CN110688373A - OFFSET method based on logistic regression - Google Patents
OFFSET method based on logistic regression Download PDFInfo
- Publication number
- CN110688373A CN110688373A CN201910877298.7A CN201910877298A CN110688373A CN 110688373 A CN110688373 A CN 110688373A CN 201910877298 A CN201910877298 A CN 201910877298A CN 110688373 A CN110688373 A CN 110688373A
- Authority
- CN
- China
- Prior art keywords
- model
- offset
- data
- exposure
- logistic regression
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
The invention discloses an OFFSET method based on logistic regression. The invention has the following implementation steps: step 1, confirming default factors influencing customers, including operator data, consumption behavior data, loan application information, terminal equipment information, active geographic position information, credit score and the like; step 2, collecting, converting, quantifying and deriving related data; step 3, constructing a model to carry out iteration and operation; step 4, effect verification; the invention combines the logistic regression and offset method to solve the problem that systematic characteristic data are missing in a part of data sources in a certain time period, and builds a default model in steps and carries out instance verification. From the comparison of the results, the stability and the accuracy of the model are obviously improved after the offset method is applied.
Description
Technical Field
The invention provides an OFFSET method based on logistic regression based on data information generated by consumption staging users and by combining machine learning, feature engineering and other related technologies. The invention predicts the default probability by using an offset method in logistic regression.
Background
The existing data information generated by consuming users in stages has the problem that a lot of feature data is missing in a part of data sources in a certain period, for example, access to a third party data (represented by X1) is started from a certain intermediate point, so that users before the access point (represented by C1) completely miss the data sources. In the case of such systematic loss of data, it is not reasonable to directly process the characteristics of the data source into missing values or null values, because the loss of the data source is not random loss, and the conventional data padding or filling method can cause serious data deviation. In addition, if modeling is performed only with samples where the feature X1 is not missing, the information contained in the user C1 sample before the data source is accessed will be lost, and the accuracy of the model will be affected.
Disclosure of Invention
The invention aims to provide an OFFSET method based on logistic regression.
The technical scheme adopted by the invention for solving the technical problem comprises the following steps:
step 1, confirming default factors influencing customers, including operator data, consumption behavior data, loan application information, terminal equipment information, active geographic position information, credit score and the like;
step 2, collecting, converting, quantifying and deriving related data;
step 3, constructing a model to carry out iteration and operation;
step 4, effect verification;
the confirmation of the factor affecting the default of the customer in the step 1 is specifically realized as follows:
the operator data comprises a call detail sheet, a short message detail sheet and a recharging record of the application client;
the consumption behavior data mainly refers to on-line consumption data, including the consumption amount of shopping and the types of consumed commodities each month; extracting features reflecting the income level and the consumption habits of the customers through the online consumption data;
the loan application information refers to application information of a client on other loan platforms, and currently, the number of applications and historical loan conditions are applied on a plurality of platforms;
the terminal device information comprises a device type, an app list and various app liveness degrees;
the credit scores of the clients comprise online and offline consumption stages, sesame credit scores and financing lease credit conditions;
the step 2of collecting, converting, quantifying and deriving the relevant data means that after establishing dimensions that may affect the default of the customer in the step 1, the corresponding data is converted and quantified, and the variables required by the generation are derived and processed.
