Quotation accuracy attribution analysis method and device, equipment and storage medium

Abstract

The invention relates to a quotation accuracy attribution analysis method and device, equipment and a storage medium. The method comprises the following steps: acquiring original quotation accuracy data; performing feature engineering processing on the original quotation accuracy data to obtain multi-dimensional target features; inputting the multi-dimensional target features into a preset prediction model for prediction to obtain a prediction result of the multi-dimensional target features; and inputting the prediction result into a preset attribution analysis model for analysis to obtain an attribution analysis result. According to the method, the overall trend change can be captured through the preset prediction model, and then the multi-dimensional attribution analysis is performed through the preset attribution analysis model to obtain the specific attribution analysis result, so that the existing problem can be quickly solved directly according to the attribution analysis result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a quotation accuracy attribution analysis method and device, equipment and a storage medium. BACKGROUND

[0002] In a hotel distribution system, quotation accuracy (i.e. the consistency of the quotation returned by the platform and the actual bookable result) is a key indicator for measuring service reliability and user experience. Abnormal decrease of this indicator is often caused by multiple heterogeneous factors, including but not limited to: supplier interface response delay, hotel data source synchronization lag, and business dimension combination abnormality (such as specific length of stay (LOS), lead time, meal type, bed type, country, hotel ID or supplier ID, etc.).

[0003] Related technologies usually use time series models (such as ARIMA, Prophet) or machine learning models (such as XGBoost) for accuracy prediction analysis.

[0004] However, since a single time series model (such as ARIMA, Prophet) or machine learning model (such as XGBoost) can only output numerical prediction or abnormal alarm, it cannot provide root cause explanation to the business dimension, resulting in inability to directly solve the existing problems. SUMMARY

[0005] Embodiments of the present application provide a quotation accuracy attribution analysis method, device, equipment and storage medium, aiming to solve the problem that traditional single time series model (such as ARIMA, Prophet) or machine learning model can only output numerical prediction or abnormal alarm, and cannot provide root cause explanation to the business dimension.

[0006] In a first aspect, embodiments of the present application provide a quotation accuracy attribution analysis method, which comprises:

[0007] Obtaining original quotation accuracy data;

[0008] Performing feature engineering processing on the original quotation accuracy data to obtain multi-dimensional target features;

[0009] Inputting the multi-dimensional target features into a preset prediction model for prediction to obtain a prediction result of the multi-dimensional target features;

[0010] Inputting the prediction result into a preset attribution analysis model for analysis to obtain an attribution analysis result.

[0011] Further technical solutions are as follows: the feature engineering processing on the original quotation accuracy data to obtain multi-dimensional target features comprises:

[0012] Within a preset sliding time window, feature construction and calculation are performed based on the original quotation accuracy data to obtain multi-dimensional target features, wherein the multi-dimensional target features include minute-level statistical features, historical accuracy features, frequency features, change rate features, and combined features.

[0013] A further technical solution is that before the original quotation accuracy data is processed by feature engineering to obtain multi-dimensional target features, the method further includes:

[0014] The original quotation accuracy data is preprocessed to obtain preprocessed quotation accuracy data, wherein the preprocessing includes missing value filling, outlier removal, and time alignment operations.

[0015] A further technical solution is that the attribution analysis result includes the average marginal contribution of each feature to the prediction result, the prediction result is input into a preset attribution analysis model for analysis to obtain an attribution analysis result, which includes:

[0016] A first feature set is obtained, and the first feature set contains multiple features.

[0017] A first feature is taken as a target feature, and based on the first feature set and the target feature, the average marginal contribution of the target feature to the prediction result is determined.

[0018] The next feature of the target feature is taken as the target feature, and the step of determining the average marginal contribution of the target feature to the prediction result based on the first feature set and the target feature is returned until the average marginal contribution of each feature to the prediction result is obtained.

