CN113055208B - Method, device and equipment for identifying information identification model based on transfer learning - Google Patents
Method, device and equipment for identifying information identification model based on transfer learning Download PDFInfo
- Publication number
- CN113055208B CN113055208B CN201911377177.2A CN201911377177A CN113055208B CN 113055208 B CN113055208 B CN 113055208B CN 201911377177 A CN201911377177 A CN 201911377177A CN 113055208 B CN113055208 B CN 113055208B
- Authority
- CN
- China
- Prior art keywords
- logistic regression
- information
- service data
- regression model
- model
- 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.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
- H04L41/14—Network analysis or design
- H04L41/147—Network analysis or design for predicting network behaviour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Evolutionary Computation (AREA)
- Pure & Applied Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Algebra (AREA)
- Operations Research (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The embodiment of the invention discloses a method, a device, equipment and a storage medium for identifying an information identification model based on transfer learning. The method comprises the following steps: acquiring service data of a plurality of areas; respectively extracting first characteristic information in the service data of a plurality of areas; extracting the same second characteristic information in the first characteristic information of the multiple regions by adopting a preset characteristic extraction analysis model; training based on the service data including the second characteristic information in the plurality of regions to obtain a common parameter; and migrating the common parameters to a preset second logistic regression model of the region, and identifying and calculating the user carrying rate of each region. According to the information identification model identification method, device, equipment and storage medium based on transfer learning provided by the embodiment of the invention, the common parameters are transferred to the characteristic parameters of each region, so that the workload and the timeliness of each region in the process of respectively establishing the identification model are reduced, and the accuracy of the identification model identification process is improved.
Description
Technical Field
The present invention relates to the field of mobile communications technologies, and in particular, to a method, an apparatus, a device, and a storage medium for identifying an information identification model based on transfer learning.
Background
At present, the popularization of users in the communication industry is gradually saturated, users with newly increased stock are mainly realized by digging opponent users, and with the implementation of number portability, in order to prevent the number portability service from impacting the current market of operators, a mathematical model is urgently needed to be established to predict potential carry-out early warning users and maintain the system in advance.
However, at present, because the data volume of the carrying-out users in each region (province) is small, the model cannot fully learn the characteristics of the carrying-out users, the model is easy to have the phenomenon of under-fitting, and the prediction effect is not ideal.
If expert scoring is carried out on the potential carry-out early warning users, the effect is quite inaccurate due to the fact that the dependency on service personnel is high and the influence of a large subjective factor exists.
Therefore, when the prediction is carried out on each region independently at present, due to the fact that models need to be established independently, the problems that carrying-out user identification models are established repeatedly, workload is large, time consumption is long, and effects are poor exist.
Disclosure of Invention
The embodiment of the invention provides an information identification model identification method, device, equipment and storage medium based on transfer learning, which can establish a general model according to the characteristic information of carrying-out users in each region and output general parameters, then transfer the general parameters into each region model comprising the characteristic information of each region and optimize the general parameters, identify the probability of potential easy carrying-out of the users in each region by using the optimized model, reduce the workload in the process of respectively establishing identification models in each region and improve the accuracy of the identification model identification process.
In a first aspect, a training method for an information recognition model based on transfer learning is provided, the information recognition model being used for recognizing the probability of potential portability of a user, and the method includes:
acquiring service data of a plurality of areas;
respectively extracting first characteristic information in the service data of the plurality of areas;
extracting the same second characteristic information in the first characteristic information of the multiple regions by adopting a preset characteristic extraction analysis model;
training based on the service data including the second feature information in the plurality of regions to obtain a first logistic regression model, wherein the first logistic regression model includes output parameters;
and respectively transferring the output parameters of the first logistic regression model to a preset second logistic regression model of at least one region, and respectively optimizing the preset second logistic regression model of the corresponding region based on the service data of at least one region to obtain the optimized at least one second logistic regression model.
In some implementations of the first aspect, the extracting first feature information in the service data of the multiple regions respectively includes:
at least one piece of first characteristic information in the service data of the plurality of areas is extracted respectively.
In some implementations of the first aspect, the extracting at least one piece of first feature information in the service data of the plurality of areas respectively includes:
respectively acquiring at least one piece of characteristic information in the service data of a plurality of areas;
sorting according to the importance of at least one characteristic information;
and acquiring at least one piece of first characteristic information with larger importance after the first characteristic information is sorted according to the importance.
In some implementations of the first aspect, the extracting first feature information in the service data of the plurality of areas respectively includes:
and respectively extracting first characteristic Information in the service data of the plurality of areas by adopting an Information Value (IV) analysis algorithm, a recursive characteristic algorithm or a random forest algorithm.
In some implementations of the first aspect, training based on the service data including the second feature information in the plurality of regions to obtain the first logistic regression model includes:
training based on the service data including the second feature information in the plurality of regions to obtain an initial logistic regression model;
and optimizing parameters corresponding to the first characteristic information in the initial logistic regression model based on the first characteristic information in the service data of the multiple regions to obtain a first logistic regression model.
