Information statistics management platform based on artificial intelligence

Abstract

The invention discloses an artificial intelligence-based information statistics management platform, which relates to the technical field of information statistics management, and comprises an artificial intelligence unit, a data set sharing unit, an expandability unit, a solution division unit and an execution unit, the problems that an existing platform is not clear in data integration and use scene, low in expandability and resource utilization rate, insufficient in model interpretation and the like exist, analysis and processing are conducted through a data set common unit, an expandability unit and a division unit of the platform, the real-time performance and accuracy of data are guaranteed, therefore, faster and more accurate service decisions are supported, and the service quality of the platform is improved. The overall data management efficiency and quality are improved; the optimal configuration and cost control of resources are realized, and the real-time processing requirements of high-concurrency and high-load scenes are met; the transparency of model decision is improved, clear model explanation is provided for users, and decision optimization is supported.

Description

Technical Field

[0001] The present invention relates to the technical field of information statistics management, and in particular to an information statistics management platform based on artificial intelligence. Background Art

[0002] In the existing information statistics management platforms, the application of artificial intelligence (AI) technology has gradually become a trend, especially in the fields of data processing, predictive analysis, and automated statistics. These platforms can efficiently process large amounts of data, explore deep connections between data, and improve the intelligence level of information statistics and management.

[0003] However, existing platforms often lack consistency in integrating multi-source data, resulting in inefficiency in cross-departmental data sharing and use, and easily creating data silos; they often face problems such as low computing resource utilization and low task processing efficiency when processing big data, and they are insufficient in providing explanations for model results, making it difficult to help users understand the role of features in predictions. Summary of the invention

[0004] The purpose of the present invention is to solve the above-mentioned problems and to propose an information statistics management platform based on artificial intelligence.

[0005] The object of the present invention can be achieved by the following technical solutions: an information statistics management platform based on artificial intelligence, comprising a data set sharing unit, an extensibility unit and a decomposition unit;

[0006] The data set is used to build a data center, where artificial intelligence and the business department perform data identification to record the type, frequency and purpose of the data obtained by each department; then data is collected by establishing an automated data collection pipeline; the file import tool is used to import the data into the integration platform for format conversion, and the data is converted into a unified format; data from multiple different data sources are merged, and customer information is combined with corresponding order data to form a unified customer view; scheduling tools are used to regularly obtain the latest data from each data source;

[0007] The scalability unit performs scalability processing on the platform through artificial intelligence, specifically:

[0008] A certain number of different extended processing methods are created by artificial intelligence, including method 1, method 2, method 3, ..., method n, and different methods or combinations of methods are selected according to different processing scenarios; the number of computing tasks and usage scenarios are analyzed to obtain the number of tasks, the task timeliness, and the scenario timeliness values, and the corresponding processing method is selected or combined optimization is performed according to the calculated number of tasks, the task timeliness, and the scenario timeliness values;

[0009] The decomposition unit is used to perform interpretable analysis on the statistical information, specifically:

[0010] The weight of each feature is obtained by interpreting the local data;

[0011] The marginal contribution value of each feature in all feature combinations is calculated to get the average marginal contribution.

[0012] As a preferred embodiment of the present invention, it also includes an artificial intelligence unit and an execution unit;

[0013] The artificial intelligence unit intelligently identifies the needs of business departments by building a data middle platform. It also dynamically optimizes resource allocation through load balancing and resource monitoring. The artificial intelligence unit also participates in generating local explanatory models to quantify and visualize the impact of each feature.

[0014] The execution unit is responsible for automatically scheduling data collection, allocating processing tasks, dynamically adjusting resource utilization, and visualizing analysis results.

[0015] As a preferred implementation of the present invention, the specific process of analyzing the number of calculation tasks and usage scenarios to obtain the task quantity scale, task timeliness and scenario timeliness value is as follows:

[0016] By calculating the number of tasks, the formula is: Output task quantity scale RSG, Dx is the total computing requirement of the task, C i is the processing capacity of each node, C i =α×CPU speed+β×memory size+γ×network bandwidth, α, β, γ are weight coefficients; calculate the task timeliness through the formula: Output task timeliness Tsx, Nc is the number of nodes involved in the calculation; evaluate the task timeliness and system resource utilization, whether the task can be completed in time in a specific scenario; by formula Output scenario time value Tcs, U is resource utilization, Output resource utilization U, Cs is the actual amount of resources used, and Cz is the total amount of resources allocated.