2. The OFFSET method based on logistic regression as claimed in claim 1, wherein the model constructed in step 3 is iterated and operated, specifically implemented as follows:
3.1 the characteristic engineering comprises the processing of abnormal values and missing values of the characteristics, data transformation and characteristic selection;
①, removing samples of abnormal values of basic feature items of the trading platform, and filling the missing situation of part of feature items;
②, for the record that the third party multi-head feature item is not matched with, no filling is done, and the null value problem is considered in the algorithm adopted subsequently;
③ the operator data provides related call detail, short message detail and recharge records, the user generates new derived variables around the call time, call time and calling and called conditions, new derived variables around the short message quantity and receiving and sending conditions, and new derived variables according to recharge mode, money amount and frequency, thereby obtaining derived characteristics;
3.2 construction of models
3.2.1 model training description:
1. based on the original characteristics and the derivative characteristics generated in the characteristic engineering step, constructing a plurality of models by adopting a plurality of characteristic combinations, and finally selecting an optimal model through a plurality of evaluation indexes;
2. taking 65% of all samples as a training set of the model for model training; taking the remaining 35% as a test set of the model for evaluating the training result of the model;
3. the model training is divided into three stages;
stage one: training the samples by using a logistic regression model, and using a training set full-scale sample, namely the characteristic X1 which comprises C1 but does not contain the systematic deletion of the user group C1;
and a second stage: obtaining a regression model score value of each sample;
and a third stage: taking the model score value of the second stage as an offset item, and further optimizing the model effect by using an offset method based on the user samples containing the characteristics X1 except C1;
the three phases fully consider the information contained in the user group C1 with the missing feature X1 and the user group without the missing feature X1;
3.2.2offset algorithm
The offset method stems from the application of a poisson regression model, in which the "exposure" can be considered as the offset to the right of the equation, i.e., the offset term;
log(E(Y|x))＝log(exposure)+θ’x
wherein E (Y | x) represents an average value of the explained variable Y when each explained variable x is fixed; exposure represents an offset; theta' represents a parameter to be estimated, and is in a vector form; x represents each explanatory variable, in vector form;
the log (exposure) is simultaneously subtracted from the left side and the right side of the formula to obtain the following formula:
wherein E (Y | x) represents an average value of the explained variable Y when each explained variable x is fixed; exposure represents an offset; theta' represents a parameter to be estimated, and is in a vector form; x represents each explanatory variable, in vector form;
in practical applications, an offset () may be used to specify a variable representing "exposure", the code being as follows:
glm(y～offset(log(exposure))+x,family＝poisson(li nk＝log))
logistic regression is a generalized linear regression, i.e., y ═ wTx + b, wherein w and b are parameters to be solved, and a logistic regression basic formula is as follows:
wherein y represents an interpreted variable; w represents a partial regression coefficient to be estimated; b is a constant term coefficient to be estimated;
after the above formula is transformed, the following results are obtained:
then, combining with the offset method, the Logistic regression model finally becomes:
wherein, w0Represents an offset variable; λ is the corresponding coefficient, and the value is 1;
when running the logistic regression model, the variable representing "exposure" is specified with offset (), when the code is as follows:
glm(y～offset(logit(exposure))+x,family＝binomial(link＝“logit”))
the exposure is the model result based on the information of the full amount of samples except the missing feature X1, and the user sample without the missing feature X1 can improve the model effect based on the exposure.
The invention has the following beneficial effects:
the invention combines the logistic regression and offset method to solve the problem that systematic characteristic data are missing in a part of data sources in a certain time period, and builds a default model in steps and carries out instance verification. From the comparison of the results, the stability and the accuracy of the model are obviously improved after the offset method is applied.
Drawings
FIG. 1 is a flow chart of model building according to the present invention.
Detailed Description
The invention is further illustrated by the following figures and examples.
As shown in fig. 1, the implementation steps of the present invention are as follows:
step 1, confirming default factors influencing customers, including operator data, consumption behavior data, loan application information, terminal equipment information, active geographic position information, credit score and the like.
Step 2, collecting, converting, quantifying and deriving related data;
step 3, constructing a model to carry out iteration and operation;
and 4, verifying the effect.
The confirmation of the factor affecting the default of the customer in the step 1 is specifically realized as follows:
the operator data comprises call details, short message details, recharging records and the like of the application client, and because the records of the operator data are numerous and disordered, corresponding fields, such as the number of calls, the duration and the proportion of the number of night calls in the last X months, need to be further integrated and extracted. In the process of building the wind control model, related data needs to be fully mined so as to improve the accuracy of the model.
The consumption behavior data mainly refers to on-line consumption data, including the consumption amount of shopping and the types of consumed commodities each month. By using these data, features reflecting the income level and consumption habits of the customer can be extracted.
