient analysis technology prediction model in customer attribute estimates, combined with the causal model and basic estimate model, the problem of large deviations in manual estimates in the existing technology is solved, and more accurate user attribute estimates and abnormal traffic capture is achieved.
Patent Information
- Application Number
- CN202411745571.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-29
- Publication Date
- 2025-03-04
AI Technical Summary
In the prior art, customer attribute estimates rely on manual labor, resulting in large prediction deviations, especially during abnormal traffic periods such as holiday promotions, which cannot capture abnormal fluctuations in time, resulting in poor prediction results.
The prediction model construction method based on causal coefficient analysis technology is adopted, and the causal coefficients of each factor affecting the target user attributes are determined through the combination of the causal model and the basic prediction model, and the causal coefficients of each factor affecting the target user attributes are obtained, and the final prediction target user attributes are obtained.
It improves the accuracy of user attribute estimates, can timely capture the impact of abnormal traffic such as holiday promotions, and improves the prediction effect of user number and order quantity.
Smart Images
Figure CN119558739A_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present disclosure relate to the field of customer marketing technology, and in particular to an estimation model construction method, an estimation model construction device, and an electronic device. Background Art
[0002] At present, the behaviors of customers in the logistics industry during the contract period are diverse and the data accumulated in the customer business data are complex. How to accurately estimate some attribute data of customers (such as life cycle value) is an urgent problem that needs to be solved.
[0003] However, the inventors found that existing customer attribute estimation is usually done manually. Generally, a rough estimation is made based on historical order volume, revenue, customer complaint rate and other indicators, and such estimation often has large deviations. In addition, customer attributes are often affected by factors such as holiday promotions, which may cause abnormal fluctuations in operating data in the short term. Manual estimation usually fails to capture abnormal traffic such as holiday promotions in a timely manner, and predictions often lag, which greatly reduces the prediction effect.
[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention
[0005] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.
[0006] Some embodiments of the present disclosure propose an estimation model construction method, an estimation model construction device, an electronic device, a computer-readable medium and a computer program product to solve one or more of the technical problems mentioned in the above background technology section.
[0007] In a first aspect, some embodiments of the present disclosure provide a method for constructing an estimation model, comprising: determining the causal coefficients of factors affecting the attributes of a target user based on sample data and a causal model in an estimation model, wherein the sample data includes feature data sequences under multiple data dimensions, and label data representing the attributes of sample target users under different times and different data dimensions; selecting the causal coefficient of a target factor from the determined causal coefficients, wherein the target factor is a factor that matches a set time and a set dimension; obtaining a preliminary estimation result based on the sample data and a basic estimation model in the estimation model, wherein the basic estimation model is used to estimate the attributes of the target user at a set time; obtaining the estimated target user attributes output by the estimation model based on the causal coefficient of the target factor and the preliminary estimation result.
[0008] In some embodiments, based on the sample data and the causal model in the estimation model, the causal coefficients of various factors affecting the target user attributes are determined, including: selecting an adapted causal model according to the data characteristics of different feature data sequences in the sample data; obtaining the causal effects of various factors affecting the target user attributes based on the sample data and the selected causal model; and transforming the causal effects of various factors to obtain the causal coefficients of various factors.
[0009] In some embodiments, based on the sample data and the selected causal model, the causal effects of various factors affecting the target user attributes are obtained, including: performing a logarithmic transformation on the sample data, and inputting the transformed sample data into the selected causal model to obtain the causal effects of various factors affecting the target user attributes.
[0010] In some embodiments, the causal effect of each factor is transformed to obtain the causal coefficient of each factor, including: performing exponential transformation on the causal effect of each factor to obtain the transformed causal coefficient; evaluating the transformed causal coefficient, and determining the evaluated causal coefficient as the causal coefficient of each factor.
[0011] In some embodiments, the method also includes: extracting relevant feature data from the acquired original data, wherein the relevant feature data includes at least one of the following: target factor data, confusion factor data, and / or other dimensional feature data; combining the extracted relevant feature data according to a set time length to obtain a feature data sequence, and marking the target user attributes within the set time length to generate sample data.
[0012] In some embodiments, relevant feature data is extracted from the acquired original data, including: performing word segmentation processing on the unstructured data in the acquired original data to obtain word data; classifying the word data using a tree structure, and using the information of each node in the tree structure as a data dimension to obtain feature data under different data dimensions.
[0013] In some embodiments, the method further includes: tuning the estimation model according to the estimated target user attributes and the corresponding label data in the sample data to continue training the tuned model; in response to determining that the estimation model training is completed, evaluating the estimation model.