[0019] A further technical solution is that the average marginal contribution of the target feature to the prediction result is determined based on the first feature set and the target feature, which includes:

[0020] Based on the first feature set and the target feature, a second feature set is obtained, wherein the second feature set does not contain the target feature.

[0021] The first feature quantity corresponding to the first feature set and the second feature quantity corresponding to the second feature set are determined.

[0022] Based on the first feature quantity, the second feature quantity, and a preset weight coefficient, the average marginal contribution of the target feature to the prediction result is obtained.

[0023] Further technical solutions are that after the average marginal contribution degree of the target feature to the prediction result is obtained, the method further includes:

[0024] determining whether the average marginal contribution degree exceeds a preset contribution degree threshold value;

[0025] generating an alarm prompt if the average marginal contribution degree exceeds the preset contribution degree threshold value.

[0026] Further technical solutions are that the multi-dimensional target feature includes attribution analysis results including global feature importance ranking information and single-sample local explanation information.

[0027] In a second aspect, the embodiments of the present application further provide an attribution analysis device for quote accuracy, which includes modules for executing the above method.

[0028] In a third aspect, the embodiments of the present application further provide a computer device, which includes a memory and a processor, the memory stores a computer program, and the processor implements the above method when executing the computer program.

[0029] In a fourth aspect, the embodiments of the present application further provide a computer readable storage medium, which stores a computer program, and the computer program can implement the above method when executed by a processor.

[0030] The embodiments of the present application provide an attribution analysis method, device, equipment and storage medium for quote accuracy. The method includes: obtaining original quote accuracy rate data; performing feature engineering processing on the original quote accuracy rate data to obtain multi-dimensional target features; inputting the multi-dimensional target features into a preset prediction model for prediction to obtain prediction results of the multi-dimensional target features; and inputting the prediction results into a preset attribution analysis model for analysis to obtain attribution analysis results.

[0031] The embodiments of the present application input multi-dimensional target features into a preset prediction model for prediction, and then use a preset attribution analysis model to analyze the prediction results output by the preset prediction model. In this way, the preset prediction model and the preset attribution analysis model are combined, the overall trend change is captured through the preset prediction model, multi-dimensional attribution analysis is performed through the preset attribution analysis model, specific attribution analysis results are obtained, and then existing problems can be quickly solved directly according to the attribution analysis results. BRIEF DESCRIPTION OF DRAWINGS

[0032] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application.

[0033] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, for those skilled in the field, other drawings can also be obtained based on these drawings without any creative effort.

[0034] One or more embodiments are illustrated by way of example in the drawings that are not intended to be limiting of the embodiments. Like references numerals in the drawings denote like elements, unless otherwise specified. The drawings in which:

[0035] Figure 1 A flowchart of a first embodiment of a bid accuracy attribution analysis method provided by the present application;

[0036] Figure 2 An interaction processing schematic diagram of various modules in a bid accuracy attribution analysis device provided by the present application;

[0037] Figure 3 A structural schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0038] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort fall within the scope of protection of the present application.

[0039] The following disclosure provides many different embodiments, or examples, for implementing different structures of the present application. For the purpose of simplification, the components and arrangements of the specific examples are described in the following. Of course, they are only examples and the purpose is not to limit the present application. In addition, the present application can repeatedly refer to numbers and / or letters in different examples. Such repetition is for the purpose of simplification and clarity, and it does not indicate the relationship between the various embodiments and / or arrangements discussed.

[0040] It should be understood that when used in the specification and the appended claims, the terms "comprise" and "include" indicate the presence of described features, integers, steps, operations, elements, and / or components, but do not exclude one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0041] It should also be understood that the terms used herein are for the purpose of describing particular embodiments and are not intended to limit the application. As used in this specification and the appended claims, the singular forms "a," "an" and "the" are intended to include plural referents unless the context clearly dictates otherwise.

[0042] It should further be understood that the term "and / or" as used herein refers to any combination of one or more of the associated listed items, and all possible combinations, and includes these combinations.