[0017] As a preferred embodiment of the present invention, the specific process of the processing method or the combination optimization is as follows:

[0018] When the task quantity scale RSG is greater than the corresponding preset threshold, the task is split into multiple subtasks and assigned to different nodes through parallel processing; when the scenario timeliness value Tcs is less than the corresponding preset threshold, the platform load is light and suitable for parallel processing;

[0019] When the scenario timeliness value Tcs is greater than the corresponding preset threshold, the cloud platform elastic expansion function is used to dynamically increase virtual machines or computing nodes to relieve load pressure; when the task timeliness RSG approaches the platform limit, the timeliness is reduced by adding resources;

[0020] When the calculated task quantity, task timeliness, and scenario timeliness values ​​meet the conditions of method 1: Method 1 conditions: RSG>N / θ, and Tcs≤target timeliness, that is, when the task scale is greater than the number of nodes and the scenario resources are sufficient, method 1 is selected;

[0021] The condition for selecting method 2 is: Tcs>>target timeliness / κ; that is, when the scenario timeliness value is greater than the target timeliness / κ, computing resources must be increased to meet the timeliness requirement, and method 2 is selected;

[0022] In complex scenarios, system performance is further optimized through combination; parallel processing + elastic expansion of tasks with large scale and high load; first split the tasks into parallel processing, and dynamically expand computing resources when the resource load approaches the bottleneck.

[0023] As a preferred embodiment of the present invention, the specific process of obtaining the weight of each feature by interpreting the local data is as follows:

[0024] A local explanation model is established. The local explanation model selects a sample x to be explained, generates multiple similar data points x' near the instance, and adds tiny random noise to the original sample features; each generated data point x' is input into the original black box model to obtain the predicted output value f(x'), where f represents the black box model, and approximates the output of f(x) by constructing a local weighted linear model; the local model is defined as g(x'), which is a linear model and is expressed as: Among them, ω h is the weight of feature h, representing the contribution of feature h to x; define a weighted regression loss function L(f,g,π x ),Right now: Among them, the data set of X, π x (x') is the similarity weight between sample x' and target sample x; finally, the weight ω of each feature h is obtained through the linear model h , which is to explain the influence measure of feature h when the target sample point is x; and the influence measure of feature h when the target sample point is x is displayed in a graphical form through the execution unit.

[0025] As a preferred embodiment of the present invention, the specific process of calculating the marginal contribution value of each feature in all feature combinations to obtain the average marginal contribution is:

[0026] Calculate the marginal contribution of each feature in all feature combinations: For each feature h, calculate the marginal contribution in different combinations, that is, the predicted change after adding feature h; if there are H features, then there are 2 H-1 The marginal contribution value of feature h is calculated for each feature combination; for each combination, if the feature combination is B, the marginal contribution is: marginal contribution = f(B∪{h})-f(B), and then the average marginal contribution of the feature is calculated by the formula: in, represents the average marginal contribution of feature h, and |B| is the size of set B. Similarly, the execution unit visualizes the average marginal contribution value of each feature.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. The present invention constructs a data middle platform through a data set sharing unit. The artificial intelligence unit can efficiently identify and meet the diverse data needs of business departments and clarify the data usage scenarios; the automated data collection pipeline unifies the format of multi-source data, realizes the seamless integration of cross-departmental data, and provides a comprehensive customer view for each department; through regular scheduling, the system ensures the real-time and accuracy of the data, thereby supporting faster and more accurate business decisions and improving the efficiency and quality of overall data management.

[0029] 2. The present invention provides a variety of intelligent expansion solutions through the scalability unit constructed by artificial intelligence, which effectively improves the resource utilization and task processing efficiency of the system in a complex environment; the first method adopts multi-node parallel computing to decompose and distribute big data tasks to each node, and realizes load balancing, so that each node can flexibly process data according to its own computing power, greatly accelerating the big data processing speed; the second method uses the elastic expansion function of the cloud platform to dynamically increase or decrease resources according to the load situation to ensure that the system can achieve efficient operation in different load scenarios; this multi-mode and combinable expansion strategy realizes the optimal allocation of resources and cost control, and meets the real-time processing requirements of high concurrency and high load scenarios.