[0014] In some embodiments, tuning the estimation model includes: using a grid search method to determine the type of causal model to tune the model; and / or evaluating feature data in the estimation model according to Occam's razor principle to tune the feature data of the model.
[0015] In some embodiments, the method further includes: in response to receiving an estimation request, obtaining an estimation result of the estimation request through an estimation model, wherein the estimation request is used to indicate an estimated target user attribute; and storing the estimation result in a database for query.
[0016] In a second aspect, some embodiments of the present disclosure provide an estimation model construction device, comprising: a causal coefficient determination unit, configured to determine the causal coefficients of factors affecting the target user attributes based on sample data and the causal model in the estimation model, wherein the sample data includes feature data sequences under multiple data dimensions, and label data representing the sample target user attributes under different times and different data dimensions; a target coefficient selection unit, configured to select the causal coefficient of the target factor from the determined causal coefficients, wherein the target factor is a factor matching the set time and the set dimension; a preliminary estimation unit, configured to obtain a preliminary estimation result based on the sample data and the basic estimation model in the estimation model, wherein the basic estimation model is used to estimate the target user attributes at the set time; an estimated attribute output unit, configured to obtain the estimated target user attributes output by the estimation model based on the causal coefficient of the target factor and the preliminary estimation result.
[0017] In some embodiments, the causal coefficient determination unit is further configured to select an adaptive causal model based on the data characteristics of different feature data sequences in the sample data; obtain the causal effects of each factor affecting the target user attributes based on the sample data and the selected causal model; and transform the causal effects of each factor to obtain the causal coefficient of each factor.
[0018] In some embodiments, the causal coefficient determination unit is further configured to perform a logarithmic transformation on the sample data, and input the transformed sample data into the selected causal model to obtain the causal effect of each factor affecting the target user attribute.
[0019] In some embodiments, the causal coefficient determination unit is further configured to perform an exponential transformation on the causal effect of each factor to obtain a transformed causal coefficient; evaluate the transformed causal coefficient, and determine the evaluated causal coefficient as the causal coefficient of each factor.
[0020] In some embodiments, the estimation model building device also includes a sample data generation unit, which is configured to extract relevant feature data from the acquired original data, wherein the relevant feature data includes at least one of the following: target factor data, confusion factor data, and / or other dimensional feature data; perform feature combination on the extracted relevant feature data according to a set time length to obtain a feature data sequence, and mark the target user attributes within the set time length to generate sample data.
[0021] In some embodiments, the sample data generation unit is further configured to perform word segmentation processing on the unstructured data in the acquired original data to obtain word data; classify the word data using a tree structure, and use the node information in the tree structure as a data dimension to obtain feature data under different data dimensions.
[0022] In some embodiments, the estimation model building device also includes a model tuning unit, which is configured to tune the estimation model according to the estimated target user attributes and the corresponding label data in the sample data to continue training the tuned model; in response to determining that the estimation model training is completed, the estimation model is evaluated.
[0023] In some embodiments, the model tuning unit is further configured to use a grid search method to determine the type of causal model to tune the model; and / or evaluate the feature data in the estimation model according to the Occam's razor principle to tune the feature data of the model.
[0024] In some embodiments, the estimation model building device also includes a storage unit, which is configured to obtain an estimation result of the estimation request through the estimation model in response to receiving the estimation request, wherein the estimation request is used to indicate the estimation of the target user attributes; and store the estimation result in a database for query.
[0025] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by one or more processors, the one or more processors implement the estimation model building method described in any implementation method in the first aspect above.
[0026] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the estimation model construction method described in any implementation manner in the above-mentioned first aspect is implemented.
[0027] In a fifth aspect, some embodiments of the present disclosure provide a computer program product, including a computer program, which, when executed by a processor, implements the estimation model construction method described in any implementation method in the above-mentioned first aspect.
[0028] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: the estimation model construction method of some embodiments of the present disclosure can improve the accuracy of the model estimation results. Specifically, the manual estimation of some attribute data of existing customers (such as life cycle value) is often very dependent on historical experience, and abnormal traffic such as holiday promotions is not captured in time. For example, during major promotions such as Labor Day, Spring Festival, and Double Eleven, when abnormal fluctuations in operating data occur in the short term, predictions often lag. It is impossible to give a more accurate user attribute estimate within the abnormal traffic cycle window such as holiday promotions, which affects the increase in the number of customers and the number of orders.