[0043] As used in this specification and the appended claims, the term "if" can be interpreted as meaning "when" or "once" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrase "if it is determined" or "if [a described condition or event] is detected" can be interpreted to mean "once it is determined" or "in response to determining" or "once [the described condition or event] is detected" or "in response to detecting [the described condition or event]", depending on the context.

[0044] To solve the above problems, the application provides a quotation accuracy attribution analysis method, which can first capture overall trend changes through a preset prediction model, then perform multi-dimensional attribution analysis through a preset attribution analysis model to obtain specific attribution analysis results, and then directly solve the existing problems according to the attribution analysis results.

[0045] Referring to Figure 1 , Figure 1 The flowchart of the first embodiment of the quotation accuracy attribution analysis method provided by the application is shown in the figure, and the quotation accuracy attribution analysis method comprises the following steps:

[0046] Step 110: Obtain original quotation accuracy data.

[0047] Step 120: Perform feature engineering processing on the original quotation accuracy data to obtain multi-dimensional target features.

[0048] Step 130: Input the multi-dimensional target features into a preset prediction model for prediction to obtain a prediction result of the multi-dimensional target features.

[0049] Step 140: Input the prediction result into a preset attribution analysis model for analysis to obtain an attribution analysis result.

[0050] The embodiment predicts by inputting the multi-dimensional target feature into a preset prediction model, and analyzes the prediction result output by the preset prediction model by using a preset attribution analysis model. In this way, the preset prediction model and the preset attribution analysis model are combined, the overall trend change is captured by the preset prediction model, multi-dimensional attribution analysis is performed by the preset attribution analysis model, and specific attribution analysis results are obtained. Then, the existing problems can be quickly solved according to the attribution analysis results.

[0051] In some possible embodiments, before the original quotation accuracy rate data is processed by feature engineering to obtain multi-dimensional target features, refer to the second embodiment of the quotation accuracy rate attribution analysis method provided in the present application. The quotation accuracy rate attribution analysis method includes the following steps:

[0052] Step 210: Obtain original quotation accuracy rate data.

[0053] In some embodiments, quotation request data (i.e., original quotation accuracy rate data) can be collected in real time from a log database D, and a single request is defined as:

[0054] r i ={t i ,s i ,a i ,h i ,c i ,LOS i ,LT i ,meal i ,bed i ,issue i}。

[0055] Wherein, t i is the request timestamp, s i is the supplier ID, a i is the supplier sub-account ID, h i is the hotel ID, c i is the hotel country code, LOS i is the length of stay (Length of Stay), LT i is the lead time (Leadtime), meal i is the meal type, bed i is the bed type, and issue i ∈{0·1} is the accuracy rate label, which can determine whether there is data inconsistency.

[0056] In this way, through minute-level data collection, real-time performance can be improved, and early warning of accuracy rate decline can be performed.

[0057] Step 220: pre-processing the original quotation accuracy rate data to obtain pre-processed quotation accuracy rate data.

[0058] The pre-processing includes missing value filling, outlier removal, and time alignment operations.

[0059] In this way, by pre-processing the original quotation accuracy rate data, the quality of the data can be improved.

[0060] Step 230: performing feature engineering processing on the pre-processed quotation accuracy rate data to obtain multi-dimensional target features.

[0061] In some possible implementations, the feature engineering processing on the original quotation accuracy rate data to obtain multi-dimensional target features includes: based on the original quotation accuracy rate data, performing feature construction and calculation within a pre-set sliding time window to obtain multi-dimensional target features, wherein the multi-dimensional target features include minute-level statistical features, historical accuracy rate features, frequency features, change rate features, and combined features.

[0062] After the pre-processed quotation accuracy rate data is pre-processed, feature construction and calculation can be performed based on the pre-processed quotation accuracy rate data.