In some implementation manners of the first aspect, based on first feature information in the service data of the multiple regions, a parameter corresponding to the first feature information in the initial logistic regression model is optimized to obtain a first logistic regression model, and the following formula and relationship are satisfied:
wherein l (θ) is an evaluation value, y i As the behavior information of the users included in the service data of the plurality of areas,
X＝[x 0 ,x 1 ,x 2 ...x n ]w = [ W ] for first characteristic information in traffic data of a plurality of regions 0 ,w 1 ,w 2 ...w n ]The parameters are corresponding to the first characteristic information in the initial logistic regression model;
adjusting the parameter W corresponding to the first characteristic information, and when the corresponding l (theta) obtains the maximum value, adjusting the parameter W p Is an output parameter.
In some implementations of the first aspect, the preset second logistic regression models of the plurality of regions respectively include characteristic parameters of corresponding regions; respectively transferring the output parameters of the first logistic regression model to a preset second logistic regression model of at least one region, and respectively optimizing the preset second logistic regression model of the corresponding region based on the service data of the at least one region to obtain the optimized at least one second logistic regression model, wherein the method comprises the following steps:
and respectively optimizing the output parameters in the preset second logistic regression models of the corresponding regions and the characteristic parameters of the corresponding regions based on the service data of at least one region to obtain at least one optimized second logistic regression model.
In a second aspect, a method for identifying an information identification model based on transfer learning is provided, and the method includes:
acquiring first characteristic information in service data of an area to be identified;
and obtaining the potential portability probability of the user corresponding to the service data based on the first characteristic information and the second logistic regression model of the region.
In some implementations of the second aspect, the method further comprises:
and when the probability of the potential portability meets a preset threshold value, determining the user as the potential portable user.
In a third aspect, a training apparatus for information recognition model based on transfer learning is provided, the apparatus includes:
the acquisition module is used for acquiring the service data of a plurality of areas;
the processing module is used for respectively extracting first characteristic information in the service data of the plurality of areas;
the processing module is further used for extracting the same second characteristic information in the first characteristic information of the multiple regions by adopting a preset characteristic extraction analysis model;
the processing module is further used for training based on the service data including the second feature information in the multiple regions to obtain a first logistic regression model, and the first logistic regression model comprises output parameters;
the processing module is further configured to migrate the output parameters of the first logistic regression model to a preset second logistic regression model of at least one region, and optimize the preset second logistic regression model of the corresponding region based on the service data of the at least one region to obtain the optimized at least one second logistic regression model.
In some implementations of the third aspect,
the processing module is further configured to extract at least one piece of first feature information in the service data of the multiple regions, respectively.
In some implementations of the third aspect,
the processing module is further used for respectively acquiring at least one piece of characteristic information in the service data of the plurality of areas;
the processing module is also used for sequencing according to the importance of at least one piece of characteristic information;
the processing module is further used for acquiring at least one piece of first characteristic information with larger importance after the first characteristic information is sorted according to importance.
In some implementations of the third aspect,
and the processing module is also used for respectively extracting first characteristic information in the service data of the plurality of areas by adopting an information value IV analysis algorithm, a recursive characteristic algorithm or a random forest algorithm.
In some implementations of the third aspect,
the processing module is further used for training based on the service data including the second feature information in the multiple regions to obtain an initial logistic regression model;
the processing module is further configured to optimize parameters corresponding to the first feature information in the initial logistic regression model based on the first feature information in the service data of the multiple regions, so as to obtain a first logistic regression model.
In some implementations of the third aspect,
optimizing parameters corresponding to the first characteristic information in the initial logistic regression model based on the first characteristic information in the service data of the multiple regions to obtain a first logistic regression model, wherein the first logistic regression model satisfies the following formula and relationship:
where l (θ) is the evaluation value, y i As the behavior information of the users included in the service data of the plurality of areas,
X＝[x 0 ,x 1 ,x 2 ...x n ]w = [ W ] for first characteristic information in traffic data of a plurality of regions 0 ,w 1 ,w 2 ...w n ]The parameters are corresponding to the first characteristic information in the initial logistic regression model;
when l (theta) corresponding to the adjustment parameter W is maximum, the adjusted parameter W p Is an output parameter.
In some implementations of the third aspect,
and the processing module is further used for optimizing the output parameters in the preset second logistic regression model of the corresponding region and the characteristic parameters of the corresponding region respectively based on the service data of the at least one region to obtain at least one optimized second logistic regression model.
In a fourth aspect, an apparatus for identifying a model based on information recognition of transfer learning is provided, the apparatus comprising:
the acquisition module is used for acquiring first characteristic information in the service data of the area to be identified;
and the processing module is used for obtaining the potential portable probability of the user corresponding to the service data based on the first characteristic information and the second logistic regression model of the region.
In some implementations of the fourth aspect,
and the processing module is further used for determining that the user is the potential portable user when the probability of the potential portable user meets a preset threshold value.
In a fifth aspect, an electronic device is provided, which includes: a processor and a memory storing computer program instructions;
the processor may implement the first aspect and the method for training the information recognition model based on the transfer learning in some implementations of the first aspect when executing the computer program instructions, or may implement the second aspect and the method for recognizing the information recognition model based on the transfer learning in some implementations of the second aspect when executing the computer program instructions.
In a sixth aspect, a storage medium is provided, on which computer program instructions are stored, which computer program instructions, when executed by a processor, implement the first aspect and the method for training an information recognition model based on migratory learning in some implementations of the first aspect, or, when executed by a processor, implement the second aspect and the method for recognition of an information recognition model based on migratory learning in some implementations of the second aspect.