[0029] In addition, manual evaluation makes it difficult to clarify the impact of different factors on customers' future attributes. The samples corresponding to these factors are often sparse, and the factors interact with each other, making it difficult to estimate the impact of the factors on the user attributes (lifetime value) of existing logistics customers. In addition, customer behavior is complex and mutually influential in business data, and the costs of performing related actions on existing customers (such as renewal, promotional incentives, etc.) vary. Therefore, customer lifecycle value estimation can guide staff to choose the optimal (profit-maximizing) intervention action to avoid waste of resources caused by rising operating costs.
[0030] Based on this, the estimation model construction method of some embodiments of the present disclosure can construct an estimation model based on causal coefficient analysis technology. The estimation model of the present disclosure includes a causal model and a basic estimation model. By analyzing the sample data through the causal model, the causal coefficient of each factor affecting the user value can be determined. That is, different influencing factors can be modeled and analyzed based on the causal model. And the causal coefficient of the appropriate influencing factor can be applied to the preliminary estimation result of the basic estimation model in a timely manner. Thereby, the final estimated target user attributes (such as life cycle value) after adjustment of the major influencing factors are obtained. In this way, the problem of poor estimation of time series variables caused by the untimely capture of abnormal traffic such as holiday promotions can be solved. This will improve the accuracy of user attribute estimation, which is conducive to increasing the number of users and the number of orders. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.
[0032] Figure 1 is a flowchart of some embodiments of the estimation model building method disclosed in the present invention;
[0033] Figure 2 are flow charts of other embodiments of the estimation model building method disclosed herein;
[0034] Figure 3A It is a schematic diagram of the data processing flow for causal coefficient estimation;
[0035] Figure 3B is an architectural diagram of an exemplary system in which some embodiments of the present disclosure may be applied;
[0036] Figure 4 It is a schematic diagram of the structure of some embodiments of the estimation model building device disclosed in the present invention;
[0037] Figure 5 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION
[0038] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.
[0039] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.
[0040] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.
[0041] In addition, the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".
[0042] Figure 1 The process 100 of some embodiments of the method for building an estimation model according to the present disclosure is shown. The method may include the following steps:
[0043] Step 101 , based on sample data and the causal model in the estimation model, determine the causal coefficient of each factor affecting the target user's attributes.
[0044] In some embodiments, the execution subject of the estimation model construction method disclosed in the present invention (such as a model construction and training server) can be connected to other electronic devices (such as user terminals, databases, etc.) through wired or wireless connections. Here, the staff can send the sample data used to train the model directly to the execution subject through the user terminal, so that the execution subject can build and train the estimation model. For another example, the staff can also store the sample data in the database in advance. In this case, after receiving the relevant instructions sent by the staff, the execution subject can obtain the sample data from the database to build and train the estimation model.
[0045] Here, the sample data may include feature data sequences under multiple data dimensions, as well as label data that characterize the sample target user attributes under different times and different data dimensions. Among them, the data dimensions can be set according to actual needs, for example, it can include relevant data of abnormal traffic such as holiday promotions, such as specified business data such as order volume and return volume; another example is a specified time period (granularity), such as day, week, month, quarter, year, etc.; another example is related actions performed by staff on users, such as item promotion, maintaining or improving user levels, etc. The above-mentioned target user attributes can also be set according to actual needs, such as customer lifecycle value (CLV). Among them, CLV, also known as LTV (life time value), is an effective prediction of the future profits of customers (users), which is used to measure how valuable a customer (user) is to the company over a period of time.
[0046] In some embodiments, Figure 2 As shown, the estimation model of the present invention may include a causal model for causal coefficient estimation. Among them, the causal model is different from the traditional machine learning model, and what is explored and pursued is the causal mechanism behind the physical object. The causal model here is set according to actual needs, such as linear causal model (LinearDML (Double Machine Learning)), tree series causal model (CausalForestDML), etc. Here, the execution subject can determine the causal coefficients of various factors affecting the attributes of the target user based on the sample data and the causal model in the estimation model.
[0047] As an example, first, the execution subject can select an adaptive causal model according to the data characteristics of different feature data sequences in the sample data. That is, a suitable causal model can be selected for different data features. For example, a linear causal model can be selected for linear data, and a tree series causal model can be selected for hierarchical data. Then, based on the sample data and the selected causal model, the causal effect of each factor affecting the target user's attributes can be obtained. For example, the execution subject can input the sample data into the selected causal model, analyze the sample data through the causal model, and obtain the causal effect of each factor affecting the (sample) target user's attributes.