[0063] In some embodiments, the minute-level statistical features include sliding statistical features such as mean, variance, maximum value, minimum value, etc.

[0064] For example, the mean can be calculated using the following formula 1:

[0065]

[0066] where n is the total number of data points, i is the index (from 1 to n), x i is the specific value of the ith data.

[0067] The mean can be calculated using the following formula 2:

[0068]

[0069] The maximum value can be calculated using the following formula 3:

[0070]

[0071] The minimum value can be calculated using the following formula 3:

[0072]

[0073] The slope can be calculated using the following formula 5:

[0074]

[0075] In some embodiments, the rate of change feature includes a rate of change, which can be calculated using Formula 6:

[0076]

[0077] In some embodiments, the historical accuracy rate feature includes a lag feature and a target encoding feature, where the lag feature can be calculated using Formula 7 below:

[0078]

[0079] The target encoding feature can be calculated using Formula 8 and Formula 9 below:

[0080]

[0081] where N s is the number of historical requests for supplier s, and N h is the number of historical requests for hotel h.

[0082] Since existing technologies mostly use single time series features (such as daily accuracy rate) or a small number of static features, they cannot capture multi-dimensional information such as minute-level statistics, historical accuracy rate, frequency change, and risk change at the same time. In the present application, minute-level statistical features, historical accuracy rate features, frequency type features, and rate of change type features are introduced in feature engineering, and features are constructed for different business entities (supplier, supplier account, customer ID, hotel ID, country code, etc.), covering the combination space of time dimension and entity dimension, which can improve the joint modeling capability of the prediction model for time series and multi-dimensional features.

[0083] Step 240: inputting the multi-dimensional target feature into a preset prediction model for prediction to obtain a prediction result of the multi-dimensional target feature.

[0084] Step 250: inputting the prediction result into a preset attribution analysis model for analysis to obtain an attribution analysis result.

[0085] In some possible implementations, the attribution analysis result includes an average marginal contribution degree of each feature to the prediction result, and step 250, i.e., inputting the prediction result into a preset attribution analysis model for analysis to obtain an attribution analysis result, includes:

[0086] Step 251: obtaining a first feature set, the first feature set containing a plurality of features.

[0087] Step 252: taking a first feature as a target feature, and determining an average marginal contribution degree of the target feature to the prediction result based on the first feature set and the target feature.

[0088] Step 253: taking the next feature of the target feature as the target feature, returning to the step of determining the average marginal contribution of the target feature to the prediction result based on the first feature set and the target feature, until the average marginal contribution of each feature to the prediction result is obtained.

[0089] In some possible implementations, the determining of the average marginal contribution of the target feature to the prediction result based on the first feature set and the target feature comprises:

[0090] Step 2521: obtaining a second feature set based on the first feature set and the target feature, wherein the second feature set does not contain the target feature.

[0091] Step 2522: determining a first feature quantity corresponding to the first feature set and a second feature quantity corresponding to the second feature set.

[0092] Step 2523: obtaining the average marginal contribution of the target feature to the prediction result based on the first feature quantity, the second feature quantity, and a preset weight coefficient.

[0093] In some embodiments, the contribution of each feature in the bid data in the past few minutes can be calculated.

[0094] Specifically, for steps 2521-2523, in some embodiments, reference can be made to the following formula 10.

[0095]

[0096] wherein j is the target feature, φ j is the average marginal contribution of feature j to the prediction result, F is a set of all features, i.e., the first feature set, S is a subset of the feature set, i.e., the second feature set, and does not contain the target feature j, |F| is the total number of features, i.e., the first feature quantity, |S| is the second feature quantity, |S|!(|F|-|S|-1)! is a preset weight coefficient, which is derived from the Shapley value theory and is used to ensure the fairness of the contribution allocation of all features, and f S (x S ) is the prediction result of the prediction model using only the feature subset S (i.e., the second feature set). S∪{j} (x S∪{j} ) is the prediction result of the prediction model when the feature subset S is added with the target feature such as feature j.