[0030] 3. The present invention performs detailed interpretability analysis on statistical information by decomposing units to help users intuitively understand the role of each feature in model prediction; by constructing a local explanatory model, the influence weight of each feature on a specific sample is displayed, and it is visualized in the form of a bar chart, scatter plot or heat map, so that key features and non-key features are clear at a glance; in addition, by calculating the marginal contribution value and its average value of each feature in different feature combinations, users can quantify and evaluate the independent contribution, positive or negative impact of each feature on the prediction results; this analysis improves the transparency of model decisions, provides users with a clear model explanation, and supports decision optimization. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to facilitate understanding by those skilled in the art, the present invention is further described below with reference to the accompanying drawings.

[0032] Figure 1 It is a principle block diagram of the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0034] It should be understood that the terms “include” and “comprising” used in the specification and claims of the present disclosure indicate the presence of described features, integers, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or collections thereof.

[0035] It should also be understood that the terms used in this disclosure are only for the purpose of describing specific embodiments and are not intended to limit the disclosure. As used in this disclosure and claims, the singular forms of "a", "an", and "the" are intended to include the plural forms unless the context clearly indicates otherwise. It should also be further understood that the term "and / or" used in this disclosure and claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations.

[0036] See also Figure 1 As shown, an information statistics management platform based on artificial intelligence is applied in a computer, comprising an artificial intelligence unit, a data set sharing unit, an extensibility unit, a decomposition unit and an execution unit;

[0037] Build a data center through data set sharing units, and use artificial intelligence and business departments (such as sales, marketing, human resources, etc.) to identify data, clarify the data needs and goals of each department, and understand the specific usage scenarios of the data; and write detailed requirements documents to record the type, frequency and purpose of the data that each department hopes to obtain;

[0038] Then, we establish an automated data collection pipeline to collect data, obtain real-time online services and third-party data platform data by calling APIs; obtain internal data of the database system through data query tools; import the data into the integration platform through file import tools for format conversion, and convert it into a unified format; for example, the date format may need to be unified into the same standard form, or the product classification methods in different systems may need to be unified; merge data from multiple different data sources, combine customer information in the system with corresponding order data, and form a unified customer view for easy analysis; finally, use scheduling tools to regularly obtain the latest data from each data source, automate the data collection process, and ensure the timeliness and accuracy of the data.

[0039] The scalability unit uses artificial intelligence to process the scalability of the platform, specifically:

[0040] Through artificial intelligence, a certain number of different extended processing methods are created, including method 1, method 2, method 3, ..., method n; and different methods or combinations of methods are selected according to different processing scenarios; Method 1: By splitting the data processing tasks to multiple nodes for parallel processing, the large data set is divided into multiple small pieces, each of which can be processed independently; then the task is assigned, the computing tasks are assigned to different computing nodes, and the task load is balanced;

[0041] If the amount of data that needs to be processed by the task is set to D, there are N nodes, and the goal is to allocate the same amount of data to each node, through the formula: Output the amount of data D allocated to the i-th node i , where D is the total amount of data and N is the number of nodes;

[0042] If the computing power of the nodes is inconsistent, different data volumes are allocated to different nodes, and the processing power of each node is set to C. i (C i =α×CPU speed (GHz)+β×memory size (GB)+γ×network bandwidth (Mbps), where α, β, and γ are weight coefficients adjusted according to the importance of the task). The load balancing formula is: Output the amount of data D allocated to the i-th node i ,in, is the sum of the computing power of all nodes. According to this formula, nodes with strong computing power will be allocated more data, while nodes with weak computing power will be allocated less data. For example, there are 3 nodes with computing power of C1=3.2, C2=2.0, and C3=1.6C, then: Assuming the total data volume D = 1000 units of data, the corresponding data volume is allocated to each node according to the formula: D2=294.12, D3=235.29; since the first node has stronger computing power, it is allocated about 470.59 data units; the second node is allocated 294.12 data units; and the third node is allocated 235.29 data units.