[0048] It should be noted that the causal model is very advantageous in solving the problem that the samples corresponding to abnormal factors are sparse, and the factors interact with each other, which makes it difficult to estimate the true impact of factors on the customer life cycle value (referred to as customer value) in the growth of logistics stock customers. The impact of confounding factors is mainly eliminated by establishing a model between confounding factors and target factors. However, the causal model is an additive model, which is used to analyze how much incremental causal effect the treatment factor can bring. And customer value estimation is a multiplicative model. Therefore, when analyzing the impact of abnormal factors on customer value, what is needed is the causal coefficient rather than the causal effect. In order to solve this contradiction, the causal model needs to be adapted and transformed to obtain the causal coefficient. Therefore, the execution subject can transform the causal effect of each factor to obtain the causal coefficient of each factor. As an example, the execution subject can perform an exponential transformation on the causal effect of each factor to obtain the transformed causal coefficient.
[0049] Optionally, the execution subject can pass Figure 3A The processing flow shown in Figure 1 is used to estimate the causal coefficient. Figure 3AAs shown, it can include four parts: logarithmic transformation, causal model selection, exponential transformation, and causal coefficient evaluation. First, the target user attributes are logarithmically transformed to adapt to the design of traditional causal model additive modeling. That is, the execution subject can logarithmically transform the sample data. Next, the appropriate causal model is selected according to different data characteristics. Then, the transformed sample data can be input into the selected causal model to obtain the causal effect of each factor affecting the target user attribute. After that, the causal effect of each factor is exponentially transformed. That is, the causal effect estimated by the causal model is exponentially transformed, that is, the causal coefficient is converted from the causal effect output in the previous step. Through mathematical transformation, it can be seen that the prediction result of the causal model is transformed from additive effect to multiplicative effect after exponential transformation. The transformed data is the causal coefficient we want. Finally, the transformed causal coefficient can be evaluated, and the causal coefficient that has been evaluated can be determined as the causal coefficient of each factor. Here, we mainly focus on whether the p-value of the causal effect estimate is significant. If it is significant, the estimated causal coefficient has high reliability. In addition, we can also observe the range of the causal coefficient value to see whether it deviates far from 1.0.
[0050] Step 102: Select the causal coefficient of the target factor from the determined causal coefficients.
[0051] In some embodiments, when the causal coefficients of each factor are obtained in step 101, the execution entity can select the causal coefficient of the target factor from the determined causal coefficients. Among them, the target factor is a factor that matches the set time and the set dimension. The set time and the set dimension here are usually the time and dimension of the target user attribute to be predicted. It can be understood that in the model training process, the set time and the set dimension are usually the time and data dimension in the sample data. In other words, the execution entity can select the factor coefficient that matches the prediction day and granularity, such as the causal coefficient of the big promotion activity time for a big promotion activity.
[0052] Step 103, obtaining a preliminary estimation result according to the sample data and the basic estimation model in the estimation model.
[0053] In some embodiments, Figure 2 As shown, the estimation model of the present disclosure may also include a basic estimation model. Among them, the basic estimation model can be used to estimate the attributes of the target user at a set time. The basic estimation model here can also be set according to actual needs, and can adopt the user attribute estimation model commonly used in related technologies, such as the Transformer deep learning model. Here, the execution subject can also input sample data into the basic estimation model to obtain the preliminary estimation results of the target user attributes under the above-mentioned set time and set dimension.
[0054] It should be noted that, in order to improve the efficiency of model construction and training, the basic estimation model here can adopt the user attribute estimation model that has been trained in the relevant technology. Alternatively, while improving efficiency and meeting the usage scenario of this embodiment, the basic estimation model here can also adopt a pre-trained model.
[0055] Step 104, obtaining the estimated target user attributes output by the estimation model according to the causal coefficient of the target factor and the preliminary estimation result.
[0056] In some embodiments, the execution entity may obtain the estimated target user attributes output by the estimation model based on the causal coefficient of the target factor in step 102 and the preliminary estimation result in step 103. As an example, the causal coefficient of the target factor may be multiplied by the preliminary estimation result, and the result obtained may be output as the final estimation data of the estimation model. Optionally, the execution entity may also perform a weighted summation on the result of the multiplication and the preliminary estimation result, thereby outputting the calculation result as the estimated target user attributes.