[0097] wherein when the average marginal contribution is a positive value, it represents an improvement in the prediction result, and when the average marginal contribution is a negative value, it represents a decrease in the prediction result.​

[0098] In some possible implementations, the attribution analysis module can first use gradient boosting trees (LightGBM / XGBoost) to perform binary classification modeling prediction, and can use a target function as shown in the following formula 11.

[0099]

[0100] Wherein, L is a binary classification loss function (Logloss), and Ω is a regularization term.

[0101] In some implementations, the global feature importance ranking can be obtained using the following formula 12.

[0102]

[0103] Wherein, N is the number of samples.

[0104] In this way, the SHAP value calculation and the Attention weight analysis can meet the attribution needs of the business and enhance the explainability.

[0105] In some possible implementations, the multi-dimensional target feature includes global feature importance ranking information and single-sample local explanation information.

[0106] In this way, by introducing SHAP value calculation in attribution analysis, not only the global feature contribution degree can be provided, but also the single-sample local explanation can be performed to determine the influencing factors of a specific request.

[0107] In some possible implementations, after obtaining the average marginal contribution degree of the target feature to the prediction result, the method further includes:

[0108] Step 31: determining whether the average marginal contribution degree exceeds a preset contribution degree threshold.

[0109] Step 32: if the average marginal contribution degree exceeds the preset contribution degree threshold, generating an alarm prompt and recording a log.

[0110] Based on the above embodiments, a specific application scenario of a hotel is taken as an example for illustration, and the following flow is referred to:

[0111] 1) Data collection:

[0112] For example, the quotation request data in a specified time range, i.e., the original quotation accuracy data, is extracted from the database table rateaccuracy.rate_accuracy_channel.

[0113] Specifically, the collected experimental data can be real quote request logs from a global hotel distribution platform, with a time span of 90 consecutive days, and a total sample size of about 120 million request records.

[0114] 2) Data preprocessing:

[0115] 2-1) Filter records with missing supplier ID or hotel ID;

[0116] 2-2) Bucket by minute, generate statistical features within the window.

[0117] 3) Feature construction, including minute-level statistical features, historical features, frequency features, change rate features, and combined features;

[0118] Among them, the minute-level statistical features include: accuracy mean, variance, maximum value, minimum value within the window; historical features include the accuracy mean and change rate of the past k minutes; frequency features are the proportion of the number of occurrences of a certain feature value within the window, and change rate features are the accuracy change speed; combined features can be supplier-hotel combinations, hotel-room type combinations, etc.

[0119] 3) Prediction model training and feature weight calculation:

[0120] Use LightGBM to train binary classification on whether there is a quote inconsistency, and get the accuracy label output by the prediction model, where the accuracy label can be 1 or 0, 0 indicating consistent data and 1 indicating inconsistent data.

[0121] 4) Attribution analysis and verification:

[0122] 4-1) Use SHAP values to calculate global feature importance and local explanations for individual requests, and get attribution analysis results including global feature importance ranking information and single-sample local explanation information.

[0123] 4-2) Sort the feature contribution degree of the time period when the accuracy decreases

[0124] 4-3) Compare high-contribution features with actual business logs to verify attribution accuracy

[0125] 4-4) When the contribution degree of a certain dimension feature (i.e. target feature) exceeds the pre-set contribution degree threshold θ, it indicates an anomaly, and the system automatically generates an alarm and records the log.

[0126] 5) Experimental results

[0127] For example, when the model identifies that the contribution degree of "Supplier ID = 123" is 0.42 (significantly higher than other features), it is consistent with the actual operation and maintenance logs;

[0128] When the model identifies that the contribution degree of "hotel ID = 456" is 0.37, it is consistent with the actual operation and maintenance log.