The loan application information refers to application information of a client on other loan platforms, such as the number of applications and historical loan situations on a plurality of platforms at present.
The terminal device information comprises a device type, an app list, various app activeness and the like.
The credit scores of the clients comprise online and offline consumption stages, sesame credit scores, financing lease credit conditions and the like.
The acquisition, transformation, quantification and derivation of the related data in the step 2 are specifically realized as follows:
after establishing the dimensions that may affect the customer's default in step 1, the corresponding data is transformed and quantified, and derivative processed to generate variables that may help explain the model and have commercial significance.
And 3, iterating and calculating the constructed model, specifically realizing the following steps:
3.1 feature engineering includes processing of outliers and missing values of features, data transformation, and feature selection.
①, removing the sample of abnormal value of the basic characteristic item of the trading platform, and filling the case that part of the characteristic item is missing.
②, for the record that the multi-head feature item of the third party does not match, no filling is done, and the algorithm adopted subsequently considers the null value problem.
③ the operator data provides related call detail list, short message detail list and charging record, the user generates new derived variable around the call time length, call time and calling and called conditions in different time periods, generates new derived variable around the number of short messages, receiving and sending conditions, and generates new derived variable according to charging mode, amount and frequency.
3.2 construction of models
3.2.1 this project model training description:
1. based on the original characteristics and the derivative characteristics generated in the characteristic engineering step, the project adopts multiple characteristic combinations to construct multiple models, and finally selects the optimal model through multiple evaluation indexes.
2. Taking 65% of all samples as a training set of the model for model training; the remaining 35% was used as a test set of models to evaluate the training results of the models.
3. The model training is divided into three stages.
Stage one: training the samples using a logistic regression model (using training set full-scale samples, i.e., feature X1 including C1 but not including the systematic absence of user group C1);
and a second stage: obtaining a regression model score value of each sample;
and a third stage: with the model score value of phase two as the offset term, based on the user samples containing the features X1 except C1, the offset method is used to further optimize the model effect.
The three phases fully consider the information contained in the user group C1 with the missing feature X1 and the user group without the missing feature X1;
3.2.2offset algorithm principle
The offset method stems from the application of a poisson regression model.
The Poisson Regression Model is one of the generalized linear models (generalized linear Model) with logarithmic change as the connecting function (canonical function), and one of the assumptions of this Model is that its interpreted variables obey Poisson distribution.
The poisson distribution may also be applied to ratio data, i.e. the ratio of the number of occurrences of an event to its measurement time or measurement range. For example, a biologist measures the number of tree species in a forest and the ratio variable is the number of tree species per square kilometer. The population-scientists are concerned with the number of human deaths per population year (person-year). Generally, the ratio variable expresses the number of times the event occurs per unit time. In these examples, the variables "square meters" and "population years" are the so-called "Exposure" (Exposure).
In Poisson's regression, the "exposure" can be viewed as the offset to the right of the equation, i.e., the offset term.
log(E(Y|x))＝log(exposure)+θ'x
Wherein E (Y | x) represents an average value of the explained variable Y when each explained variable x is fixed; exposure represents an offset; theta' represents a parameter to be estimated, and is in a vector form; x represents each explanatory variable, in vector form.
The log (exposure) is simultaneously subtracted from the left side and the right side of the formula to obtain the following formula:
wherein E (Y | x) represents an average value of the explained variable Y when each explained variable x is fixed; exposure represents an offset; theta' represents a parameter to be estimated, and is in a vector form; x represents each explanatory variable, in vector form.
In practical applications, such as when running a generalized linear model in R software, the variable representing "exposure" can be specified with offset (), and the code is as follows:
glm(y～offset(log(exposure))+x,family＝poisson(li nk＝log))
logistic regression is also a generalized linear regression (generalized linear model) in which y is wTx + b, where w and b are the parameters to be solved. Logistic regression transforms the interpreted variables by dividing the probability of impending occurrence by the probability of non-occurrence and taking the logarithm. This transformation changes the contradiction between the left and right value intervals of the equation and the curve relationship between the dependent variable and the independent variable. The reason for this is that the probability of occurrence and non-occurrence becomes a ratio, which is a buffer, and the value range is expanded, and then logarithmic transformation is performed, and the whole dependent variable is changed.