[0057] It can be seen from the above description that the estimation model construction method of some embodiments of the present disclosure can construct an estimation model based on causal coefficient analysis technology. The estimation model of the present disclosure includes a causal model and a basic estimation model. By analyzing the sample data through the causal model, the causal coefficients of various factors affecting the target user attributes can be determined. That is, different influencing factors can be modeled and analyzed based on the causal model. And the causal coefficients of appropriate influencing factors can be applied to the preliminary estimation results of the basic estimation model in a timely manner. Thereby, the final estimated target user attributes after adjustment of the major influencing factors are obtained. In this way, the problem of poor estimation of time series variables caused by the untimely capture of abnormal traffic such as holiday promotions can be solved. This will improve the accuracy of user attribute estimation, which is conducive to increasing the number of users and orders.
[0058] Continue to see Figure 2 , which shows a flowchart of some other embodiments of the estimation model construction method disclosed in the present invention. The method may include the following steps:
[0059] Step A: Feature Engineering for Causal Coefficient Estimation
[0060] In some embodiments, first, relevant feature data can be extracted from the acquired raw data. Among them, the relevant feature data may include at least one of the following: target factor data, confounding factor data, and / or other dimensional feature data, etc. As an example, the Treatment variable (i.e., target factor) can be selected first, such as a big promotion day. Then, confounding variables (i.e., confounding factors) can be selected, such as holidays, midweek weekends, abnormal weather, etc. And modeling can also be performed by selecting other multi-dimensional features as covariates. Among them, the feature dimensions may include: merchant portrait information (location, tertiary industry, main category, contract product, etc.), logistics waybill information (commodity SKU, category, route, etc.), POP (point of sale advertising) merchant mall information (browse, click, consult, add to cart, favorite, etc.), statistical feature information (main route, single quantity / package quantity extreme value, single quantity / package quantity mean value, etc.).
[0061] It is understandable that structured data can usually be directly extracted from the original data. However, in order to increase the diversity of feature data and improve the accuracy of model prediction results, the unstructured data in the original data can also be processed to extract relevant feature data. Specifically, for the unstructured data in the original data, word segmentation can be performed first to obtain word data. Then, a tree structure can be used to classify the word data. After that, the information of each node in the tree structure can be used as a data dimension to obtain feature data under different data dimensions.
[0062] In other words, for unstructured text information such as staff visit information, customer complaint information, and operational quality monitoring, we can segment the middle-level features of different levels through Chinese word segmentation and tree structure classification. It should be noted that we will classify the above text information (customer feedback issues). The information category is usually a system, which will have first-level, second-level, and third-level classifications to form a tree structure. For example, the first-level "platform problem" can include the second-level "operation problem", and the second-level "operation problem" can include the third-level "lost item" problem.
[0063] Next, when the relevant feature data is obtained, the extracted relevant feature data can be combined according to the set time length to obtain a feature data sequence, and the target user attributes within the set time length can be marked to generate sample data.
[0064] As an example, all features can be combined into feature slices by time, and the time difference feature between the historical time and the current time can be generated. In other words, a set of feature slices is often within a certain time period, such as a day. The generated time difference feature is used as the time feature of the feature slice, indicating the degree of discontinuity. In addition, since there is a lot of feature data, the pipeline mechanism can also be used here to serialize the processing mechanism of all feature data and finally summarize it into a total feature attribute.
[0065] In other words, we can collect the merchant's historical order information, waybill information, product details, merchant portraits, staff's online and offline visit feedback, customer complaints, claims, fulfillment data, and other features that have the greatest impact on the target user's attributes, thereby constructing the basic sample data of the <basic query combination, timestamp> combination, and input it as the basic data into the estimation model based on causal coefficient analysis technology, which can realize the estimation of target user attributes at different time granularities.
[0066] Step B: Estimation of causal coefficients
[0067] For this step, reference may be made to the relevant description of step 101 in the above embodiment, which will not be repeated here.
[0068] Step C: Establishment of prediction model based on causal coefficient analysis technique
[0069] For this step, reference may be made to the relevant description of steps 102-104 in the above embodiment, which will not be repeated here.
[0070] Step D: Estimation model tuning based on causal coefficient analysis techniques
[0071] In some embodiments, the estimation model can be tuned according to the estimated target user attributes and the corresponding label data in the sample data to continue training the tuned model until the model training is completed. Here, the method of model tuning is also not limited. As an example, the execution entity can determine the loss function value of the estimation model based on the estimated target user attributes and the corresponding sample target user attributes. Then, the model parameters of the estimation model are adjusted according to the loss function value. It can be understood that the causal model and the basic estimation model in the model can be adjusted at the same time, or only the causal model in the model can be adjusted.