[0129] Therefore, the application supports minute-level data collection and feature calculation, realizes minute-level accuracy trend prediction and threshold alarm by combining a prediction model, and improves real-time performance.

[0130] In addition, the feature construction and model design provided by the application can be applied to a multi-source heterogeneous data scene, and can be extended to ticket distribution, car rental platforms and other distribution businesses in addition to the hotel distribution industry.

[0131] Corresponding to the above offer accuracy attribution analysis method, the application also provides an offer accuracy attribution analysis device. The offer accuracy attribution analysis device includes a module for executing the above offer accuracy attribution analysis method, and the offer accuracy attribution analysis device can be configured in a desktop computer, a tablet computer, a laptop computer, and the like. terminal.

[0132] Referring to Figure 2 , the offer accuracy attribution analysis method module can include a data collection module, a feature engineering module, a prediction model module, and an attribution analysis module, and the functions of each module are as follows:

[0133] The data collection module is configured to obtain raw offer accuracy data.

[0134] The feature engineering module is configured to perform feature engineering processing on the raw offer accuracy data to obtain multi-dimensional target features.

[0135] The prediction model module is configured to perform prediction based on the multi-dimensional target features to obtain a prediction result of the multi-dimensional target features.

[0136] The attribution analysis module is configured to perform analysis based on the prediction result to obtain an attribution analysis result.

[0137] As shown in Figure 3 , the application embodiment provides a computer device, which includes a processor 111, a communication interface 112, a memory 113, and a communication bus 114, wherein the processor 111, the communication interface 112, and the memory 113 communicate with each other through the communication bus 114.

[0138] The memory 113 is configured to store a computer program.

[0139] In an embodiment of the application, the processor 111 is configured to execute the program stored in the memory 113 to implement the offer accuracy attribution analysis method provided by any one of the preceding method embodiments, including:

[0140] Obtaining raw offer accuracy data.

[0141] The original quotation accuracy rate data is subjected to feature engineering processing to obtain multi-dimensional target features;

[0142] The multi-dimensional target features are input into a preset prediction model for prediction to obtain a prediction result of the multi-dimensional target features;

[0143] The prediction result is input into a preset attribution analysis model for analysis to obtain an attribution analysis result.

[0144] A further technical solution is that the original quotation accuracy rate data is subjected to feature engineering processing to obtain multi-dimensional target features, including:

[0145] Within a preset sliding time window, multi-dimensional target features are obtained based on feature construction and calculation of the original quotation accuracy rate data, wherein the multi-dimensional target features include minute-level statistical features, historical accuracy rate features, frequency features, change rate features, and combined features.

[0146] A further technical solution is that before the original quotation accuracy rate data is subjected to feature engineering processing to obtain multi-dimensional target features, the method further includes:

[0147] The original quotation accuracy rate data is preprocessed to obtain preprocessed quotation accuracy rate data, wherein the preprocessing includes missing value filling, outlier removal, and time alignment operations.

[0148] A further technical solution is that the attribution analysis result includes average marginal contribution degrees of each feature to the prediction result, and the prediction result is input into a preset attribution analysis model for analysis to obtain an attribution analysis result, including:

[0149] A first feature set is obtained, and the first feature set contains multiple features;

[0150] A first feature is taken as a target feature, and based on the first feature set and the target feature, an average marginal contribution degree of the target feature to the prediction result is determined;

[0151] The next feature of the target feature is taken as a target feature, and the step of determining the average marginal contribution degree of the target feature to the prediction result based on the first feature set and the target feature is returned to until average marginal contribution degrees of each feature to the prediction result are obtained.

[0152] A further technical solution is that the average marginal contribution degree of the target feature to the prediction result is determined based on the first feature set and the target feature, including:

[0153] obtaining a second feature set based on the first feature set and the target feature, wherein the second feature set does not contain the target feature;

[0154] determining a first feature quantity corresponding to the first feature set and a second feature quantity corresponding to the second feature set;

[0155] obtaining an average marginal contribution degree of the target feature to the prediction result based on the first feature quantity, the second feature quantity, and a preset weight coefficient.