The embodiment of the invention provides an identification method, an identification device, identification equipment and a storage medium of an information identification model based on transfer learning, which can establish a general model according to the characteristic information of carrying-out users in each area and output general parameters, then transfer the general parameters to each area model comprising the characteristic information of each area and optimize the general parameters, identify the potential carrying-out probability of the users in each area by using the optimized model, reduce the workload and time delay in the process of respectively establishing the identification model in each area, and improve the accuracy of the identification model identification process.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required to be used in the embodiments of the present invention will be briefly described below, and for those skilled in the art, other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a schematic flowchart of a training method of an information recognition model based on transfer learning according to an embodiment of the present invention;
fig. 2 is a schematic flowchart of an identification method of an information identification model based on transfer learning according to an embodiment of the present invention;
FIG. 3 is a schematic structural diagram of a training apparatus for information recognition model based on transfer learning according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of an identification apparatus for identifying a model based on information of transfer learning according to an embodiment of the present invention;
fig. 5 is a block diagram of an exemplary hardware architecture of an electronic device according to an embodiment of the present invention.
Detailed Description
Features and exemplary embodiments of various aspects of the present invention will be described in detail below, and in order to make objects, technical solutions and advantages of the present invention more apparent, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not to be construed as limiting the invention. It will be apparent to one skilled in the art that the present invention may be practiced without some of these specific details. The following description of the embodiments is merely intended to provide a better understanding of the present invention by illustrating examples of the present invention.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrases "comprising 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or apparatus that comprises the element.
At present, the popularity of the users in the communication industry tends to saturate gradually by operators, and the users with newly increased stock mainly rely on digging up the competitive environment of the opponent. Under the environment of 'number portability to network' in China, in order to reduce the loss of users to other operators for network conversion and to deal with the impact of comprehensively releasing the number portability service on the current market, a mathematical model is urgently needed to be established to predict potential carry-out early warning users and to carry out a pull-in maintenance system in advance so as to prevent the loss of the users.
The non-pilot area (province) carries out user sample data due to lack of number portability switching, mainly through analyzing the market condition of the area, determining variables and corresponding weights which possibly influence carrying out based on pilot area experience and self business experience, constructing an expert scoring model, outputting carrying-out risk user scores of each user, wherein the higher the score is, the larger the carrying-out risk is, and finally screening potential carrying-out users according to the user scores. In addition, a two-stage nested fixed-point estimation method is adopted to estimate model structure parameters according to personal monthly network information data, and a prediction result of user behavior is calculated by utilizing the estimated parameters; considering the prediction results of the plurality of prediction models and weighting to obtain a final prediction result; there are various technical solutions considering the relationship between the user's working information and the off-network, and considering that the user has no risk of off-network if the user's working information is the working information related to the mobile service field.
However, in the existing multiple technical schemes, because the data volume of the carrying-out user is small, the expert scoring has high dependency on the expert, the subjectivity is strong, the training adjustment parameters are not combined with the data, and the constructed models are mutually independent, when the individual prediction is carried out on each region, the problems of large workload, long time consumption and low recognition rate are caused by repeatedly establishing the carrying-out user recognition model.
In order to solve the problems that when each area is individually predicted in the prior art, a carrying-out user identification model needs to be repeatedly established, the workload is large, the time consumption is long, and the identification rate is low, the embodiment of the invention uses an identification method of an information identification model based on transfer learning, a general model is established according to the characteristic information of the carrying-out user in each area, general parameters are output, the general parameters are transferred to each area model comprising the characteristic information of each area and are optimized, the user easy to carry out identification by using the optimized model, the potential easy-to-carry-out probability of the user in each area is determined, the workload in the process of respectively establishing the identification model in each area is long in time, the accuracy of the identification model identification process is improved, and the problems that the workload is large, the time consumption is long, and the identification rate is low in the prior art are solved.
The embodiment of the invention provides a method, a device, equipment and a storage medium for identifying an information identification model based on transfer learning. The technical solutions provided by the embodiments of the present invention are described below with reference to the accompanying drawings.
Fig. 1 shows a flowchart of a training method of an information recognition model based on transfer learning according to an embodiment of the present invention. As shown in fig. 1, the training method of the information recognition model based on the transfer learning may include:
s101: and acquiring service data of a plurality of areas.
Specifically, since the operator is mainly divided into regions (province, city) and the user characteristics tend to a certain region, the region in the service data of the multiple regions may be province, city, or a region with obvious characteristics for the operator.
The service data may refer to feature information of the user, such as traffic information (including a time period for using the traffic and a size of the used traffic), call information (including a call duration and a time point of the call), shopping information, a keyword for searching, and the like.
S102: first characteristic information in the service data of the plurality of areas is extracted respectively.
Specifically, at least one first feature information in the service data of the plurality of regions may be extracted, the first feature information may be ranked according to the importance of the at least one feature information, and the at least one first feature information with a higher importance after ranking according to the importance may be obtained.
Specifically, an Information Value (IV) algorithm may be adopted, that is, by measuring the prediction capability of the variables, calculating and analyzing the IV Value of the feature Information, and screening out the feature Information having a better prediction capability for the target variable, so as to obtain important feature Information, that is, the first feature Information, affecting the carrying-out of the user in the service data.