[0072] Optionally, a grid search method can also be used to determine the type of causal model in order to tune the model. This tuning method is mainly for the selection of models such as linear causal models, causal forests, counterfactual predictions (deepIV), etc. We can use the grid search method to obtain the optimal model effect. And / or, the feature data in the estimation model can also be evaluated according to the Occam's razor principle to tune the feature data of the model. This tuning method mainly analyzes the feature importance of the model and the correlation between features. According to the Occam's razor principle, the feature dimension is reduced as much as possible to avoid overfitting.
[0073] Step E: Estimation Model Evaluation Based on Causal Coefficient Analysis Techniques
[0074] In some embodiments, when the estimation model training is completed, the estimation model can also be evaluated. The evaluation method here is also not limited. For example, 0.2 parts of the sample data are used to evaluate the model. The evaluation indicators can mainly use indicators of the regression model, such as WMAPE (Weighted Mean Absolute Percentage Error, an indicator for measuring the accuracy of the prediction model), WAE (Weighted Average of Errors, a measure of the error between the model prediction results and the actual observation results), and MAPE (Mean Absolute Percentage Error).
[0075] Step F: Prediction and use of the estimated model based on causal coefficient analysis techniques
[0076] In some embodiments, Figure 3B As shown, after the estimation model evaluation is completed, the estimation model can be used to estimate the target user attributes (such as customer value) online. That is, when an estimation request is received, the estimation result of the estimation request can be obtained through the estimation model. Among them, the estimation request can be used to indicate the estimated target user attributes, such as including the estimated time and data dimension.
[0077] Furthermore, the estimated results can be stored in the database for query. In other words, the real-time business information of merchants can be collected through the online database, mainly the waybill log. The processed real-time business information is then spliced with offline features, user intervention actions, etc., and input into the model as a first-level feature combination. After real-time reasoning, the final estimated results can be obtained. The results are then written to the database (such as es, search server) for data backtracking and model optimization, which is also convenient for instant query.
[0078] In this case, the query conditions input by the front end can be combined according to the query needs of the staff, and the enumeration values can be matched in the database according to the key value. That is, the original value of the daily estimated data can be read from the database. Finally, all the query results are clustered to obtain the final actual prediction results. Subsequently, real-time post-processing calculations and risk control limit calculations can be performed on the server to obtain the estimated final customer LTV estimate and display the specific query results on the front end. In this way, the staff can not only accurately complete the budget planning of the new project based on the query results, but also refine the follow-up reasons based on the difference between the customer value estimated by the model and the manual inventory estimate, and improve the user's insight ability.
[0079] In actual application scenarios, such as in the supply chain order increase and potential mining project, compared with the model without superimposed causal coefficients, the estimation accuracy of this algorithm during the promotion period increased from 74.1% to 87.5%. Online, through the PSM (Propensity Score Matching) method, the differences between customers are eliminated, the customer data is diverted, and the customers are divided into an experimental group and a control group. A total of 2,000 customers in each group were subjected to AB experiments. As shown in the following table, the P value shows that there is no significant difference in the advertising code scanning behavior between the two groups of customers.
[0080]
[0081] The experimental group used the prediction model based on causal coefficients to predict the big promotion, and the control group used the model without superimposed causal coefficients to predict the big promotion. As shown in the following table, the average order volume of 2,000 customers in the experimental group was 10,206.7565; the average order volume of the control group was 7,558.9416. The experimental group significantly increased by 35.03% compared with the control group (p value 0.0040).
[0082]
[0083] The estimation model construction method of the disclosed embodiment is mainly aimed at estimating the customer life cycle value in the growth of logistics stock customers. In order to enable staff to accurately estimate the customer value after performing intervention actions, balance the marketing budget (specify marketing time period and intervention cost), recover revenue under abnormal traffic conditions such as holiday promotions, reduce negative revenue brought by abnormal customers and other multiple indicators, causal coefficient analysis technology can be used to realize the estimation of logistics stock customer value and provide accurate estimation capabilities. This can assist staff in specifying accurate and highly available marketing intervention actions to maximize user growth.
[0084] Further references Figure 4 , as a response to the above Figure 1 To implement the method shown in FIG. 3 , the present disclosure provides some embodiments of an estimation model building device. These device embodiments are Figure 1 The device can be applied to various electronic devices.
[0085] like Figure 4 As shown, an estimation model construction device 400 of some embodiments may include: a causal coefficient determination unit 401, configured to determine the causal coefficients of various factors affecting the target user attributes based on sample data and the causal model in the estimation model, wherein the sample data includes feature data sequences under multiple data dimensions, and label data representing the sample target user attributes under different times and different data dimensions; a target coefficient selection unit 402, configured to select the causal coefficient of the target factor from the determined causal coefficients, wherein the target factor is a factor that matches the set time and the set dimension; a preliminary estimation unit 403, configured to obtain a preliminary estimation result based on the sample data and the basic estimation model in the estimation model, wherein the basic estimation model is used to estimate the target user attributes at the set time; an estimated attribute output unit 404, configured to obtain the estimated target user attributes output by the estimation model based on the causal coefficient of the target factor and the preliminary estimation result.