[0043] Then, each node processes the shard data locally and performs complexity calculation: through phase 1, the input data is processed and converted to generate a set of key-value pairs, and each node is responsible for processing the data shard assigned to it; if the data size is S, the number of nodes is N, and the data shard size of each node is Si, each node is evenly allocated The computational time complexity of this stage can be expressed as: Where S is the total size of the input data, N is the total number of nodes, and C1 is the computational complexity of a single data unit in phase one, that is, the time required to process each data unit. It should be noted that the computational complexity of a single data unit depends on the specific processing task and algorithm. For example, for simple data preprocessing, the string is divided into words, and each word outputs a key-value pair. Then the computational complexity is at a constant level, that is, O(1) because the time required to process each unit remains basically unchanged. If complex operations are required to process each data unit, and each data unit needs to be sorted or searched, then its computational complexity will be higher, which may be linear O(n) or logarithmic O(logn), depending on the specific content of the data processing.

[0044] Through stage 2, the output (intermediate result) of stage 1 is redistributed to each node for reduction calculation. If the total size of the intermediate result is R, the calculation time complexity of stage 2 can be expressed as: in, C2 is the size of the intermediate result processed by each node, and the computational complexity of a single data unit in stage one. It should be noted that in stage two, the output (key-value pairs) of stage one will be aggregated and reduced. The computational complexity C2 of a single data unit also depends on the specific tasks and algorithms of stage two. If stage two only performs simple accumulation or counting of key-value pairs, such as summing the same key-value pairs, the computational complexity may be a constant level O(1), because the time to process each unit is fixed. If stage two involves sorting, searching, or other complex operations, the computational complexity may be linear O(n) or logarithmic O(logn), where n is the amount of intermediate data allocated to stage two.

[0045] Finally, the partial calculation results of all nodes are combined into the final output time through the network.

[0046] Method 2: Through the elastic expansion function of the cloud platform, dynamically adjust computing and storage resources to meet the expansion needs of the system under different loads;

[0047] In the cloud platform, artificial intelligence continuously monitors the usage of various resources of the cloud platform, such as CPU, memory, network bandwidth, etc., to ensure that the system can meet user needs while efficiently utilizing resources; through the formula: Output resource utilization U (such as CPU utilization), Cs is the actual amount of resources used (such as CPU time or memory consumption), and Cz is the total amount of allocated resources; when the utilization U exceeds a certain set threshold (such as 80%), the platform triggers an expansion operation through the execution unit to increase resources, such as automatically starting a new virtual machine or container to increase computing power, and the resource scheduler reallocates existing tasks to new nodes to ensure load balancing. When expanding resources, the platform needs to dynamically allocate new computing resources; if n new virtual machines need to be started, the capacity of each virtual machine is Rvm (the computing power of each virtual machine), then the total expanded resources are: Rzvm = n×Rvm Output the total amount of expanded resources Rzvm; assign tasks to these newly started virtual machines to achieve load balancing.

[0048] When the utilization rate U is lower than a certain set threshold (such as 20%), the platform releases unnecessary resources through the execution unit to reduce costs; the amount of released resources Rsvm can be calculated by the following formula: Rsvm = Rzvm-Rsg outputs the amount of releasable resources Rsvm, where Rsg is the actual amount of resources currently required;

[0049] Artificial intelligence analyzes the number of computing tasks and usage scenarios; by calculating the number and scale of tasks, the formula is: Output task quantity scale RSG, Dx is the total computing requirement of the task, C i The processing capacity of each node (C i =α×CPU speed (GHz)+β×memory size (GB)+γ×network bandwidth (Mbps), where α, β, and γ are weight coefficients adjusted according to the importance of the task).

[0050] Calculate the task timeliness using the formula: Output task timeliness Tsx (indicates the completion time of the task under ideal conditions), Nc is the number of nodes involved in the calculation; evaluate the task timeliness and system resource utilization, whether the task can be completed in time in a specific scenario; by formula Output scenario time value Tcs (reflects the actual completion time of the task under current resources and load conditions), U is resource utilization.

[0051] Then, based on the calculated task quantity, task timeliness, and scenario timeliness, the AI ​​system can select an appropriate processing method or perform combined optimization.