The method can also adopt a recursive feature algorithm, repeatedly construct a model, put the best feature information selected each time to one side, and repeat the process on the rest feature information until all the feature information is traversed, wherein the selected sequence is the sequencing of the feature information, and can screen out the features with the top ranking as important feature information to obtain the important feature information which influences the carrying-out of the user in the service data, namely, the first feature information.
The importance of the feature information to the model can be well evaluated through the importance of the feature information output by the random forest algorithm, and the feature information with the top rank is screened out through sorting the importance of the random forest model, so that the important feature information affecting the carrying-out of the user in the business data, namely the first feature information, can be obtained.
After the important feature information affecting the carry-out of the user in the service data of the multiple regions is respectively extracted by using the acquired IV analysis algorithm, the recursive feature algorithm or the random forest algorithm, that is, after the first feature information, because the important feature information of each region is different, when the general model of each region is calculated, the important feature information affecting the carry-out of the user in each region needs to be extracted again, and the process enters S103.
S103: and extracting the same second characteristic information in the first characteristic information of the plurality of regions by adopting a preset characteristic extraction analysis model.
Specifically, a preset feature extraction analysis model can be adopted to screen out features common to all regions from important features affecting user carry-out.
S104: and training the service data including the second characteristic information in the plurality of regions to obtain a first logistic regression model.
Wherein the first logistic regression model includes an output parameter.
Specifically, an initial logistic regression model is obtained by training service data including second feature information in a plurality of regions, and parameters corresponding to first feature information in the initial logistic regression model are optimized based on first feature information in the service data of the plurality of regions to obtain a first logistic regression model, wherein the process also meets formula (1).
Specifically, the formula (1) may be:
wherein, the first and the second end of the pipe are connected with each other,
l (θ) is a log-likelihood estimation value, i.e., an evaluation value.
y i The behavior information of the user included in the service data of the plurality of areas may be specifically whether the user has taken away or whether there is a sign of taking away.
W＝[w 0 ,w 1 ,w 2 ...w n ]And the parameters are parameters corresponding to first characteristic information preset in the logistic regression model.
w T The method comprises the steps of obtaining a first characteristic information of a logistic regression model, and obtaining a transposed matrix of parameters corresponding to the first characteristic information preset in the logistic regression model.
X＝[x 0 ,x 1 ,x 2 ...x n ]Is first characteristic information in the service data of the plurality of areas.
Adjusting the parameter W corresponding to the first characteristic information, and when the log-likelihood estimation value l (theta) corresponding to the logistic regression model obtains the maximum value, adjusting the adjusted parameter W p Is an output parameter of the first logistic regression model, i.e., a characteristic parameter common to the respective regions.
S105: and respectively transferring the output parameters of the first logistic regression model to a preset second logistic regression model of at least one region, and respectively optimizing the preset second logistic regression model of the corresponding region based on the service data of at least one region to obtain the optimized at least one second logistic regression model.
Specifically, the output parameters in the preset second logistic regression model of the corresponding region and the characteristic parameters of the corresponding region are optimized respectively based on the service data of at least one region, so as to obtain at least one optimized second logistic regression model.
Although the common characteristic parameters of all the characteristic parameters of each region account for the main part, the characteristic characteristics of each region also affect the final effect of the model due to different service conditions of each region, so that the common characteristic parameters need to be migrated to the model of each region, and the output parameters of the first logistic regression model are migrated to the preset second logistic regression model of at least one region to satisfy the formula (2).
W t ＝W p +W s (2)
Wherein, the first and the second end of the pipe are connected with each other,
W t to increase the parameters of the models of the regions after the parameters are migrated.
W p Is an output parameter of the first logistic regression model, i.e., a characteristic parameter common to the respective regions.
W s Characteristic parameters specific to each region.
Specifically, after the output parameters of the first logistic regression model are respectively migrated to the preset second logistic regression model of at least one region, the process of combining the common parameters of each region with the characteristic parameters specific to each region is completed, the combined parameters are close to the model parameters of the second logistic regression model of each region, but the precision of the model parameters needs to be optimized again to ensure the identification precision of the model, the optimization process satisfies formula (3), and formula (3) is:
wherein, the first and the second end of the pipe are connected with each other,
l (θ) is a log likelihood estimation value, i.e., an evaluation value.
y i The behavior information of the user included in the service data of the area may specifically be whether the user has taken away or whether there is a sign of taking away.
W＝[w 0 ,w 1 ,w 2 ...w n ]And adding the parameters of each region model after the migration parameters are added corresponding to the first characteristic information in the second logistic regression model.
w T The method is a transpose matrix of the parameters of each area model after the parameters are migrated.
X＝[x 0 ,x 1 ,x 2 ...x n ]Is the first characteristic information in the service data of the region.
Adjusting a parameter W corresponding to the first feature information, and when the log-likelihood estimation value l (θ) corresponding to the second logistic regression model obtains a maximum value, the adjusted parameter W is a parameter of the second logistic regression model, that is, an optimized feature parameter of each region, and the parameter is used for identifying the carrying-out probability of each region user, where a specific identification method is shown in fig. 2.
Fig. 2 shows a schematic flowchart of an identification method of an information identification model based on transfer learning according to an embodiment of the present invention. As shown in fig. 2, the identification method of the information identification model based on the transfer learning may include:
s201: and acquiring service data including the first characteristic information in the area to be identified.