The logistic regression basic formula is as follows:
wherein y represents an interpreted variable; w represents a partial regression coefficient to be estimated; b is the constant term coefficient to be estimated.
After the above formula is transformed, the following results are obtained:
then, combining with the offset method, the Logistic regression model finally becomes:
wherein, w0Represents an offset variable; λ is the corresponding coefficient, typically 1.
When running the logistic regression model in R, the variable representing "exposure" can be specified with offset (), when the code is as follows:
glm(y～offset(logit(exposure))+x,family＝binomial(link＝“logit”))
in the invention, the exposure is the model result obtained based on the information of the full amount of samples except the missing characteristic X1, and the user sample without the missing characteristic X1 further improves the model effect on the basis of the exposure.
3.2.3 evaluation index adopted by the project model
The evaluation of the model effect of the project is based on AUC and KS of the model before and after the application of the offset method.
3.2.4offset algorithm boosting effect
In stage one, the model based on the logistic regression algorithm works as follows:
on the basis of logistic regression, a characteristic variable X is added, and a model obtained by applying the offset method is represented as follows:
from the comparison of the results, the stability and the accuracy of the model are obviously improved after the offset method is applied.
Claims (2)
1. An OFFSET method based on logistic regression is characterized by comprising the following steps:
step 1, confirming default factors influencing customers, including operator data, consumption behavior data, loan application information, terminal equipment information, active geographic position information, credit score and the like;
step 2, collecting, converting, quantifying and deriving related data;
step 3, constructing a model to carry out iteration and operation;
step 4, effect verification;
the confirmation of the factor affecting the default of the customer in the step 1 is specifically realized as follows:
the operator data comprises a call detail sheet, a short message detail sheet and a recharging record of the application client;
the consumption behavior data mainly refers to on-line consumption data, including the consumption amount of shopping and the types of consumed commodities each month; extracting features reflecting the income level and the consumption habits of the customers through the online consumption data;
the loan application information refers to application information of a client on other loan platforms, and currently, the number of applications and historical loan conditions are applied on a plurality of platforms;
the terminal device information comprises a device type, an app list and various app liveness degrees;
the credit scores of the clients comprise online and offline consumption stages, sesame credit scores and financing lease credit conditions;
the step 2of collecting, converting, quantifying and deriving the relevant data means that after establishing dimensions that may affect the default of the customer in the step 1, the corresponding data is converted and quantified, and the variables required by the generation are derived and processed.
2. The OFFSET method based on logistic regression as claimed in claim 1, wherein the model constructed in step 3 is iterated and operated, specifically implemented as follows:
3.1 the characteristic engineering comprises the processing of abnormal values and missing values of the characteristics, data transformation and characteristic selection;
①, removing samples of abnormal values of basic feature items of the trading platform, and filling the missing situation of part of feature items;
②, for the record that the third party multi-head feature item is not matched with, no filling is done, and the null value problem is considered in the algorithm adopted subsequently;
③ the operator data provides related call detail, short message detail and recharge records, the user generates new derived variables around the call time, call time and calling and called conditions, new derived variables around the short message quantity and receiving and sending conditions, and new derived variables according to recharge mode, money amount and frequency, thereby obtaining derived characteristics;
3.2 construction of models
3.2.1 model training description:
1. based on the original characteristics and the derivative characteristics generated in the characteristic engineering step, constructing a plurality of models by adopting a plurality of characteristic combinations, and finally selecting an optimal model through a plurality of evaluation indexes;
2. taking 65% of all samples as a training set of the model for model training; taking the remaining 35% as a test set of the model for evaluating the training result of the model;
3. the model training is divided into three stages;
stage one: training the samples by using a logistic regression model, and using a training set full-scale sample, namely the characteristic X1 which comprises C1 but does not contain the systematic deletion of the user group C1;
and a second stage: obtaining a regression model score value of each sample;
and a third stage: taking the model score value of the second stage as an offset item, and further optimizing the model effect by using an offset method based on the user samples containing the characteristics X1 except C1;
the three phases fully consider the information contained in the user group C1 with the missing feature X1 and the user group without the missing feature X1;
3.2.2offset algorithm
The offset method stems from the application of a poisson regression model, in which the "exposure" can be considered as the offset to the right of the equation, i.e., the offset term;
log(E(Y|x))＝log(exp osure)+θ′x
wherein E (Y | x) represents an average value of the explained variable Y when each explained variable x is fixed; exposure represents an offset; theta' represents a parameter to be estimated, and is in a vector form; x represents each explanatory variable, in vector form;
the log (exposure) is simultaneously subtracted from the left side and the right side of the formula to obtain the following formula:
wherein E (Y | x) represents an average value of the explained variable Y when each explained variable x is fixed; exposure represents an offset; theta' represents a parameter to be estimated, and is in a vector form; x represents each explanatory variable, in vector form;
in practical applications, an offset () may be used to specify a variable representing "exposure", the code being as follows:
glm(y～offset(log(exposure))+x,family＝poisson(li nk＝log))
logistic regression is a generalized linear regression, i.e., y ═ wTx + b, wherein w and b are parameters to be solved, and a logistic regression basic formula is as follows:
wherein y represents an interpreted variable; w represents a partial regression coefficient to be estimated; b is a constant term coefficient to be estimated;
after the above formula is transformed, the following results are obtained:
then, combining with the offset method, the Logistic regression model finally becomes:
wherein, w0Represents an offset variable; λ is the corresponding coefficient, and the value is 1;
when running the logistic regression model, the variable representing "exposure" is specified with offset (), when the code is as follows:
glm(y～offset(logit(exposure))+x,family＝binomial(link＝“logit”))
the exposure is the model result based on the information of the full amount of samples except the missing feature X1, and the user sample without the missing feature X1 can improve the model effect based on the exposure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910877298.7A CN110688373A (en) | 2019-09-17 | 2019-09-17 | OFFSET method based on logistic regression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910877298.7A CN110688373A (en) | 2019-09-17 | 2019-09-17 | OFFSET method based on logistic regression |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110688373A true CN110688373A (en) | 2020-01-14 |
Family
ID=69109373
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910877298.7A Pending CN110688373A (en) | 2019-09-17 | 2019-09-17 | OFFSET method based on logistic regression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110688373A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111523798A (en) * | 2020-04-21 | 2020-08-11 | 武汉市奥拓智能科技有限公司 | Automatic modeling method, device and system and electronic equipment thereof |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2839943A1 (en) * | 2011-07-06 | 2013-01-10 | Fred Bergman Healthcare Pty Ltd | Improvements relating to event detection algorithms |
| CN103260624A (en) * | 2010-12-07 | 2013-08-21 | 泰华制药工业有限公司 | Use of laquinimod for reducing fatigue, improving functional status, and improving quality of life in multiple sclerosis patients |
| KR20170137514A (en) * | 2016-06-03 | 2017-12-13 | 서상훈 | Personal credit rating device, method and computer programs based on psychometric data |
| CN108154430A (en) * | 2017-12-28 | 2018-06-12 | 上海氪信信息技术有限公司 | A kind of credit scoring construction method based on machine learning and big data technology |