[0156] Further, after obtaining the average marginal contribution degree of the target feature to the prediction result, the method further includes:

[0157] determining whether the average marginal contribution degree exceeds a preset contribution threshold;

[0158] If the average marginal contribution degree exceeds the preset contribution threshold, an alarm prompt is generated. Further, the multi-dimensional target feature includes attribution analysis results including global feature importance ranking information and single-sample local explanation information.

[0159] It can be understood by those skilled in the art that all or part of the processes in the method of the above-mentioned embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a storage medium, which is a computer readable storage medium. The computer program is executed by at least one processor in the computer system to implement the process steps of the above-mentioned embodiment of the method.

[0160] Therefore, the embodiment of the present application also provides a computer readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the attribution analysis method of the bid accuracy rate provided by any one of the above-mentioned method embodiments are implemented, including:

[0161] obtaining original bid accuracy rate data;

[0162] performing feature engineering processing on the original bid accuracy rate data to obtain multi-dimensional target features;

[0163] inputting the multi-dimensional target features into a preset prediction model for prediction to obtain a prediction result of the multi-dimensional target features;

[0164] inputting the prediction result into a preset attribution analysis model for analysis to obtain attribution analysis results.

[0165] Further, the feature engineering processing on the original bid accuracy rate data to obtain multi-dimensional target features includes:

[0166] Within a preset sliding time window, feature construction and calculation are performed based on the original quotation accuracy data to obtain multi-dimensional target features, wherein the multi-dimensional target features include minute-level statistical features, historical accuracy features, frequency features, change rate features, and combined features.

[0167] A further technical solution is that before the original quotation accuracy data is processed by feature engineering to obtain multi-dimensional target features, the method further comprises:

[0168] The original quotation accuracy data is preprocessed to obtain preprocessed quotation accuracy data, wherein the preprocessing includes missing value filling, outlier removal, and time alignment operations.

[0169] A further technical solution is that the attribution analysis result includes the average marginal contribution of each feature to the prediction result, the prediction result is input into a preset attribution analysis model for analysis to obtain an attribution analysis result, which includes:

[0170] A first feature set is obtained, and the first feature set contains multiple features.

[0171] The first feature is taken as a target feature, and based on the first feature set and the target feature, the average marginal contribution of the target feature to the prediction result is determined.

[0172] The next feature of the target feature is taken as the target feature, and the step of determining the average marginal contribution of the target feature to the prediction result based on the first feature set and the target feature is returned to until the average marginal contribution of each feature to the prediction result is obtained.

[0173] A further technical solution is that determining the average marginal contribution of the target feature to the prediction result based on the first feature set and the target feature includes:

[0174] Based on the first feature set and the target feature, a second feature set is obtained, wherein the second feature set does not contain the target feature.

[0175] The first feature quantity corresponding to the first feature set and the second feature quantity corresponding to the second feature set are determined.

[0176] Based on the first feature quantity, the second feature quantity, and a preset weight coefficient, the average marginal contribution of the target feature to the prediction result is obtained.

[0177] A further technical solution is that after the average marginal contribution of the target feature to the prediction result is obtained, the method further comprises:

[0178] determine whether the average marginal contribution degree exceeds a preset contribution degree threshold value;

[0179] generate an alarm prompt if the average marginal contribution degree exceeds the preset contribution degree threshold value. Further, the multi-dimensional target feature includes attribution analysis results including global feature importance ranking information and single-sample local explanation information.

[0180] The storage medium is an entity, non-transient storage medium, for example, can be a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk or an optical disk, and various entity storage media that can store program codes. The computer readable storage medium can be non-volatile or volatile.