S202: and obtaining the potential portability probability of the user corresponding to the service data based on the service data and the second logistic regression model of the region.
Specifically, the probability that the user in the current region to be identified wants to take out is determined according to the acquired service data including the first feature information in the region to be identified and the optimized parameter of the region to be identified output by the second logistic regression model, and a calculation formula (4) in the process is as follows:
wherein p is the probability of the user being taken out.
x＝w 0 x 0 +w 1 x 1 +w 2 x 2 +...+w n x n 。
w 0 ,w 1 ,w 2 ...w n The optimized parameters W of the second logistic regression model include parameters for each feature information x in the traffic data of the region to be identified.
x 0 ,x 1 ,x 2 ...x n The characteristic information in the service data of the area to be identified.
Specifically, when the probability of potential carry-out meets a preset probability threshold, the user is determined to be a potential carry-out user, that is, when the calculated p is greater than or equal to the preset threshold, the user is considered to be a carry-out user, so that the predictive identification of the potential carry-out user is realized.
As a specific embodiment, typical feature information (first feature information) of a carry-out user may be extracted according to service data of two different regions within a period of time, where the typical feature may include basic information, binding behavior, traffic circle information, consumption behavior, different network information, consultation complaints, home group information, broadband service, payment method, network coverage, carry-over query 11 dimensions, 62 important fields, extract features (second feature information) common to two regions among the 62 features, adjust a common logistic regression model according to data including the features common to the two regions, select a parameter that maximizes log-likelihood estimation as an output parameter, combine the output parameter as a common feature parameter with characteristic feature parameters of each region, further adjust all features of each region model, and use a parameter of log-likelihood estimation when the maximum value is taken as a logistic regression parameter of each region. And (4) carrying out prediction identification on the potential carrying-out users by using the logistic regression parameters of each region, and calculating the carrying-out probability of the test set users. And selecting a proper threshold value to divide the carrying-out users by combining the market conditions of all the areas so as to determine the carrying-out conditions of the final users.
Therefore, the identification method of the information identification model based on the transfer learning provided by the embodiment of the invention can establish the general model according to the characteristic information of the carrying-out users in each area and output the general parameters, then the general parameters are transferred to each area model comprising the characteristic information of each area and optimized, the optimized model is used for identifying the potential carrying-out probability of the users in each area, the workload in the process of respectively establishing the identification model in each area is reduced, the time and the labor are prolonged, and the accuracy of the identification model identification process is improved
Corresponding to the embodiment of the training method of the information identification model based on the transfer learning and the identification method of the information identification model based on the transfer learning, the embodiment of the invention also provides a training device of the information identification model based on the transfer learning and an identification device of the information identification model based on the transfer learning.
As shown in fig. 3, fig. 3 is a schematic structural diagram of a training apparatus for information recognition model based on transfer learning according to an embodiment of the present invention.
The training device for the information recognition model based on the transfer learning can comprise: an acquisition module 301 and a processing module 302, wherein,
an obtaining module 301, configured to obtain service data of multiple areas.
The processing module 302 is configured to extract first feature information in the service data of multiple areas, respectively.
The processing module 302 is further configured to extract, by using a preset feature extraction analysis model, the same second feature information in the first feature information of the multiple regions.
The processing module 302 is further configured to train based on the service data including the second feature information in the plurality of regions to obtain a first logistic regression model, where the first logistic regression model includes an output parameter.
The processing module 302 is further configured to migrate the output parameters of the first logistic regression model to a preset second logistic regression model of at least one region, and optimize the preset second logistic regression model of the corresponding region based on the service data of the at least one region, so as to obtain the optimized at least one second logistic regression model.
The processing module 302 is further configured to extract at least one first feature information in the service data of the multiple regions, respectively.
The processing module 302 is further configured to obtain at least one feature information in the service data of the multiple regions, respectively.
The processing module 302 is further configured to rank according to importance of the at least one feature information.
The processing module 302 is further configured to obtain at least one piece of first feature information with a greater importance after being sorted according to importance.
The processing module 302 is further configured to extract first feature information in the service data of the multiple regions respectively by using an information value IV analysis algorithm, a recursive feature algorithm, or a random forest algorithm.
The processing module 302 is further configured to train based on the service data including the second feature information in the plurality of regions, so as to obtain an initial logistic regression model.
The processing module 302 is further configured to optimize, based on first feature information in the service data of the multiple regions, a parameter corresponding to the first feature information in the initial logistic regression model to obtain a first logistic regression model.
Wherein, based on the first characteristic information in the service data of a plurality of areas, the parameters corresponding to the first characteristic information in the initial logistic regression model are optimized to obtain a first logistic regression model which satisfies the following formulas and relations,
where l (θ) is the evaluation value, y i As the behavior information of the users included in the service data of the plurality of areas,
X＝[x 0 ,x 1 ,x 2 ...x n ]w = [ W ] as first characteristic information in traffic data of a plurality of zones 0 ,w 1 ,w 2 ...w n ]The parameters are parameters corresponding to the first characteristic information in the initial logistic regression model.
When l (theta) corresponding to the adjustment parameter W is maximum, the adjusted parameter W p Is an output parameter.