| CN108256691A (en) * | 2018-02-08 | 2018-07-06 | 成都智宝大数据科技有限公司 | Refund Probabilistic Prediction Model construction method and device |
| CN109063931A (en) * | 2018-09-06 | 2018-12-21 | 盈盈（杭州）网络技术有限公司 | A kind of model method for predicting freight logistics driver Default Probability |
| CN109636591A (en) * | 2018-12-28 | 2019-04-16 | 浙江工业大学 | A kind of credit scoring card development approach based on machine learning |
-
2019
- 2019-09-17 CN CN201910877298.7A patent/CN110688373A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103260624A (en) * | 2010-12-07 | 2013-08-21 | 泰华制药工业有限公司 | Use of laquinimod for reducing fatigue, improving functional status, and improving quality of life in multiple sclerosis patients |
| CA2839943A1 (en) * | 2011-07-06 | 2013-01-10 | Fred Bergman Healthcare Pty Ltd | Improvements relating to event detection algorithms |
| KR20170137514A (en) * | 2016-06-03 | 2017-12-13 | 서상훈 | Personal credit rating device, method and computer programs based on psychometric data |
| CN108154430A (en) * | 2017-12-28 | 2018-06-12 | 上海氪信信息技术有限公司 | A kind of credit scoring construction method based on machine learning and big data technology |
| CN108256691A (en) * | 2018-02-08 | 2018-07-06 | 成都智宝大数据科技有限公司 | Refund Probabilistic Prediction Model construction method and device |
| CN109063931A (en) * | 2018-09-06 | 2018-12-21 | 盈盈（杭州）网络技术有限公司 | A kind of model method for predicting freight logistics driver Default Probability |
| CN109636591A (en) * | 2018-12-28 | 2019-04-16 | 浙江工业大学 | A kind of credit scoring card development approach based on machine learning |
Non-Patent Citations (1)
| Title |
|---|
| PIAOYU94: "R offset", 《HTTPS://BLOG.CSDN.NET/PIAOYU94/ARTICLE/DETAILS/80275610?IVK_SA=1024320U》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111523798A (en) * | 2020-04-21 | 2020-08-11 | 武汉市奥拓智能科技有限公司 | Automatic modeling method, device and system and electronic equipment thereof |
| CN111523798B (en) * | 2020-04-21 | 2023-09-01 | 武汉市奥拓智能科技有限公司 | Automatic modeling method, device, system and electronic equipment thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11397955B2 (en) | System and method for estimating CO2 emission | |
| US20140310059A1 (en) | System , method and computer program forecasting energy price | |
| CN110634021A (en) | Big data based vehicle estimation method, system, device and readable storage medium | |
| CN112990583B (en) | Method and equipment for determining model entering characteristics of data prediction model | |
| CN110689427A (en) | Consumption stage default probability model based on survival analysis | |
| CN105488163A (en) | Information pushing method and apparatus | |
| CN113449753B (en) | Service risk prediction method, device and system | |
| CN108416619B (en) | Consumption interval time prediction method and device and readable storage medium | |
| CN114723481A (en) | Data processing method and device, electronic equipment and storage medium | |
| CN110688373A (en) | OFFSET method based on logistic regression | |
| EP3764310A1 (en) | Prediction task assistance device and prediction task assistance method | |
| CN110544165B (en) | Credit risk score card creating method and device and electronic equipment | |
| Belomestny et al. | Semitractability of optimal stopping problems via a weighted stochastic mesh algorithm | |
| CN104850711B (en) | A kind of electromechanical product design standard system of selection | |
| CN115170304B (en) | Method and device for extracting risk feature description | |
| CN110956318A (en) | Method and device for predicting based on pre-constructed prediction model | |
| CN107562695B (en) | Load data processing method and device for distribution transformer | |
| CN113343577B (en) | Parameter optimization method, device, equipment and medium based on machine learning | |
| CN112950279A (en) | Accurate marketing strategy model construction method and device based on machine learning | |
| CN114548620A (en) | Logistics punctual insurance service recommendation method and device, computer equipment and storage medium | |
| CN118195280B (en) | Integrated task scheduling method, system and device based on integrated nesting model | |
| CN115482112A (en) | Data processing method, data processing device, computer equipment and storage medium | |
| CN116957397A (en) | Customer emotion perception method, device, equipment and storage medium | |
| CN115204508A (en) | Object prediction method and device and electronic equipment | |
| CN115345640A (en) | Product demand prediction method, product demand prediction device, computer equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200114 |