[0086] In some embodiments, the causal coefficient determination unit 401 can be further configured to select an adaptive causal model based on the data characteristics of different feature data sequences in the sample data; obtain the causal effects of various factors affecting the target user attributes based on the sample data and the selected causal model; and transform the causal effects of each factor to obtain the causal coefficient of each factor.
[0087] In some embodiments, the causal coefficient determination unit 401 may be further configured to perform a logarithmic transformation on the sample data, and input the transformed sample data into a selected causal model to obtain the causal effects of various factors affecting the target user attributes.
[0088] In some embodiments, the causal coefficient determination unit 401 can be further configured to perform an exponential transformation on the causal effect of each factor to obtain a transformed causal coefficient; evaluate the transformed causal coefficient, and determine the evaluated causal coefficient as the causal coefficient of each factor.
[0089] In some embodiments, the estimation model building device 400 may also include a sample data generation unit (not shown in the figure), which is configured to extract relevant feature data from the acquired original data, wherein the relevant feature data includes at least one of the following: target factor data, confusion factor data, and / or other dimensional feature data; perform feature combination on the extracted relevant feature data according to a set time length to obtain a feature data sequence, and mark the target user attributes within the set time length to generate sample data.
[0090] In some embodiments, the sample data generation unit can be further configured to perform word segmentation processing on the unstructured data in the acquired original data to obtain word data; classify the word data using a tree structure, and use the node information in the tree structure as a data dimension to obtain feature data under different data dimensions.
[0091] In some embodiments, the estimation model building device 400 may also include a model tuning unit (not shown in the figure), which is configured to tune the estimation model according to the estimated target user attributes and the corresponding label data in the sample data to continue training the tuned model; in response to determining that the estimation model training is completed, the estimation model is evaluated.
[0092] In some embodiments, the model tuning unit can be further configured to use a grid search method to determine the type of causal model to tune the model; and / or evaluate the feature data in the estimation model according to the Occam's razor principle to tune the feature data of the model.
[0093] In some embodiments, the estimation model building device 400 may also include a storage unit (not shown in the figure), which is configured to obtain an estimation result of the estimation request through an estimation model in response to receiving an estimation request, wherein the estimation request is used to indicate the estimation of target user attributes; and store the estimation result in a database for query.
[0094] It can be understood that the units recorded in the estimation model building device 400 are similar to those in the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the estimation model building device 400 and the units included therein, and will not be described in detail here.
[0095] Reference below Figure 5 , which shows a structural schematic diagram of an electronic device 500 suitable for implementing some embodiments of the present disclosure. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.
[0096] like Figure 5As shown, the electronic device 500 may include a processing device 501 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 502 or a program loaded from a storage device 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.
[0097] Typically, the following devices may be connected to the I / O interface 505: input devices 506 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 507 including, for example, a speaker, a vibrator, etc.; storage devices 508 including, for example, a disk, a hard disk, a memory card, etc.; and communication devices 509. The communication devices 509 may allow the electronic device 500 to communicate with other devices wirelessly or by wire to exchange data. Figure 5 The electronic device 500 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 5 Each block shown in the figure may represent one device, or may represent multiple devices as required.
[0098] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.
[0099] It should be noted that the computer-readable medium recorded in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.
[0100] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (Hyper Text Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.
[0101] The computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: determines the causal coefficients of various factors affecting the target user attributes based on the sample data and the causal model in the estimation model, wherein the sample data includes feature data sequences under multiple data dimensions, and label data representing the sample target user attributes under different times and different data dimensions; selects the causal coefficient of the target factor from the determined causal coefficients, wherein the target factor is a factor that matches the set time and the set dimension; obtains a preliminary estimation result based on the sample data and the basic estimation model in the estimation model, wherein the basic estimation model is used to estimate the target user attributes at the set time; obtains the estimated target user attributes output by the estimation model based on the causal coefficient of the target factor and the preliminary estimation result.
[0102] In addition, computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).
[0103] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.
[0104] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be arranged in a processor, for example, may be described as: a processor including a causal coefficient determination unit, a target coefficient selection unit, a preliminary estimation unit, and an estimated attribute output unit. Among them, the names of these units do not constitute a limitation on the units themselves in certain cases, for example, the causal coefficient determination unit may also be described as a "unit for determining the causal coefficients of various factors affecting the target user attributes".