[0052] For example: when the task quantity scale RSG is larger than the corresponding preset threshold, the task is split into multiple subtasks and assigned to different nodes through parallel processing to increase processing efficiency; when the scenario timeliness value Tcs is less than the corresponding preset threshold, the platform load is light and suitable for parallel processing, and no additional expansion resources are required; the conditions for selecting method 1 are: RSG>N\ / θ, and Tcs≤target timeliness, θ is a positive number, and the specific size is set by artificial intelligence, that is, when the task scale is larger than the number of nodes and the scenario resources are sufficient, method 1 is selected;

[0053] When the scenario timeliness value Tcs is greater than the corresponding preset threshold, the cloud platform elastic expansion function is used to dynamically increase virtual machines or computing nodes to relieve load pressure; when the task timeliness RSG approaches the platform limit, the timeliness can be reduced by increasing resources; the condition for selecting method 2 is: Tcs> target timeliness / κ, κ is a positive number, and the specific size is set by artificial intelligence; that is, when the scenario timeliness value is greater than the target timeliness, computing resources must be increased to meet the timeliness requirements, and method 2 is selected;

[0054] In complex scenarios, system performance can be further optimized through a combination of methods; parallel processing + elastic expansion (method one + method two) has a large task scale and a high system load; first split the task into parallel processing, and dynamically expand computing resources when the resource load approaches a bottleneck.

[0055] The decomposition unit is used to perform interpretable analysis on statistical information, specifically:

[0056] To interpret local data, a local interpretation model is established. The local interpretation model selects a sample x to be explained, generates multiple similar data points x' near the instance, and adds tiny random noise to the original sample features. Each generated data point x' is input into the original black box model to obtain the predicted output value f(x'), where f represents the black box model, and a local weighted linear model is constructed to approximate the output of f(x). The local model is defined as g(x'), which is a linear model and can be expressed as: Among them, ω h is the weight of feature h, representing the contribution of feature h to x; h = 1, 2, ..., p, p is the total number of features. In order to make the linear model better fit the local behavior of the black box model, a weighted regression loss function L(f, g, π x ),Right now:

[0057] Among them, the data set of X, π x(x') is the similarity weight between sample x' and target sample x; finally, through the weight ω of each feature h in the linear model h That is, to explain the influence measure of feature h at the target sample point x; and to display the influence measure of feature h at the target sample point x in a graphical form through the execution unit; using visualization methods such as bar charts, scatter plots or heat maps; It should be noted that: by visualizing these weights, users can clearly know which features play a key role in the prediction of the model and which features have a smaller or unimportant impact.

[0058] Then calculate the marginal contribution of each feature in all feature combinations: for each feature h, calculate the marginal contribution in different combinations, that is, the predicted change after adding feature h; if there are H features, then there are 2 H-1 The marginal contribution value of feature h can be calculated for each feature combination; for each combination, if the feature combination is B, the marginal contribution is: marginal contribution = f(B∪{h})-f(B), and then the average marginal contribution of the feature is calculated by the formula: in, represents the average marginal contribution of feature h, and |B| is the size of set B. Similarly, the execution unit visualizes the average marginal contribution value of each feature. It should be noted that users can see the positive or negative contribution of the feature to the prediction result, its importance in the model decision, and how much information it independently contributes to the model.

[0059] The preferred embodiments of the present invention disclosed above are only used to help explain the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only specific implementation methods. Obviously, many modifications and changes can be made according to the content of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can understand and use the present invention well. The present invention is limited only by the claims and their full scope and equivalents.

Claims

1. An information statistics management platform based on artificial intelligence, including a data set sharing unit, an extensibility unit and a decomposition unit; characterized by: The data set is used to build a data center, where artificial intelligence and the business department perform data identification to record the type, frequency and purpose of the data obtained by each department; then data is collected by establishing an automated data collection pipeline; the file import tool is used to import the data into the integration platform for format conversion, and the data is converted into a unified format; data from multiple different data sources are merged, and customer information is combined with corresponding order data to form a unified customer view; Use scheduling tools to regularly obtain the latest data from various data sources; The scalability unit performs scalability processing on the platform through artificial intelligence, specifically: A certain number of different extended processing methods are created by artificial intelligence, including method 1, method 2, method 3, ..., method n, and different methods or combinations of methods are selected according to different processing scenarios; the number of computing tasks and usage scenarios are analyzed to obtain the number of tasks, the task timeliness, and the scenario timeliness values, and the corresponding processing method is selected or combined optimization is performed according to the calculated number of tasks, the task timeliness, and the scenario timeliness values; The decomposition unit is used to perform interpretable analysis on the statistical information, specifically: The weight of each feature is obtained by interpreting the local data; The marginal contribution value of each feature in all feature combinations is calculated to get the average marginal contribution.