The processing module 302 is further configured to optimize an output parameter in a preset second logistic regression model of a corresponding region and a characteristic parameter of the corresponding region based on the service data of the at least one region, respectively, to obtain at least one optimized second logistic regression model.
As shown in fig. 4, fig. 4 is a schematic structural diagram of an identification apparatus for identifying a model based on information of transfer learning according to an embodiment of the present invention.
The recognition apparatus for recognizing a model based on information of transfer learning may include: an acquisition module 401 and a processing module 402, wherein,
the obtaining module 401 is configured to obtain first feature information in the service data of the area to be identified.
A processing module 402, configured to obtain a potential portability probability of the user corresponding to the service data based on the first feature information and the second logistic regression model of the region.
The processing module 402 is further configured to determine that the user is a potential portable user when the probability of potential portable is satisfied with a preset threshold.
The identification device of the information identification model based on the transfer learning provided by the embodiment of the invention can establish the general model according to the characteristic information of the carrying-out users in each region and output the general parameters, then transfer the general parameters to each region model comprising the characteristic information of each region and optimize the general parameters, and use the optimized model to identify the potential carrying-out probability of the users in each region, thereby reducing the workload in the process of respectively establishing the identification model in each region and prolonging the accuracy of the identification model identification process.
Fig. 5 is a block diagram illustrating an exemplary hardware architecture of an electronic device capable of implementing an information recognition model recognition method and apparatus based on transfer learning according to an embodiment of the present invention. As shown in fig. 5, the electronic device 500 includes an input device 501, an input interface 502, a central processor 503, a memory 504, an output interface 505, and an output device 506. The input interface 502, the central processing unit 503, the memory 504, and the output interface 505 are connected to each other through a bus 510, and the input device 501 and the output device 506 are connected to the bus 510 through the input interface 502 and the output interface 505, respectively, and further connected to other components of the computing device 500.
Specifically, the input device 501 receives input information from the outside and transmits the input information to the central processor 503 through the input interface 502; the central processor 503 processes input information based on computer-executable instructions stored in the memory 504 to generate output information, temporarily or permanently stores the output information in the memory 504, and then transmits the output information to the output device 506 through the output interface 505; output device 506 outputs the output information outside of computing device 500 for use by a user.
That is, the electronic device shown in fig. 5 may also be implemented as a training device based on an information recognition model for transfer learning and a recognition device based on an information recognition model for transfer learning, and the training device based on the information recognition model for transfer learning and the recognition device based on the information recognition model for transfer learning may include: a memory storing computer-executable instructions; and a processor, which when executing the computer executable instructions, may implement the training method of the information recognition model based on the transfer learning shown in fig. 1 and the recognition method of the information recognition model based on the transfer learning shown in fig. 2 in the embodiment of the present invention.
An embodiment of the present invention further provides a computer-readable storage medium, where the computer-readable storage medium stores computer program instructions; the computer program instructions, when executed by a processor, implement the training method of the information recognition model based on the transfer learning shown in fig. 1 and the recognition method of the information recognition model based on the transfer learning shown in fig. 2 according to the embodiment of the present invention.
It is to be understood that the invention is not limited to the precise arrangements and instrumentalities shown. For the sake of brevity, a detailed description of known methods is omitted in this area. In the above embodiments, several specific steps are described and shown as examples. However, the method processes of the present invention are not limited to the specific steps described and illustrated, and those skilled in the art can make various changes, modifications and additions or change the order between the steps after comprehending the spirit of the present invention.
The functional blocks shown in the above-described structural block diagrams may be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, it may be, for example, an electronic circuit, an Application Specific Integrated Circuit (ASIC), suitable firmware, plug-in, function card, or the like. When implemented in software, the elements of the invention are the programs or code segments used to perform the required tasks. The program or code segments may be stored in a machine-readable medium or transmitted by a data signal carried in a carrier wave over a transmission medium or a communication link. A "machine-readable medium" may include any medium that can store or transfer information. Examples of a machine-readable medium include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio Frequency (RF) links, and so forth. The code segments may be downloaded via computer networks such as the internet, intranet, etc.
It should also be noted that the exemplary embodiments mentioned in this patent describe some methods or systems based on a series of steps or devices. However, the present invention is not limited to the order of the above-described steps, that is, the steps may be performed in the order mentioned in the embodiments, may be performed in an order different from the order in the embodiments, or may be performed simultaneously.
As described above, only the specific embodiments of the present invention are provided, and it can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working processes of the system, the module and the unit described above may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again. It should be understood that the scope of the present invention is not limited thereto, and any person skilled in the art can easily conceive various equivalent modifications or substitutions within the technical scope of the present invention, and these modifications or substitutions should be covered within the scope of the present invention.
Claims (13)
1. A training method of an information recognition model based on transfer learning is characterized in that the information recognition model is used for recognizing the probability of potential portability of a user, and the method comprises the following steps:
acquiring service data of a plurality of areas;
respectively extracting first characteristic information in the service data of the plurality of areas;
extracting the same second characteristic information in the first characteristic information of the multiple regions by adopting a preset characteristic extraction analysis model;
training based on the service data including the second feature information in the multiple regions to obtain a first logistic regression model, wherein the first logistic regression model includes output parameters;
and respectively transferring the output parameters of the first logistic regression model to a preset second logistic regression model of at least one region, and respectively optimizing the preset second logistic regression model of the corresponding region based on the business data of at least one region to obtain at least one optimized second logistic regression model.