[0105] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.
[0106] Some embodiments of the present disclosure also provide a computer program product, including a computer program, which, when executed by a processor, implements any of the above-mentioned estimation model construction methods.
[0107] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.
Claims
1. A method for constructing an estimation model, comprising: Determine the causal coefficients of various factors affecting the target user attributes based on the sample data and the causal model in the estimation model, wherein the sample data includes feature data sequences under multiple data dimensions and label data representing the sample target user attributes under different times and different data dimensions; Selecting a causal coefficient of a target factor from the determined causal coefficients, wherein the target factor is a factor that matches a set time and a set dimension; Obtaining a preliminary estimation result according to the sample data and a basic estimation model in the estimation model, wherein the basic estimation model is used to estimate the attributes of the target user at the set time; According to the causal coefficient of the target factor and the preliminary estimation result, the estimated target user attribute output by the estimation model is obtained.
2. The method for constructing an estimation model according to claim 1, wherein: The causal coefficients of various factors affecting the target user attributes are determined based on the sample data and the causal model in the estimation model, including: Select an appropriate causal model based on the data characteristics of different feature data sequences in the sample data; Based on the sample data and the selected causal model, the causal effect of each factor affecting the target user attribute is obtained; The causal effect of each factor is transformed to obtain the causal coefficient of each factor.
3. The method for constructing an estimation model according to claim 2, wherein: The causal effects of various factors affecting the target user attributes are obtained based on the sample data and the selected causal model, including: The sample data is logarithmically transformed, and the transformed sample data is input into a selected causal model to obtain the causal effect of each factor affecting the target user's attributes.
4. The method for constructing an estimation model according to claim 2, wherein: The causal effect of each factor is transformed to obtain the causal coefficient of each factor, including: The causal effect of each factor is exponentially transformed to obtain the transformed causal coefficient; The transformed causal coefficients are evaluated, and the evaluated causal coefficients are determined as the causal coefficients of each factor.
5. The method for constructing an estimation model according to claim 1, wherein: The method further comprises: Extracting relevant feature data from the acquired original data, wherein the relevant feature data includes at least one of the following: target factor data, confounding factor data, and / or other dimensional feature data; The extracted relevant feature data are combined according to the set time length to obtain a feature data sequence, and the target user attributes within the set time length are marked to generate sample data.
6. The method for constructing an estimation model according to claim 5, wherein: The extracting relevant feature data from the acquired raw data includes: Perform word segmentation processing on the unstructured data in the acquired original data to obtain word data; The word data are classified by using a tree structure, and each node information in the tree structure is used as a data dimension to obtain feature data under different data dimensions.
7. The method for constructing an estimation model according to claim 1, wherein: The method further comprises: According to the estimated target user attributes and the corresponding label data in the sample data, the estimation model is optimized to continue training the optimized model; In response to determining that the estimation model training is complete, the estimation model is evaluated.
8. The method for constructing an estimation model according to claim 7, wherein: The step of tuning the estimation model includes: Using a grid search method to determine the type of the causal model to tune the model; and / or The characteristic data in the estimation model is evaluated according to the principle of Occam's razor to optimize the characteristic data of the model.
9. The method for constructing an estimation model according to any one of claims 1 to 8, wherein: The method further comprises: In response to receiving the estimation request, obtaining an estimation result of the estimation request through the estimation model, wherein the estimation request is used to indicate an estimated target user attribute; The estimated results are stored in a database for query.
10. A prediction model building device, comprising: A causal coefficient determination unit is configured to determine the causal coefficients of various factors affecting the target user attributes based on the sample data and the causal model in the estimation model, wherein the sample data includes feature data sequences under multiple data dimensions and label data representing the sample target user attributes under different times and different data dimensions; A target coefficient selection unit is configured to select a causal coefficient of a target factor from the determined causal coefficients, wherein the target factor is a factor that matches a set time and a set dimension; A preliminary estimation unit, configured to obtain a preliminary estimation result according to the sample data and a basic estimation model in the estimation model, wherein the basic estimation model is used to estimate the attributes of the target user at the set time; The estimated attribute output unit is configured to obtain the estimated target user attribute output by the estimation model according to the causal coefficient of the target factor and the preliminary estimation result.
11. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the estimation model building method as described in any one of claims 1-9.
12. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the estimation model construction method as described in any one of claims 1 to 9 is implemented.
13. A computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the estimation model construction method as described in any one of claims 1 to 9.