[0181] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been described in the above description in general terms. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. A person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0182] In several embodiments provided in the present application, it should be understood that the disclosed apparatus and method can be implemented in other ways. For example, the apparatus embodiments described above are only schematic. For example, the division of each unit is only a logical function division, and actual implementation can have another division manner. For example, a plurality of units or components can be combined or integrated into another system, or some features can be omitted or not executed.

[0183] The steps in the method embodiments of the present application can be adjusted, combined and deleted in sequence according to actual needs. The units in the apparatus embodiments of the present application can be combined, divided and deleted according to actual needs. In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0184] The integrated unit, if implemented in the form of a software function unit and sold or used as an independent product, can be stored in a storage medium. Based on such understanding, the technical solutions of the present application essentially or say the part that contributes to the prior art, or the whole or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, a terminal, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application.

[0185] In the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the relevant description of other embodiments.

[0186] Obviously, those skilled in the art can make various modifications and variations to the present application without departing from the spirit and scope of the present application. Thus, any modifications and variations of the present application that fall within the scope of the claims of the present application and their equivalent technologies shall be included in the scope of the present application.

[0187] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any modifications or replacements that those skilled in the art can easily think of within the technical scope disclosed by the present application shall be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.

Claims

1. An attribution analysis method for quotation accuracy, characterized in that, The attribution analysis method for the accuracy of the quote includes: Obtain the accuracy data of the original quote; The original quotation accuracy data is subjected to feature engineering to obtain multi-dimensional target features; The multidimensional target features are input into a preset prediction model for prediction, and the prediction results of the multidimensional target features are obtained. The prediction results are input into a preset attribution analysis model for analysis to obtain the attribution analysis results.

2. The method according to claim 1, characterized in that, The step of performing feature engineering on the original quotation accuracy data to obtain multi-dimensional target features includes: Within a preset sliding time window, features are constructed and calculated based on the original quote accuracy data to obtain multidimensional target features, which include minute-level statistical features, historical accuracy features, frequency features, rate of change features, and combined features.

3. The method according to claim 1, characterized in that, Before performing feature engineering on the original quote accuracy data to obtain multidimensional target features, the method further includes: The original quote accuracy data is preprocessed to obtain preprocessed quote accuracy data. The preprocessing includes operations such as missing value imputation, outlier removal, and time alignment.

4. The method according to claim 1, characterized in that, The attribution analysis results include the average marginal contribution of each feature to the prediction results. The process of inputting the prediction results into a preset attribution analysis model for analysis to obtain the attribution analysis results includes: Obtain a first feature set, which contains multiple features; The first feature is used as the target feature, and the average marginal contribution of the target feature to the prediction result is determined based on the first feature set and the target feature. Let the next feature of the target feature be taken as the target feature, and return to the step of determining the average marginal contribution of the target feature to the prediction result based on the first feature set and the target feature, until the average marginal contribution of each feature to the prediction result is obtained.

5. The method according to claim 4, characterized in that, The step of determining the average marginal contribution of the target feature to the prediction result based on the first feature set and the target feature includes: Based on the first feature set and the target feature, a second feature set is obtained, wherein the second feature set does not include the target feature; Determine the number of first features corresponding to the first feature set, and the number of second features corresponding to the second feature set; Based on the first number of features, the second number of features, and the preset weighting coefficients, the average marginal contribution of the target feature to the prediction result is obtained.

6. The method according to claim 5, characterized in that, After obtaining the average marginal contribution of the target feature to the prediction result, the method further includes: Determine whether the average marginal contribution exceeds a preset contribution threshold; If the average marginal contribution exceeds a preset contribution threshold, an alarm will be generated.

7. The method according to claim 1, characterized in that, The multidimensional target features include attribution analysis results that include global feature importance ranking information and single-sample local explanation information.

8. An attribution analysis device for quotation accuracy, characterized in that, Includes a module for performing the method as described in any one of claims 1-7.

9. A computer device, characterized in that, The computer device includes a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, can implement the method as described in any one of claims 1-7.

Images