2. The information statistics management platform based on artificial intelligence according to claim 1 is characterized in that: It also includes an artificial intelligence unit and an execution unit; The artificial intelligence unit intelligently identifies the needs of business departments by building a data middle platform. It also dynamically optimizes resource allocation through load balancing and resource monitoring. The artificial intelligence unit also participates in generating local explanatory models to quantify and visualize the impact of each feature. The execution unit is responsible for automatically scheduling data collection, allocating processing tasks, dynamically adjusting resource utilization, and visualizing analysis results.

3. The information statistics management platform based on artificial intelligence according to claim 1 is characterized in that: The specific process of analyzing the number of calculation tasks and usage scenarios to obtain the task quantity scale, task timeliness and scenario timeliness value is as follows: By calculating the number of tasks, the formula is: Output task quantity scale RSG, Dx is the total computing requirement of the task, C i is the processing capacity of each node, C i =α×CPU speed+β×memory size+γ×network bandwidth, α, β, γ are weight coefficients; calculate the task timeliness through the formula: Output task timeliness Tsx, Nc is the number of nodes involved in the calculation; evaluate the task timeliness and system resource utilization, whether the task can be completed in time in a specific scenario; by formula Output scenario time value Tcs, U is resource utilization, Output resource utilization U, Cs is the actual amount of resources used, and Cz is the total amount of resources allocated.

4. The information statistics management platform based on artificial intelligence according to claim 3 is characterized in that: The specific process of the processing method or combined optimization is as follows: When the task quantity scale RSG is greater than the corresponding preset threshold, the task is split into multiple subtasks and assigned to different nodes through parallel processing; when the scenario timeliness value Tcs is less than the corresponding preset threshold, the platform load is light and suitable for parallel processing; When the scenario timeliness value Tcs is greater than the corresponding preset threshold, the cloud platform elastic expansion function is used to dynamically increase virtual machines or computing nodes to relieve load pressure; when the task timeliness RSG approaches the platform limit, the timeliness is reduced by adding resources; When the calculated task quantity, task timeliness, and scenario timeliness values ​​meet the conditions of method 1: Method 1 conditions: RSG>N / θ, and Tcs≤target timeliness, that is, when the task scale is greater than the number of nodes and the scenario resources are sufficient, method 1 is selected; The condition for selecting method 2 is: Tcs> target timeliness / κ; that is, when the scenario timeliness value is greater than the target timeliness / κ, computing resources must be increased to meet the timeliness requirement, and method 2 is selected; In complex scenarios, system performance is further optimized through combination; parallel processing + elastic expansion of tasks with large scale and high load; first split the tasks into parallel processing, and dynamically expand computing resources when the resource load approaches the bottleneck.

5. The information statistics management platform based on artificial intelligence according to claim 1 is characterized in that: The specific process of obtaining the weight of each feature by interpreting the local data is as follows: A local explanation model is established. The local explanation model selects a sample x to be explained, generates multiple similar data points x' near the instance, and adds tiny random noise to the original sample features; each generated data point x' is input into the original black box model to obtain the predicted output value f(x'), where f represents the black box model, and approximates the output of f(x) by constructing a local weighted linear model; the local model is defined as g(x'), which is a linear model and is expressed as: Among them, ω h is the weight of feature h, representing the contribution of feature h to x; define a weighted regression loss function L(f,g,π x ),Right now: Among them, the data set of X, π x (x') is the similarity weight between sample x' and target sample x; finally, the weight ω of each feature h is obtained through the linear model h , which is to explain the influence measure of feature h when the target sample point is x; and the influence measure of feature h when the target sample point is x is displayed in a graphical form through the execution unit.

6. The information statistics management platform based on artificial intelligence according to claim 5 is characterized in that: The specific process of calculating the marginal contribution value of each feature in all feature combinations to obtain the average marginal contribution is: Calculate the marginal contribution of each feature in all feature combinations: For each feature h, calculate the marginal contribution in different combinations, that is, the predicted change after adding feature h; if there are H features, then there are 2 H-1 The marginal contribution value of feature h is calculated for each feature combination; for each combination, if the feature combination is B, the marginal contribution is: marginal contribution = f(B∪{h})-f(B), and then the average marginal contribution of the feature is calculated by the formula: in, represents the average marginal contribution of feature h, and |B| is the size of set B. Similarly, the execution unit visualizes the average marginal contribution value of each feature.

Images