CN114298758A - Neural network prediction method based on particle swarm optimization - Google Patents
Neural network prediction method based on particle swarm optimization Download PDFInfo
- Publication number
- CN114298758A CN114298758A CN202111650297.2A CN202111650297A CN114298758A CN 114298758 A CN114298758 A CN 114298758A CN 202111650297 A CN202111650297 A CN 202111650297A CN 114298758 A CN114298758 A CN 114298758A
- Authority
- CN
- China
- Prior art keywords
- neural network
- particle swarm
- prediction method
- platform
- indis
- 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
Images
Abstract
The invention discloses a neural network prediction method based on particle swarm optimization, relates to the technical field of prediction methods, establishes an INDIS platform user loss prediction index system, selects and screens important characteristic attributes of INDIS platform users by using the characteristics of a Relief filtering method, and constructs a BP neural network prediction model based on a particle swarm optimization algorithm. According to the neural network prediction method based on particle swarm optimization, INDIS industrial internet platform users are taken as research objects, desensitization data of the INDIS platform users are researched and analyzed, important characteristic attributes are selected and screened by using characteristics based on a Relief filtering method, loss early warning is made for inventory clients of the INDIS platform users by constructing a BP neural network prediction model and a particle swarm improved algorithm-based neural network prediction model, and reference value is provided for operation of the INDIS platform.
Description
Technical Field
The invention relates to the technical field of prediction methods, in particular to a neural network prediction method based on particle swarm optimization.
Background
With the rapid development of the industrial internet and the communication technology, the market competition of the industrial internet platform is increasingly fierce, how to furthest save stock customers and seize newly-added customer markets is one of the most concerned problems of the industrial internet platform, for an industrial internet platform, the loss of inventory customers brings a series of problems, such as market share reduction, marketing expense increase and profit reduction, when new customers are developed, how to save more stock customers is a vital work, according to the analysis of market departments, the cost for developing the new customers is about 5 times of the cost for maintaining the stock customers, the longer the online time of the customers is, the higher the value is, the profits brought to the industrial internet platform by the stock clients even reach 16 times of those of the newly added clients, and the marketing cost and the profit loss can be greatly reduced by reducing the client loss rate. However, customer loss cannot be completely avoided under general conditions, but the industrial internet platform can reduce the customer loss rate by designing an accurate marketing scheme.
An industrial Internet platform of a space cloud network, namely an INDIS platform, provides productive service covering the whole process of an industrial chain and integrating all elements, the INDIS platform is a new-generation industrial operating system, a new-generation platform system, system engineering of a new-generation model, autonomous controllable new-generation space cloud basic service, and a national-level cross-industry cross-field industrial Internet platform with new enabling business state characteristics, the platform gathers 34 ten thousand enterprise users in a mode that a leading enterprise drives a small and medium-sized enterprise to go to the cloud, a regional company pulls the enterprise to go to the cloud and the like, and in order to help the industrial Internet platform to early warn and retain potential lost clients in time, the invention provides a method for analyzing market business by taking industrial Internet platform clients as research objects, preprocessing client data, optimizing a neural network by utilizing a particle swarm algorithm, and constructing a lost early warning model of the DICINS industrial Internet platform clients of the space cloud network, and the feasibility and the high efficiency of the model construction method are verified through a desensitization data set.
The invention provides important characteristic attribute indexes influencing whether the customers lose or not and a customer loss prediction method, and also provides a possibility coefficient of customer loss, thereby helping an industrial Internet platform to make a retrieval marketing strategy for the customers with high possibility of loss in advance, powerfully seizing market share and reducing marketing cost.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a particle swarm optimization-based neural network prediction method, which takes INDIS industrial internet platform users as research objects, researches and analyzes desensitization data of the INDIS platform users, selects and screens important characteristic attributes by using characteristics based on a Relief filtering method, and makes loss early warning on inventory clients of the INDIS platform users by constructing a BP neural network prediction model and a particle swarm improvement algorithm-based neural network prediction model so as to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme: the neural network prediction method based on particle swarm optimization comprises the following steps:
s1, determining a BP neural network structure according to business requirements, and determining the specific layer number of the neural network and the initial value of parameters related to the number of hidden nodes;
s2, encoding the particle swarm, and establishing a mapping relation between the particle swarm and the weight and the threshold;
s3, initializing the number, speed and position of particles of the particle swarm, and setting the inertial weight and learning factor related parameters of the particle swarm algorithm;
s4, calculating a fitness function of the network, and updating the speed and the position of the particle;
s5, if the initial judgment condition is met and a good enough position or the maximum iteration number is reached, taking the population extreme value of the current particle swarm as an optimal solution, and if the initial judgment condition is not met, starting new training in the previous step;
and S6, converting the current optimal solution into a weight and a threshold corresponding to the neural network, assigning values to the parameters of the BP network by using the weight and the threshold, and performing learning training again until the initially set performance requirement is met, thereby obtaining the final prediction method.