2. The method according to claim 1, wherein the extracting first feature information from the service data of the plurality of regions respectively comprises:
and respectively extracting at least one piece of first characteristic information in the service data of the plurality of areas.
3. The method according to claim 2, wherein the extracting at least one first feature information from the service data of the plurality of regions respectively comprises:
respectively acquiring at least one piece of characteristic information in the service data of a plurality of areas;
sorting according to the importance of the at least one characteristic information;
and acquiring at least one piece of first characteristic information with larger importance after the first characteristic information is sorted according to the importance.
4. The method according to any one of claims 1 to 3, wherein the extracting first feature information in the service data of the plurality of regions respectively comprises:
and respectively extracting first characteristic information in the service data of the plurality of areas by adopting an information value IV analysis algorithm, a recursive characteristic algorithm or a random forest algorithm.
5. The method of claim 1, wherein training based on the traffic data in the plurality of regions that includes the second feature information yields a first logistic regression model that includes:
training based on the service data including the second characteristic information in a plurality of areas to obtain an initial logistic regression model;
and optimizing parameters corresponding to the first characteristic information in the initial logistic regression model based on the first characteristic information in the service data of the plurality of regions to obtain the first logistic regression model.
6. The method according to claim 5, wherein the parameters corresponding to the first feature information in the initial logistic regression model are optimized based on the first feature information in the business data of the plurality of regions to obtain the first logistic regression model, and the following formula and relationship are satisfied:
wherein the content of the first and second substances,
as an evaluation value, a value of,
is that the
The likelihood function of (a) is,
for the behavior information of the users included in the service data of the plurality of areas,
，X=[
...
]w = [ 2 ] as first feature information in the service data of the plurality of regions
...
]The number n is the number of the first characteristic information in the service data of the plurality of areas;
adjusting the parameter W corresponding to the first characteristic information when corresponding
Adjusted parameter W when maximum value is obtained p Is the output parameter.
7. The method according to claim 1, wherein the preset second logistic regression models of the plurality of regions respectively include characteristic parameters of the corresponding regions; the migrating the output parameters of the first logistic regression model to a preset second logistic regression model of at least one region, and optimizing the preset second logistic regression model of the corresponding region based on the service data of at least one region to obtain at least one optimized second logistic regression model includes:
and optimizing the output parameters in the preset second logistic regression model of the corresponding region and the characteristic parameters of the corresponding region respectively based on the service data of at least one region to obtain at least one optimized second logistic regression model.
8. A recognition method of an information recognition model based on transfer learning is characterized by comprising the following steps:
acquiring first characteristic information in service data of an area to be identified;
and obtaining the potential portable probability of the user corresponding to the service data based on the first feature information and a second logistic regression model of the region, wherein the second logistic regression model is the second logistic regression model in the training method for the information recognition model based on the transfer learning according to any one of claims 1 to 7.
9. The method of claim 8, further comprising:
when the probability of potential carry-out meets a preset threshold, determining that the user is a potential carry-out user.
10. An apparatus for training an information recognition model based on transfer learning, the apparatus comprising:
the acquisition module is used for acquiring the service data of a plurality of areas;
the processing module is used for respectively extracting first characteristic information in the service data of the plurality of areas;
the processing module is further configured to extract the same second feature information in the first feature information of the multiple regions by using a preset feature extraction analysis model;
the processing module is further configured to train based on the service data including the second feature information in the plurality of regions to obtain a first logistic regression model, where the first logistic regression model includes output parameters;
the processing module is further configured to migrate the output parameters of the first logistic regression model to preset second logistic regression models of at least one region, and optimize the preset second logistic regression models of the corresponding regions based on the service data of the at least one region, so as to obtain at least one optimized second logistic regression model.
11. An apparatus for identifying a model based on information of transfer learning, the apparatus comprising:
the acquisition module is used for acquiring first characteristic information in the service data of the area to be identified;
a processing module, configured to obtain a potential portable probability of a user corresponding to the service data based on the first feature information and a second logistic regression model of the region, where the second logistic regression model is the second logistic regression model in the training method for information recognition models based on transfer learning according to any one of claims 1 to 7.
12. An electronic device, characterized in that the electronic device comprises: a processor and a memory storing computer program instructions;
the processor, when executing the computer program instructions, implements a training method of an information recognition model based on transfer learning according to any one of claims 1 to 7, or, when executing the computer program instructions, implements a recognition method of an information recognition model based on transfer learning according to any one of claims 8 to 9.