Further optimizing the technical solution, in the step S4, the PSO-BP algorithm trains the network by using the PSO algorithm to obtain a mean square error to optimize the weight and the threshold of the BP neural network.
The technical scheme is further optimized, an electrical INDIS platform user loss prediction index system is established, information of lost customers is analyzed, targeted data sets are extracted for data analysis to construct a prediction model, and loss users and normal users are defined according to requirements on INDIS platform services.
Further optimizing the technical scheme, reasonably analyzing information of lost users and normal users, and eliminating noise data interference, platform test users and platform trial users are required to be eliminated when the data set is extracted.
Further optimizing the technical scheme, the INDIS platform characteristic attributes comprise user names, sexes, ages, user identities and user ARPU values, the characteristic attributes are measured by utilizing the correlation degree based on a Relief algorithm, the correlation degree can be regarded as the weight of each characteristic attribute, when the important characteristic attributes are selected by utilizing the Relief algorithm, a threshold value tau can be appointed, only the characteristic value corresponding to the correlation degree larger than the tau is needed to be selected, the number n of the characteristics which are needed to be selected can be appointed, and then the n characteristics with the maximum correlation degree are selected.
Further optimizing the technical scheme, the Relief uses a method of an assumed interval to evaluate the characteristic attribute by using the correlation, the assumed interval refers to the maximum distance that the decision surface can move under the condition of keeping the classification of the samples unchanged, and can be expressed as:
where M (x), H (x) refer to nearest neighbors that are homogeneous with x and non-homogeneous with x.
Further optimizing the technical scheme, setting the training set D as (x1, y1), (x2, y2), · and (xm, ym), calculating the nearest neighbor xh of the same class as xi for each sample xi, and then calculating the nearest neighbor xm of a different class from xi, so that the correlation corresponding to the attribute j is:
wherein the content of the first and second substances,
representative sample xaThe value on attribute j.
The technical scheme is further optimized, the BP neural network is divided into an input layer, a hidden layer and an output layer, the input layer receives signals from the outside, the input signals sequentially pass through the hidden layers from the input layer and then are transmitted to the output layer, and the output layer transmits the signals processed by the neural network to the outside.
Further optimizing the technical scheme, the Particle Swarm Optimization (PSO) improved INDIS platform user loss prediction method optimizes the BP network by analyzing the defects in the traditional BP network, selecting the PSO algorithm to improve the defects.
Further optimizing the technical scheme, the construction of the INDIS platform user loss prediction method based on PSO-BP mainly comprises the following main parts:
1) constructing a traditional BP neural network: determining a neural network structure, and setting related parameters such as the number of layers of the network, the number of hidden nodes and the like;
2) optimizing a BP neural network by utilizing a particle swarm algorithm: encoding the particle swarm, wherein a single particle represents the weight and the value of a threshold, and then optimizing the parameters by using a particle swarm algorithm;
3) carrying out secondary training on the BP neural network: and assigning the weight value and the threshold value obtained through PSO algorithm optimization to the BP network for secondary learning training.
Compared with the prior art, the invention provides a neural network prediction method based on particle swarm optimization, which has the following beneficial effects:
according to the neural network prediction method based on particle swarm optimization, important characteristic attributes are selected and screened by utilizing characteristics based on a Relief filtering method, loss early warning is conducted on users of an INDIS platform through constructing a neural network prediction model based on a BP (Back propagation) neural network and a particle swarm optimization-based neural network prediction model, and a reference value is provided for operation of the INDIS platform.
Drawings
Fig. 1 is a schematic diagram of a neural network prediction method based on particle swarm optimization according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example (b):
referring to fig. 1, the present invention discloses a neural network prediction method based on particle swarm optimization, which comprises the following steps:
s1, determining a BP neural network structure according to business requirements, and determining the specific layer number of the neural network and the initial value of parameters related to the number of hidden nodes;
s2, encoding the particle swarm, and establishing a mapping relation between the particle swarm and the weight and the threshold;
s3, initializing the number, speed and position of particles of the particle swarm, and setting the inertial weight and learning factor related parameters of the particle swarm algorithm;
s4, calculating a fitness function of the network, and updating the speed and the position of the particles, wherein the PSO-BP algorithm is to train the network by utilizing the PSO algorithm to obtain a mean square error to optimize the weight and the threshold of the BP neural network;
s5, if the initial judgment condition is met and a good enough position or the maximum iteration number is reached, taking the population extreme value of the current particle swarm as an optimal solution, and if the initial judgment condition is not met, starting new training in the previous step;
and S6, converting the current optimal solution into a weight and a threshold corresponding to the neural network, assigning values to the parameters of the BP network by using the weight and the threshold, and performing learning training again until the initially set performance requirement is met, thereby obtaining the final prediction method.
According to the neural network prediction method based on particle swarm optimization, important characteristic attributes are selected and screened by utilizing characteristics based on a Relief filtering method, loss early warning is conducted on users of an INDIS platform through constructing a neural network prediction model based on a BP (Back propagation) neural network and a particle swarm optimization-based neural network prediction model, and a reference value is provided for operation of the INDIS platform.
As a specific optimization scheme of the embodiment, an electrical INDIS platform user attrition prediction index system is established, a prediction model is established by analyzing information of attrited customers and extracting a targeted data set for data analysis, so that the method can help the INDIS platform to take appropriate leave marketing measures for different attrition reasons,
the traditional user churn is generally defined as customer off-network, with the continuous development of the industry of the industrial internet platform and the gradual complication of the INDIS platform service, a new definition for the platform user churn is caused, the customer churn not only comprises the user logging off the platform account, but also comprises the condition that the customer does not log off the account but the ARPU value is reduced, or the customer does not log off the account but stops using the platform, and the churn user and the normal user are defined by combining the requirements on the INDIS platform service:
1) users who meet all of the following conditions are attrition users:
A. the account state of the user is a logout state and the like;
B. the user has not operated on the INDIS platform for a period of one month or more;
C. the ARPU value is less than 0, the average monthly consumption of the user is basically 0 although the user does not log off the account, and the contribution value to the platform is reduced and is considered to be lost.
2) For customers who do not lose we refer to normal customers, i.e. customers who meet all of the following conditions:
A. the account number state of the user is normal and the like;
B. the user has the actions of continuously operating on the platform, ordering and the like;
C. the ARPU value of the user for the last month is > 0.
As a specific optimization scheme of this embodiment, information of the lost user and the normal user is reasonably analyzed, and in order to eliminate noise data interference, a platform test user and a platform trial user need to be eliminated when extracting a data set.
As a specific optimization scheme of this embodiment, the pieces of INDICS platform feature attributes include a user name, a gender, an age, a user identity, and an ARPU value of a user, a large number of pieces of INDICS platform user feature attributes cause an oversize feature space of an INDICS platform user data set, which brings about two problems to the construction of a prediction method, on one hand, too many pieces of feature attributes cause a large increase in the calculation amount of a classification learning algorithm, on the other hand, a data set may contain more irrelevant features and features with weak relevance, which may reduce the accuracy of the user loss prediction method to a certain extent, so that the feature attributes are measured by using relevance based on a Relief algorithm, the relevance may be regarded as a weight of each feature attribute, when an important feature attribute is selected by using a Relief algorithm, a threshold may be specified, and only a feature value corresponding to the relevance larger than τ needs to be selected, it is also possible to specify the number n of features to be selected and then select the n features with the greatest correlation.
As a specific optimization scheme of this embodiment, the Relief evaluates the feature attribute with the correlation degree by using a method of an assumed interval, where the assumed interval refers to a maximum distance that a decision surface can move under the condition that sample classification is kept unchanged, and may be represented as:
where M (x), H (x) refer to nearest neighbors that are homogeneous with x and non-homogeneous with x.
When one feature attribute is favorable for classification, the distance of the homogeneous sample on the attribute is closer, and the distance of the heterogeneous sample on the attribute is farther, so the smaller the | | | x-h (x) | in the formula (1), the larger the | | | x-h (x) | is, the more favorable the attribute is for classification, and the assumed interval can evaluate the classification capability of each feature attribute of the users of the input platform, so that the most useful feature attribute for classification can be approximately estimated.
As a specific optimization scheme of this embodiment, let training set D be (x1, y1), (x2, y2), # and # ym, for each sample xi, calculate the nearest neighbor xh of the same class as xi, and then calculate the nearest neighbor xm of a different class from xi, then the correlation corresponding to attribute j is:
wherein the content of the first and second substances,
representative sample xaThe value on attribute j.
For discrete signature attributes:
for the continuous type feature attribute:
if the distance between xi and xh on the attribute j is smaller than the distance between xi and the nearest neighbor xm of the different category, the influence of the attribute j on the classification is larger, otherwise, the influence is smaller, and therefore the larger the value of delta j is, the stronger the influence of the attribute on the classification is.
The evaluation value of each attribute of a single sample is obtained by the formula (2), the correlation of the attribute is obtained by averaging the evaluation values of the same attribute of all samples, and the classification capability is strong when the correlation is larger.
As a specific optimization scheme of this embodiment, the BP neural network is divided into three parts, namely an input layer, a hidden layer, and an output layer, the input layer receives signals from the outside, the input signals sequentially pass through the hidden layers from the input layer, and then are transmitted to the output layer, and the output layer transmits the signals processed by the neural network to the outside.
As a specific optimization scheme of this embodiment, the method for predicting user churn of the incics platform based on Particle Swarm Optimization (PSO) is implemented by analyzing defects in a conventional BP network, optimizing the BP network by using the PSO, and improving the defects by using the PSO, so as to improve the performance of the prediction method, obtain initial values of a weight and a threshold, and further improve the performance and efficiency of the prediction method.
As a specific optimization scheme of this embodiment, in view of the above defects of the BP neural network, the present invention adopts a particle swarm algorithm to optimize the weight of the neural network, thereby constructing an INDICS platform user churn prediction method, which is an INDICS platform user churn prediction method improved based on the particle swarm algorithm (PSO), by analyzing the defects in the traditional BP network and improving the defects thereof by using the PSO algorithm, thereby improving the performance of the prediction method; the application of the gradient descent method in the traditional BP network can cause the defects of slow convergence rate, easy falling into local optimal solution and the like, and the setting of the initial parameters has obvious influence on the performance of the network; inaccurate parameter initial value assignment directly influences the performance of the neural network, increases learning time, and even easily falls into a local optimal solution.
Aiming at the problems, the particle swarm algorithm is an effective global optimization algorithm, the particle swarm algorithm is selected to optimize the BP network, so that initial values of weight and threshold are obtained, the performance and the efficiency of the prediction method are further improved, and the construction of the user loss prediction method of the INDIS platform based on PSO-BP is mainly divided into the following main parts:
1) constructing a traditional BP neural network: determining the structure of the neural network, and setting related parameters such as the number of layers of the network, the number of hidden nodes and the like.
2) Optimizing a BP neural network by utilizing a particle swarm algorithm: and encoding the particle swarm, wherein a single particle represents the value of the weight and the threshold, and then optimizing the parameters by using a particle swarm algorithm.
3) Carrying out secondary training on the BP neural network: and assigning the weight value and the threshold value obtained through PSO algorithm optimization to the BP network for secondary learning training.
The method for predicting the user loss of the INDIS platform of the PSO-BP is that a BP neural network is constructed firstly, then the BP neural network is optimized by using a PSO algorithm, then the optimal solution is restored by using a mapping relation, the weight and the threshold of the BP neural network are initialized, and secondary learning training is carried out. The value obtained after the network is optimized through the PSO algorithm is already the optimal solution in the global scope, so the learning duration can be reduced through the BP network after secondary learning training, and the defect that the traditional neural network is easy to fall into a local minimum value is effectively overcome.
The invention has the beneficial effects that: the important characteristic attributes are selected and screened by utilizing the characteristics based on the Relief filtering method, the defects in the traditional BP network are analyzed by the particle swarm algorithm-based improved INDIS platform user loss prediction method, and the defects are improved by utilizing the PSO algorithm, so that the performance of the prediction method is improved. The application of the gradient descent method in the traditional BP network can cause the defects of slow convergence rate, easy falling into local optimal solution and the like, and the setting of the initial parameters has obvious influence on the performance of the network. The particle swarm optimization is an effective global optimization algorithm, only needs to adjust a few parameters during application, and has the characteristics of simplicity and feasibility. And by constructing a neural network prediction model based on a BP neural network prediction model and a neural network prediction model based on a particle swarm optimization algorithm, loss early warning is given to INDIS platform users, and the prediction model using the PSO-BP algorithm is compared with the BP prediction model to improve the prediction accuracy, performance and efficiency.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (10)
1. The neural network prediction method based on particle swarm optimization is characterized by comprising the following steps of:
s1, determining a BP neural network structure according to business requirements, and determining the specific layer number of the neural network and the initial value of parameters related to the number of hidden nodes;
s2, encoding the particle swarm, and establishing a mapping relation between the particle swarm and the weight and the threshold;
s3, initializing the number, speed and position of particles of the particle swarm, and setting the inertial weight and learning factor related parameters of the particle swarm algorithm;
s4, calculating a fitness function of the network, and updating the speed and the position of the particle;
s5, if the initial judgment condition is met and a good enough position or the maximum iteration number is reached, taking the population extreme value of the current particle swarm as an optimal solution, and if the initial judgment condition is not met, starting new training in the previous step;
and S6, converting the current optimal solution into a weight and a threshold corresponding to the neural network, assigning values to the parameters of the BP network by using the weight and the threshold, and performing learning training again until the initially set performance requirement is met, thereby obtaining the final prediction method.
2. The particle swarm optimization-based neural network prediction method of claim 1, wherein in the step S4, the PSO-BP algorithm trains the network by using the PSO algorithm, and a mean square error is obtained to optimize the weight and the threshold of the BP neural network.
3. The particle swarm optimization-based neural network prediction method of claim 1, wherein an electrical INDIS platform user churn prediction index system is established, information of churn customers is analyzed, targeted data sets are extracted for data analysis to construct a prediction model, and churn users and normal users are defined in combination with requirements on INDIS platform services.
4. The particle swarm optimization-based neural network prediction method of claim 3, wherein information of attrition users and normal users is reasonably analyzed, and platform test users and platform trial users need to be excluded when extracting a data set in order to exclude noise data interference.
5. The particle swarm optimization-based neural network prediction method according to claim 4, wherein the INDIS platform feature attributes include a user name, a gender, an age, a user identity, and an ARPU value of a user, the relevance is selected based on a Relief algorithm to measure the feature attributes, the relevance can be regarded as a weight of each feature attribute, when the Relief algorithm is used to select an important feature attribute, a threshold τ can be specified, only feature values corresponding to the relevance larger than τ are selected, the number n of features to be selected can also be specified, and then n features with the largest relevance are selected.
6. The particle swarm optimization-based neural network prediction method of claim 5, wherein the Relief uses a method of an assumed interval to evaluate the feature attributes with correlation, and the assumed interval refers to the maximum distance that the decision surface can move under the condition of keeping the sample classification unchanged, and can be expressed as:
where M (x), H (x) refer to nearest neighbors that are homogeneous with x and non-homogeneous with x.
7. The particle swarm optimization-based neural network prediction method of claim 6, wherein the training set D is (x1, y1), (x2, y2), …, (xm, ym), for each sample xi, the nearest neighbor xh of the same class as xi is calculated, and then the nearest neighbor xm of the different class as xi is calculated, and then the correlation corresponding to the attribute j is:
wherein the content of the first and second substances,
representative sample xaThe value on attribute j.
8. The particle swarm optimization-based neural network prediction method of claim 1, wherein the BP neural network is divided into three parts, namely an input layer, a hidden layer and an output layer, the input layer receives signals from the outside, the input signals sequentially pass through the hidden layers from the input layer and then are transmitted to the output layer, and the output layer transmits the signals processed by the neural network to the outside.
9. The particle swarm optimization-based neural network prediction method of claim 8, wherein the particle swarm optimization-based INDIS platform user churn prediction method improved based on Particle Swarm Optimization (PSO) is to optimize a BP network by analyzing defects in a traditional BP network, and to improve the defects by using the PSO algorithm.
10. The particle swarm optimization-based neural network prediction method of claim 1, wherein the construction of the PSO-BP-based INDIS platform user churn prediction method mainly comprises the following main parts:
1) constructing a traditional BP neural network: determining a neural network structure, and setting related parameters such as the number of layers of the network, the number of hidden nodes and the like;
2) optimizing a BP neural network by utilizing a particle swarm algorithm: encoding the particle swarm, wherein a single particle represents the weight and the value of a threshold, and then optimizing the parameters by using a particle swarm algorithm;
3) carrying out secondary training on the BP neural network: and assigning the weight value and the threshold value obtained through PSO algorithm optimization to the BP network for secondary learning training.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111650297.2A CN114298758A (en) | 2021-12-30 | 2021-12-30 | Neural network prediction method based on particle swarm optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111650297.2A CN114298758A (en) | 2021-12-30 | 2021-12-30 | Neural network prediction method based on particle swarm optimization |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114298758A true CN114298758A (en) | 2022-04-08 |
Family
ID=80974246
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111650297.2A Pending CN114298758A (en) | 2021-12-30 | 2021-12-30 | Neural network prediction method based on particle swarm optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114298758A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105469144A (en) * | 2015-11-19 | 2016-04-06 | 东北大学 | Mobile communication user loss prediction method based on particle classification and BP neural network |
| CN107316099A (en) * | 2017-05-22 | 2017-11-03 | 沈阳理工大学 | Ammunition Storage Reliability Forecasting Methodology based on particle group optimizing BP neural network |
| CN107609583A (en) * | 2017-09-05 | 2018-01-19 | 深圳乐信软件技术有限公司 | Parameter optimization method, device, computer equipment and the storage medium of disaggregated model |
| CN110659930A (en) * | 2019-08-27 | 2020-01-07 | 深圳大学 | Consumption upgrading method and device based on user behaviors, storage medium and equipment |
-
2021
- 2021-12-30 CN CN202111650297.2A patent/CN114298758A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105469144A (en) * | 2015-11-19 | 2016-04-06 | 东北大学 | Mobile communication user loss prediction method based on particle classification and BP neural network |
| CN107316099A (en) * | 2017-05-22 | 2017-11-03 | 沈阳理工大学 | Ammunition Storage Reliability Forecasting Methodology based on particle group optimizing BP neural network |
| CN107609583A (en) * | 2017-09-05 | 2018-01-19 | 深圳乐信软件技术有限公司 | Parameter optimization method, device, computer equipment and the storage medium of disaggregated model |
| CN110659930A (en) * | 2019-08-27 | 2020-01-07 | 深圳大学 | Consumption upgrading method and device based on user behaviors, storage medium and equipment |
Non-Patent Citations (1)
| Title |
|---|
| 王泽宇: "《基于粒子群优化算法的电信客户流失预测模型的设计与实现》", 《中国优秀硕士学位论文全文数据库 (信息科技辑)》, no. 2, pages 34 - 52 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105491599B (en) | Predict the novel regression system of LTE network performance indicator | |
| Huang et al. | Applying fuzzy neural network to estimate software development effort | |
| CN107886372B (en) | Customer value discovery method | |
| CN101110699B (en) | System with network satisfaction degree estimation and early warning function and implementing method thereof | |
| KR102162765B1 (en) | System and method for automated management of customer churn based on artificial intelligence and computer program for the same | |
| Hayden et al. | Statistical methods to develop rating models | |
| CN109471982B (en) | Web service recommendation method based on QoS (quality of service) perception of user and service clustering | |
| Ramezanian | Estimation of the profiles in posteriori ELECTRE TRI: A mathematical programming model | |
| CN114817663A (en) | Service modeling and recommendation method based on class perception graph neural network | |
| CN110889560A (en) | Express delivery sequence prediction method with deep interpretability | |
| CN114584406B (en) | Industrial big data privacy protection system and method for federated learning | |
| CN113672797A (en) | Content recommendation method and device | |
| CN117453764A (en) | Data mining analysis method | |
| CN111506813A (en) | Remote sensing information accurate recommendation method based on user portrait | |
| CN114298758A (en) | Neural network prediction method based on particle swarm optimization | |
| CN111105127A (en) | Modular product design evaluation method based on data driving | |
| CN116304289A (en) | Information chain recommendation method and device for supply chain based on graphic neural network | |
| CN115187312A (en) | Customer loss prediction method and system based on deep learning | |
| JP2022120816A (en) | New product sales forecast method | |
| CN114757448A (en) | Manufacturing inter-link optimal value chain construction method based on data space model | |
| CN110232517B (en) | Mobile crowd sensing user profit selection method | |
| CN113761388A (en) | Recommendation method and device, electronic equipment and storage medium | |
| Bidyuk et al. | The algorithm for predicting the cryptocurrency rate taking into account the influence of posts of a group of famous people in social networks | |
| CN111353523A (en) | Method for classifying railway customers | |
| CN109447505A (en) | A kind of cloud manufacturing knowledge service preferred method based on user psychology behavior |
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 |