13. A computer-readable storage medium, wherein computer program instructions are stored on the computer-readable storage medium, and when executed by a processor, implement the training method of the information recognition model based on the transfer learning according to any one of claims 1 to 7, or when executed by the processor, implement the recognition method of the information recognition model based on the transfer learning according to any one of claims 8 to 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911377177.2A CN113055208B (en) | 2019-12-27 | 2019-12-27 | Method, device and equipment for identifying information identification model based on transfer learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201911377177.2A CN113055208B (en) | 2019-12-27 | 2019-12-27 | Method, device and equipment for identifying information identification model based on transfer learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113055208A CN113055208A (en) | 2021-06-29 |
| CN113055208B true CN113055208B (en) | 2023-01-13 |
Family
ID=76506331
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201911377177.2A Active CN113055208B (en) | 2019-12-27 | 2019-12-27 | Method, device and equipment for identifying information identification model based on transfer learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113055208B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644279A (en) * | 2016-07-21 | 2018-01-30 | 阿里巴巴集团控股有限公司 | The modeling method and device of evaluation model |
| CN108470253A (en) * | 2018-04-02 | 2018-08-31 | 腾讯科技（深圳）有限公司 | A kind of user identification method, device and storage device |
| CN108764915A (en) * | 2018-04-26 | 2018-11-06 | 阿里巴巴集团控股有限公司 | Model training method, data type recognition methods and computer equipment |
| CN109685644A (en) * | 2018-12-17 | 2019-04-26 | 深圳市数丰科技有限公司 | A kind of customers' credit methods of marking and device based on transfer learning |
| CN109767045A (en) * | 2019-01-17 | 2019-05-17 | 北京腾云天下科技有限公司 | A kind of prediction technique, device, calculating equipment and the medium of loss user |
| CN109962795A (en) * | 2017-12-22 | 2019-07-02 | 中国移动通信集团广东有限公司 | A kind of 4G customer churn method for early warning and system based on multidimensional union variable |
| CN110147803A (en) * | 2018-02-08 | 2019-08-20 | 北大方正集团有限公司 | Customer churn early-warning processing method and device |
| CN110602207A (en) * | 2019-09-12 | 2019-12-20 | 北京红山信息科技研究院有限公司 | Method, device, server and storage medium for predicting push information based on off-network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919317A (en) * | 2018-01-11 | 2019-06-21 | 华为技术有限公司 | A kind of machine learning model training method and device |
-
2019
- 2019-12-27 CN CN201911377177.2A patent/CN113055208B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107644279A (en) * | 2016-07-21 | 2018-01-30 | 阿里巴巴集团控股有限公司 | The modeling method and device of evaluation model |
| CN109962795A (en) * | 2017-12-22 | 2019-07-02 | 中国移动通信集团广东有限公司 | A kind of 4G customer churn method for early warning and system based on multidimensional union variable |
| CN110147803A (en) * | 2018-02-08 | 2019-08-20 | 北大方正集团有限公司 | Customer churn early-warning processing method and device |
| CN108470253A (en) * | 2018-04-02 | 2018-08-31 | 腾讯科技（深圳）有限公司 | A kind of user identification method, device and storage device |
| CN108764915A (en) * | 2018-04-26 | 2018-11-06 | 阿里巴巴集团控股有限公司 | Model training method, data type recognition methods and computer equipment |
| CN109685644A (en) * | 2018-12-17 | 2019-04-26 | 深圳市数丰科技有限公司 | A kind of customers' credit methods of marking and device based on transfer learning |
| CN109767045A (en) * | 2019-01-17 | 2019-05-17 | 北京腾云天下科技有限公司 | A kind of prediction technique, device, calculating equipment and the medium of loss user |
| CN110602207A (en) * | 2019-09-12 | 2019-12-20 | 北京红山信息科技研究院有限公司 | Method, device, server and storage medium for predicting push information based on off-network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113055208A (en) | 2021-06-29 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104572998A (en) | Updating method and device of question answer sequencing model for automatic question answer system | |
| CN110417607B (en) | Flow prediction method, device and equipment | |
| Guthrie | Real options analysis of climate-change adaptation: investment flexibility and extreme weather events | |
| CN105389349A (en) | Dictionary updating method and apparatus | |
| CN108509979A (en) | A kind of method for detecting abnormality, server and computer readable storage medium | |
| CN103761254A (en) | Method for matching and recommending service themes in various fields | |
| CN104992347A (en) | Video matching advertisement method and device | |
| CN103246985A (en) | Advertisement click rate predicting method and device | |
| CN108681751B (en) | Method for determining event influence factors and terminal equipment | |
| US20180247247A1 (en) | Method and system for search provider selection based on performance scores with respect to each search query | |
| CN103890758A (en) | Method and apparatus for deriving composite tie metric for edge between nodes of telecommunication call graph | |
| CN108416368A (en) | The determination method and device of sample characteristics importance, electronic equipment | |
| CN111510368A (en) | Family group identification method, device, equipment and computer readable storage medium | |
| CN110427574B (en) | Route similarity determination method, device, equipment and medium | |
| CN103577547A (en) | Webpage type identification method and device | |
| CN109526027B (en) | Cell capacity optimization method, device, equipment and computer storage medium | |
| Webb et al. | How does network structure and complexity in river systems affect population abundance and persistence? | |
| CN113055208B (en) | Method, device and equipment for identifying information identification model based on transfer learning | |
| CN111294227A (en) | Method, apparatus, device and medium for neural network-based traffic prediction | |
| Almomani et al. | Selecting a good stochastic system for the large number of alternatives | |
| CN107016460A (en) | User changes planes Forecasting Methodology and device | |
| CN103425579A (en) | Safety evaluation method for mobile terminal system based on potential function | |
| CN112214675B (en) | Method, device, equipment and computer storage medium for determining user purchasing machine | |
| CN115860856A (en) | Data processing method and device, electronic equipment and storage medium | |
| CN111353015A (en) | Crowdsourcing question recommendation method, device, 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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |