WO2020248916A1 - 一种信息处理方法及装置 - Google Patents

一种信息处理方法及装置 Download PDF

Info

Publication number
WO2020248916A1
WO2020248916A1 PCT/CN2020/094746 CN2020094746W WO2020248916A1 WO 2020248916 A1 WO2020248916 A1 WO 2020248916A1 CN 2020094746 W CN2020094746 W CN 2020094746W WO 2020248916 A1 WO2020248916 A1 WO 2020248916A1
Authority
WO
WIPO (PCT)
Prior art keywords
feature
users
user
dialing
target
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.)
Ceased
Application number
PCT/CN2020/094746
Other languages
English (en)
French (fr)
Inventor
易剑韬
苏镇秋
郑少杰
杨波
范增虎
江旻
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
WeBank Co Ltd
Original Assignee
WeBank Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by WeBank Co Ltd filed Critical WeBank Co Ltd
Publication of WO2020248916A1 publication Critical patent/WO2020248916A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0202—Market predictions or forecasting for commercial activities
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/03—Credit; Loans; Processing thereof

Definitions

  • the present invention relates to the technical field of financial technology (Fintech), in particular to an information processing method and device.
  • the manual collection method is a commonly used collection method. For example, if 50 users apply for a small and micro enterprise loan from the bank and have not repaid the loan after the expiry date, the corresponding contact of the 50 users can be set manually Ways (for example, it may include the number of calls to the collection, introduction language, language skills, etc.), and then the collection call can be made to the 50 users through the operator according to the contact information set manually. In this way, more manual intervention is usually required, and a better collection effect may not be obtained.
  • manually Ways for example, it may include the number of calls to the collection, introduction language, language skills, etc.
  • the operator may cause the user to completely The call cannot be answered, or, in some cases, even if the user is connected to the call, the user's willingness to repay is low. It can be seen that the use of the above method may lead to a higher number of calls accompanied by a higher rejection rate and a lower effective call-through rate, thereby making the efficiency of contact with the user lower.
  • the embodiment of the present invention provides an information processing method to improve the efficiency of collection from users.
  • an information processing method provided by an embodiment of the present invention includes:
  • the features include at least one first target feature and at least one second target feature.
  • the first target feature is a feature strongly related to dialing success
  • the second target feature is a feature strongly related to dialing failure.
  • the way the user makes contact includes the time of making a call to the user.
  • the value of the first target feature and the value of the second target feature can be used to measure the contact effect.
  • the dialing time that has a better contact effect for the user can be determined; that is, the above design obtains the dialing time for the user through pre-analysis Time, you can get a better contact effect, which can improve the efficiency of contact with users.
  • the determining the method of contacting the user according to the value of one or more target features corresponding to the user includes: combining the one or more target features corresponding to the user The value of is input into the target model to obtain the way to contact the user.
  • the target model is obtained by the following method: acquiring data of M users who dialed successfully and data of N users who failed to dial, each of the data of the M users and the data of the N users
  • the user’s data includes the value of at least one feature; the number of users with successful dialing and the number of users with failed dialing corresponding to the at least one feature are determined from the M users and the N users, and the The number of users who successfully dialed and the number of users who failed to dial corresponding to the at least one feature, respectively, and the one or more target features are filtered from the at least one feature; further, the M users and the number of users
  • the values of the one or more target features corresponding to the N users and the dialing results of the M users and the N users are trained to obtain the target model; the dialing results of the M users are used for Indicate the time of successful dialing, and the dialing results of the N users are used to indicate dialing failure.
  • the target model is trained by using the dialing time of the user who successfully dialed and the value of the user's corresponding features strongly related to the dialing success, so that the target model can be used to predict at which point in time the user can make a call
  • the target model is obtained by training the user who failed the call and the user's corresponding feature strongly related to the call failure, so that the target model can be used to predict whether the user can be Make a call (that is, predict that even if the user makes a call, the user will not answer or the user will not repay).
  • the target model obtained by using the above information training can make the target model have a better prediction effect Therefore, the contact efficiency can be improved, and useless dialing work can be avoided; and the time to make a call to the user is determined by using model training and model prediction, without manual intervention, so that the contact accuracy is higher.
  • filtering to obtain the one or more target features from the at least one feature includes: for the first feature in the at least one feature, dialing based on the corresponding first feature The number of successful users and the number of users who failed to dial, and the number of users who successfully dialed and the number of users who failed to dial corresponding to features other than the first feature in the at least one feature, to obtain the first The probability of dialing success corresponding to the feature and/or the probability of dialing failure corresponding to the first feature; further, if the probability of dialing success corresponding to the first feature is greater than a first preset threshold, determining the first feature The feature is the first target feature; and/or, if the probability of dial failure corresponding to the first feature is greater than a second preset threshold, it is determined that the first feature is the second target feature.
  • the probability of dialing success corresponding to at least one feature by determining the probability of dialing success corresponding to at least one feature, multiple features with a probability greater than the first preset threshold can be selected from the at least one feature as the first target feature strongly related to dialing success, correspondingly
  • the probability of dialing failure corresponding to at least one feature multiple features with a probability greater than the second preset threshold can be selected from the at least one feature as the second target feature strongly related to dialing failure.
  • the first target feature The second target feature and the second target feature can be used to measure the quality of the contact effect; that is to say, by determining the first target feature and the second target feature in the above manner, the dialing time can be analyzed from the contact effect, which can improve communication with users. The accuracy of the contact.
  • the at least one characteristic includes any one or more of basic attribute characteristics, overdue repayment characteristics, dialing behavior characteristics, voice interaction characteristics, and risk performance characteristics.
  • the training data can be made more comprehensive, thereby improving the accuracy of the target model obtained by training, and making the prediction effect of the target model better.
  • the determining a method of contacting the user according to the value of one or more target characteristics corresponding to the user includes: according to the one or more target characteristics corresponding to the user To obtain the dialing instruction value, the dialing instruction value is used to indicate the time to perform the dialing for the user or not to perform the dialing task; accordingly, if the dialing instruction value belongs to the first type of instruction value, it is determined to The time for the user to dial is the time point indicated by the dial indication value, and if the dial indication value is the second type indicator value, it is determined not to dial the user; the first type indication value includes It is assumed that the duration is divided into multiple dialing indication values obtained from a preset dialing time period, and the second type of indicator value is used to indicate that dialing is not performed within the preset dialing time period.
  • dialing indication values are obtained by dividing the preset dialing time period in advance.
  • the dialing indicator value can be used to indicate the time range of dialing.
  • the dialing indicator value is predicted by the target model, compared with the target model In terms of predicting the specific dialing time, the complexity of model prediction is reduced, which makes the model prediction more efficient; on the other hand, by dividing a large number of dialing time into dialing time intervals, the amount of data that needs to be predicted by the model can be reduced. Further improve the efficiency of model prediction.
  • an information processing device provided by an embodiment of the present invention includes:
  • the obtaining module is used to obtain the value of one or more target features corresponding to the user, the one or more target features including at least one first target feature and at least one second target feature, and the first target feature is and A feature that is strongly related to dialing success, and the second target feature is a feature that is strongly related to dialing failure;
  • the processing module is configured to determine a method of contacting the user according to the value of one or more target characteristics corresponding to the user, and the method of contacting the user includes the time of dialing the user .
  • the processing module is specifically configured to: input the value of one or more target features corresponding to the user into the target model to obtain the time for dialing the user.
  • the target model is obtained by the processing module in the following way: acquiring data of M users who dialed successfully and data of N users who failed to dial, the data of the M users and the data of the N users The data of each user in the data includes the value of at least one feature; the number of users who dial successfully and the number of users who fail to dial corresponding to the at least one feature is determined from the M users and the N users, respectively , And according to the number of users who successfully dialed and the number of users who failed to dial corresponding to the at least one feature, the one or more target features are filtered from the at least one feature; further, the M The values of the one or more target features corresponding to the two users and the N users and the dialing results of the M users and the N users are trained to obtain the target model; the values of the M users The dialing result is used to indicate the time when
  • the processing module is specifically configured to: for the first feature in the at least one feature, based on the number of users who dial successfully and the number of users who fail to dial corresponding to the first feature , And the number of successful dialing users and the number of failed dialing users corresponding to features other than the first feature in the at least one feature, to obtain the probability of dialing success corresponding to the first feature and/or the The probability of dialing failure corresponding to the first feature; if the probability of dialing success corresponding to the first feature is greater than a first preset threshold, it is determined that the first feature is the first target feature; and/or if the first feature is If the probability of dial failure corresponding to a feature is greater than the second preset threshold, it is determined that the first feature is the second target feature.
  • the at least one characteristic includes any one or more of basic attribute characteristics, overdue repayment characteristics, dialing behavior characteristics, voice interaction characteristics, and risk performance characteristics.
  • the processing module is specifically configured to: obtain a dialing instruction value according to the value of one or more target characteristics corresponding to the user, and the dialing instruction value is used to instruct to perform the execution on the user The dialing time or the dialing task is not performed; further, if the dialing instruction value belongs to the first type of indicator value, it is determined that the time for dialing the user is the time point indicated by the dialing instruction value. If the indicator value is a second type indicator value, it is determined not to dial the user; the first type indicator value includes multiple dial indicator values obtained by dividing a preset dial time period according to a preset duration, and the second type indicator The value is used to indicate that no dialing is performed within the preset dialing time period.
  • a computer-readable storage medium provided by an embodiment of the present invention includes instructions that, when run on a computer, cause the computer to execute the information processing method described in any of the above-mentioned first aspects.
  • a computer program product provided by an embodiment of the present invention, when it runs on a computer, causes the computer to execute the information processing method described in any of the foregoing first aspect.
  • FIG. 1 is a schematic flowchart of an information processing method according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of the execution process of an information processing method provided by an embodiment of the present invention.
  • FIG. 3 is a schematic structural diagram of an information processing device provided by an embodiment of the present invention.
  • FIG. 4 is a schematic structural diagram of a terminal device provided by an embodiment of the present invention.
  • FIG. 5 is a schematic structural diagram of a back-end device provided by an embodiment of the present invention.
  • Fintech refers to a new innovative technology brought to the financial field after the integration of information technology into the financial field.
  • the financial system can be improved by using advanced information technology to assist in financial operations, transaction execution and financial system improvements. Processing efficiency, business scale, and can reduce costs and financial risks.
  • the field of financial technology usually involves a large amount of data, such as loan data of the credit department, card purchase data of the sales department, maintenance data of the operation and maintenance department, etc. How to use technology to extract financial data from the large amount of data The characteristics required by the field have always been the goal pursued by the financial technology field.
  • the business volume of the credit department is usually very huge.
  • the credit database of the credit department can store tens of thousands or even hundreds of thousands of historical loan data.
  • the staff of the credit department can usually analyze these historical loan data according to a preset cycle (such as one day, one week, one month). If a piece of loan data is found to exceed the time limit, they can report to the The user corresponding to the piece of credit data makes a collection call.
  • a preset cycle such as one day, one week, one month.
  • the embodiments of the present invention provide an information processing method, which can improve the efficiency of contacting users.
  • the information processing method in the embodiment of the present invention is described below by taking a call for collection as an example. It is understandable that this embodiment is only an example of the information processing method in the present invention and does not constitute a limitation; that is, the present invention The information processing method in the embodiment can also be applied to other fields, and it is not limited here.
  • FIG. 1 is a schematic flowchart corresponding to an information processing method provided by an embodiment of the present invention, and the method includes:
  • Step 101 Obtain the value of one or more target characteristics corresponding to the user to be collected.
  • the one or more target features may include at least one first target feature and at least one second target feature.
  • the at least one first target feature may be a feature strongly related to the success of the collection call
  • the at least one second target feature may be a feature strongly related to the failure of the call call.
  • the credit department can store the user’s data in the first credit database while approving the loan to the user.
  • the first credit database can be used to store loans to the credit department But the data of users who have not yet repaid.
  • the user's data can include the user's basic attribute information, loan information, credit information, etc.
  • the basic attribute information can include the user's height, blood type, age, gender, home address, contact information, email address, etc.
  • the loan information can include the user A.
  • the credit information can include the user's historical loan information, and other departments' evaluation information on the user.
  • Table 1 is a schematic table of a possible first credit database provided by an embodiment of the present invention.
  • Table 1 Schematic of a possible first credit database
  • the value of the age feature can be 0, if the user’s age is any age between (35, 40), Then the value of the age feature can be 1; if the gender of the user is male, the value of the gender feature can be 1, if the gender of the user is female, the value of the gender feature can be 0; if the user’s credit is excellent , The value of the credit feature can be 1, if the user’s credit is good, the value of the credit feature can be 0; if the user’s blood type is 0, the value of the blood type feature can be 1, if the user’s blood type is B, the value of the blood type feature can be 2.
  • the value of the blood type feature can be 3.
  • the user data stored in the first credit database can be as shown in Table 2.
  • the first credit database can store multiple users and multiple user data, where the user’s basic attribute information and credit information can be represented by the values of multiple features. Take user A 3 as For example, the age of user A 3 is 40, the sex is male, the blood type is AB, and the credit is good. Therefore, the value of the age feature of user A 2 can be 40, the value of the gender feature can be male, and the value of the blood type feature is 40.
  • the value can be AB, and the value of the credit feature can be good; correspondingly, the loan amount and repayment date of multiple users can also be stored in the first credit database. As shown in Table 1, the loan amount of user A 3 can be 10 million, the repayment date can be 2019.07.01.
  • Table 1 is only an exemplary simple description, and the values corresponding to the features listed are only for the convenience of explaining the solution, and do not constitute a limitation on the solution. Understandably, the value of the feature can also be Expressed in other ways.
  • Table 2 Schematic of a possible first credit database
  • the credit repayment period A 3 sector can choose repayment period less than the current date and the user A 1 credit repayment period according to the first user database ⁇ A 1 equal to the current user User A 2 on the date is the user to be collected.
  • the credit department can evaluate user A 3 ’s repayment risk by analyzing user A 3 ’s credit, repayment period, and loan amount.
  • the credit department can also use user A 3 as the user to be collected.
  • the data of the user to be collected can be obtained from the bank's database.
  • the bank’s database can store the user’s data collected during the transaction between the various departments of the bank and the user. Therefore, the loan department can obtain the gender characteristics and the gender characteristics of the user A 1 to be collected from the bank’s database.
  • a credit collection features to be user 2 may be the first credit database collection user information of the user to be a 1 and a 2 is corrected or supplemented collection.
  • the real portrait of users to be collected can be portrayed, so that the prediction results of users to be collected based on comprehensive information are more accurate.
  • the age feature and gender feature corresponding to user A 3 to be collected can be obtained And the value of the credit feature, and then determine the time for the collection call to the user A 3 to be collected according to the value of the age feature, gender feature, and credit feature corresponding to the user A 3 to be collected, so that the time corresponding to the time of the collection call When you click to call user A 3 for collection, you can get a better collection result.
  • Step 102 According to the value of one or more target characteristics corresponding to the user to be collected, the time for the user to be collected to make a collection call is obtained.
  • users are treated collection process A 3 collection time to be dialed according to one collection corresponding to user A 3 values or more target features may be implemented manually, automatically or may be implemented as a server ,
  • the specific is not limited.
  • the following embodiments of the embodiments of the present invention specifically describe that the method of automatically implementing the server to obtain the time of the collection call for the user A 3 to be collected can be obtained without manual operation, and the efficiency of the collection call can be improved.
  • the collection can be one or more target value characteristic corresponding to the input user A 3 target model, whereby the user A 3 treatment time collection for collection dialed.
  • the target model can predict the dialing indication value based on the value of one or more target characteristics corresponding to the user A 3 to be collected. If the dialing indication value belongs to the first type of indication value, the dialing indication value can be At the indicated time point, the user A 3 to be collected is to be called for collection.
  • the dialing indication value is the second type of indication value
  • the user A 3 to be collected for collection may not be used for the collection call; where the first type of indication value may include dividing according to the preset duration
  • the second type of indicator value may be used to indicate not to perform a collection call within the preset dialing time period.
  • the preset dialing time period can be divided according to the preset time period to obtain the first to forty-eighth time period.
  • the time interval can be from 8:00 to 8:15
  • the second time interval can be from 8:15 to 8:30
  • the forty-seventh time interval can be from 19:30 to 19:45
  • the 48th time can be from 19:45 to 20:00.
  • the first type indicator value may include the first to forty-eighth dial indicator values corresponding to the first to forty-eighth time intervals, for example, the first dial indicator value is 1, and the second dial indicator value 2.
  • the forty-seventh dial indicator value is 47
  • the forty-eighth dial indicator value is 48
  • the second type indicator value can be any indicator different from the 48 indicator values included in the first type indicator value Value, such as 0.
  • dialing instruction value predicted by the target model is 3, it means that the best time for the user A 3 to be collected to call is any time from 8:30 to 8:45; if the dialing instruction predicted by the target model The value is 40, which means that the best time for the user A 3 to be collected for collection is any time between 18:00 and 18:15; if the dialing indication value predicted by the target model is 0, it means that there is no need for the user to be collected A 3 makes a collection call.
  • multiple dialing indication values are obtained by dividing the preset dialing time period in advance, and the dialing indication value can be used to indicate the time range of dialing.
  • the dialing indication value is predicted by the target model, which is compared with using the target The model predicts the specific dialing time, which can reduce the complexity of model prediction, thereby making the model prediction more efficient; on the other hand, by dividing a large number of dialing moments into dialing time intervals, the data required for model prediction can be reduced To further improve the efficiency of model prediction.
  • the information processing method provided by the embodiment of the present invention may include the first stage (that is, the training process of the target model) and the second stage (that is, the prediction process of the target model).
  • the above embodiments specifically describe the use of the target model.
  • the process of prediction the following describes the process of training to obtain the target model.
  • the credit department may store the data of the users who have repaid the loans and/or the data of the users who have made a collection call but have not yet repaid in the second credit database.
  • the second The credit database can store the data of users who actively repay, the data of users who have performed a collection call and repayment, and the data of users who have performed a collection call but have not repaid.
  • the data of multiple users who have made a call for collection can be obtained from the second credit database, and users who have successfully dialed for collection and users who have failed to call for collection can be determined from the 1,000 users.
  • the successful collection call may mean that the user connected to the call and repaid the payment on time or repaid the payment according to the time promised in the call, or the user received the call and completed the call on the day of the call.
  • Repayment, etc. accordingly, the failure of the collection call may mean that the user did not make the call when the call was made to the user, or the user connected the call but hung up midway, or the user connected the call but repaid the payment on time or Failure to complete the repayment on the day of calling the collection call, or fail to repay the payment at the time promised on the call, etc.
  • the success of the collection call and the failure of the collection call may also include other examples, which are not specifically limited.
  • the data of each user in the second credit database may include the value of at least one characteristic corresponding to the user, and the at least one characteristic may include basic attribute characteristics, overdue repayment characteristics, dialing behavior characteristics, voice interaction characteristics, Any one or more of the risk performance characteristics.
  • the basic attribute characteristics can be shown in Table 1 or Table 2;
  • the overdue repayment feature can be used to characterize the user’s overdue repayment behavior, for example, the user’s overdue repayment time, overdue repayment amount, and on-time repayment cooperation
  • Dialing behavior characteristics can be used to characterize the user's receipt of the call for collection, such as the time when the user is called, whether the user answers the call when the call is collected at that time, etc.
  • the voice interaction feature can be used to characterize the user's call for collection
  • the performance of the call such as the user’s willingness to repay and the emotion of repayment in the call collection
  • the risk performance characteristics can be used to characterize the behavior of the user after the call is received, for example, it can be based on whether the user has collected the call Evaluate the user's risk value by making repayment on the same day and the user's attitude when receiving the call for collection. This risk value can be used to characterize the effectiveness of the call for collection.
  • the training data can be made more comprehensive, thereby improving the accuracy of the target model obtained by training, and making the prediction effect of the target model better.
  • the data of 1000 users is obtained from the second credit database, and there are 600 users who have successfully dialed and 400 users who have failed to dial among the 1000 users, it can be determined from the 600 users who have successfully dialed.
  • the number of users with successful calls and 600 users determine the frequency of successful calls with at least one feature, and the number of users with failed calls can be collected according to at least one feature and the number of users with failed calls and 600 users have at least one feature.
  • Table 3 is a schematic table of the correspondence between at least one feature and the result of the call for collection according to an embodiment of the present invention.
  • the frequency of collecting users who have successfully dialed corresponding to feature a 1 is 0.8, which means that the data of 480 users out of 600 users who have successfully dialed have feature a 1 ; Therefore, if the frequency of calling users who failed to collect calls corresponding to feature a 1 is 0, it means that the data of each of the 400 users who failed to collect calls does not have feature a 1 .
  • the frequency of successful collection calls for a certain feature is greater, it means that the feature is more closely related to the success of collection calls, and the probability that multiple users with successful calls are more likely to have this feature;
  • This feature can be used as the first target feature strongly associated with the successful call of the collection; accordingly, if the frequency of the user whose call failed to be collected corresponding to a certain feature is higher, it means that the feature is more closely related to the failed call of the collection. Multiple users who fail to collect calls have a higher probability of having this feature; in this way, this feature can be used as a second target feature strongly associated with failed calls.
  • the embodiment of the present invention may use the following method to filter at least one feature to obtain a first target feature that is strongly associated with a successful call for collection and a second target feature that is strongly associated with a failed call for collection:
  • the number of users with feature a 2 among the 600 users who have successfully called N 11 that is, 180
  • the number of 600 users who have successfully called does not exist.
  • the number of users with feature a 2 is N 10 (that is, 420)
  • the number of users with feature a 2 among the 400 users that failed to collect calls is N 01 (ie 60)
  • the number of users is N 00 (ie 340).
  • N 11 calculates the corresponding expected value E 11, N 10 corresponding to the expected value E 10, N 01 corresponding to the expected value E 01, N 00 corresponding to the expected value E 00; N 11 corresponds to an example of the expectation, N 11
  • the corresponding expected value can be N 11 , the probability of collecting successful users (that is, the ratio of 600 users who successfully dialed to 1000 users), and the probability of feature a 2 (that is, the probability of users with feature a 2 in 1000 users). 2 wherein a number of the user is not present the ratio of the number) of the product.
  • the expected values E 11 corresponding to N 11 and N 11 and the expected values E 10 , N 01 corresponding to N 10 can be used corresponding to the expected value E 01, N 00, respectively, corresponding to the expected value E 00 wherein a first calculated chi-square value corresponding to a 2 and / or feature corresponding to a 2 second chi-square value; wherein, a 2 corresponding to a first feature card
  • the square value can be used to predict the probability that the hypothesis "the successful call is related to the feature a 2 " (that is, the probability of the successful call of the feature a 2 ) is true
  • the second chi-square value corresponding to the feature a 2 can be used Predict the probability that the hypothesis "the call failure of the collection call is related to the feature a 2 " (that is, the probability of the call failure corresponding to the feature a 2 ) is true.
  • the first chi-square value corresponding to feature a 2 can satisfy the following conditions:
  • t can be used to characterize the success or failure of the collection call. If the collection call is successful, t can be 1, and if the collection call fails, t can be 0; c can be used to characterize whether the feature a 2 exists, and if the feature a If 2 exists, c can be 1, and if feature a 2 does not exist, c can be 0.
  • the characteristic features obtained after a 2 square value corresponding to a first card a 2 corresponding to the predicted probability of success call collection if it is determined corresponding to a 2 characterized in collection call success probability greater than a first predetermined threshold value, characterized in Description a 2 and a strong correlation successful collection call, which can be characterized as a 2 a first target feature; and / or, the probability of obtaining the feature characteristic corresponding to a 2 second prediction chi-square value corresponding to a 2 call failure in collection
  • the probability of the call failure of the call for collection corresponding to the feature a 2 is greater than the second preset threshold, it indicates that the feature a 2 is strongly related to the call failure of the call for call collection, and the feature a 2 can be used as a second target feature.
  • the first preset threshold and the second preset threshold may be set by those skilled in the art based on experience.
  • the first preset threshold and the second preset threshold may be the same or different, and are not specifically limited.
  • multiple features with a probability greater than the first preset threshold can be selected from the at least one feature as the first target feature strongly related to the successful call of the call for call
  • multiple features with a probability greater than the second preset threshold can be selected from the at least one feature as the second target feature strongly related to the call failure of the call call, .
  • the first target feature and the second target feature can be used to measure the effect of the collection call; that is to say, the first target feature and the second target feature are determined by the above method, and the collection call time can be determined from the collection call effect. Analysis, which can improve the accuracy of ripening dialing.
  • the value of the 10 first target characteristics corresponding to the 600 users who have successfully dialed can be used and the value of the call
  • the result and the value of the 6 second target features and the result of the call to call for the 400 users who failed to collect calls were trained to obtain the target model.
  • the result of user R's collection call can be the dial indication value of the time interval corresponding to the time when the collection call is made to user R. For example, if the call is made to user R at 10:03 If a collection call is made and the collection call is successful, the result of user R's collection call can be 8.
  • the result of user R's collection call can be 48; accordingly, if the user If the R collection call succeeds, the result of the collection call of the user R can indicate the call indication value of the collection call failure, such as 0.
  • the value of the 10 first target features and the result of the call to the 600 users who successfully called and the value of the 6 second target features respectively corresponding to the call to the 400 users who failed to call can be collected and The call result is collected to construct the characteristic matrix.
  • the ten first target features and the six second target features may all be different; in this way, the feature matrix can be a matrix with 1000 rows and 17 columns.
  • the first 600 rows of the feature matrix can include the information of the 600 users who have successfully called (including the values of the 10 first target features and the result of the call), and the last 400 rows of the feature matrix can include the 400 failed calls.
  • User's information including the values of the 6 second target features and the result of the call to call
  • the first 10 columns of the feature matrix can include the values of the 10 first target features corresponding to 1000 users
  • the first 10 features of the feature matrix Columns 10 to 16 can include the values of the 6 second target features corresponding to 1000 users
  • column 17 of the feature matrix can include the call result of the call for 1000 users, and the call result of each user can be 0 ⁇ Any value between 48.
  • the feature matrix can be a matrix with 1000 rows and 17-i columns.
  • the first 600 rows of the feature matrix can include the information of the 600 users who have successfully called (including the values of the 10 first target features and the result of the call), and the last 400 rows of the feature matrix can include the 400 failed calls.
  • User information including the value of the 6 second target features and the call result); correspondingly, the first 10-i column of the feature matrix can include the 10 first target features and 6 second ones corresponding to 1000 users.
  • the values of 10-i first target features with different target features, the 11-i to 10th columns of the feature matrix can include the 10 first target features corresponding to 1000 users that are the same as the 6 second target features
  • the value of the i first target feature, the ith to 16-ith columns of the feature matrix can include the values of 6 second target features corresponding to 1000 users, and the 17-ith column of the feature matrix can include 1000
  • the collection call result corresponding to the user can be any value between 0 and 48.
  • the feature matrix can be a matrix with 1000 rows and 11 columns.
  • the first 600 rows of the feature matrix can include the information of the 600 users who have successfully called (including the values of the 10 first target features and the result of the call), and the last 400 rows of the feature matrix can include the 400 failed calls.
  • User information including the value of the 6 second target features and the call result
  • the first 4 columns of the feature matrix can include 10 first target features and 6 second target features corresponding to 1000 users
  • the 5th to 10th columns of the feature matrix can include 6 first target features that are the same as the 6 second target features among the 10 first target features corresponding to 1000 users
  • the value of, the 11th column of the feature matrix can include the call result of 1000 users, and the call result of each user can be any value between 0 and 48.
  • the feature matrix can be input to the deep learning platform to train the target model; where the deep learning platform can refer to the neural network deep learning platform, such as the Long Short-Term Memory (LSTM) platform, such,
  • the trained target model can be an LSTM neural network model; or, the deep learning platform can refer to a machine learning platform, such as a differential integrated moving average autoregressive model (ARIMA).
  • ARIMA differential integrated moving average autoregressive model
  • the trained target model can be Any one of linear model, regression model, and probability graph model is not specifically limited.
  • the deep learning platform can build an initial model in advance, which can include input gate nodes, output gate nodes, forget gate nodes, processing nodes, etc.; After the feature matrix is input to the deep learning platform, the input gate node in the initial model can analyze the features in the feature matrix. If a feature is determined to be useless, the feature can be transmitted to the forget gate node.
  • the feature can be transmitted to the output gate node; accordingly, the output gate node can learn the weight of the useful feature, for example, the threshold of the weight can be adjusted by using a preset activation function, and can be adjusted according to the result Update the preset activation function, the forget gate node can determine whether the useless information has a certain impact on the processing process of the exit node according to the processing result of the output gate node. If it is determined that the useless information has a certain impact on the processing process of the exit node, the useless information can be removed It is sent to the output gate node so that the output gate node updates the model. If it is determined that the useless information does not affect the processing process of the exit node, the useless information can be discarded.
  • the threshold of the weight can be adjusted by using a preset activation function, and can be adjusted according to the result Update the preset activation function
  • the forget gate node can determine whether the useless information has a certain impact on the processing process of the exit node according to the processing result of the output gate node. If
  • the output gate node can be trained to obtain 49 neuron nodes.
  • 49 neuron nodes 48 neuron nodes are used to characterize the time corresponding to the successful call and the remaining 1 neuron node is used for Indicates that the collection call failed.
  • 49 neuron nodes can be connected to multiple classifiers, which can be used to predict the probability of passing through 49 neuron nodes respectively, and then the neuron node with the highest probability can be selected from the 49 neuron nodes As the target neuron node, the dialing identifier corresponding to the target neuron node can be output to the user.
  • the user data input A 1 of the target model, multiple classifier 48 can be predicted probability respectively corresponding to the call instruction value, corresponding to 0 if the probability value indicating the maximum call, then the target model output may be 0, That target model predicted "a 1 user does not answer the phone when the user a 1 collection dial telephone collection" of information, so, may not need collection phone call to the user a 1, thereby increasing the efficiency of collection call, avoid performing unnecessary collection Dialing work; accordingly, if the probability corresponding to dialing indicator value 2 is the greatest, the output of the target model can be 2, that is, the target model predicts that the collection effect of calling user A 1 from 8:15 to 8:30 is better OK", so, you can choose to make a collection call to user A 1 at any time from 8:15 to 8:30, so that you do not need to manually determine the call time, and improve the accuracy and efficiency of the collection call.
  • the value of one or more target features corresponding to the user is obtained, and the method of contacting the user is obtained according to the value of one or more target features corresponding to the user.
  • the method of contacting the user includes the time for dialing the user; wherein the one or more target characteristics include at least one first target characteristic and at least one second target characteristic, and the first target characteristic
  • the second target feature is a feature that is strongly related to dialing success, and the second target feature is a feature that is strongly related to dialing failure.
  • the value of the first target feature and the value of the second target feature can be used to measure the contact effect
  • the dialing time with a better contact effect with the user can be determined; that is, in the embodiment of the present invention, through pre-analysis By obtaining the time for making a call to the user, a better contact effect can be obtained, so that the efficiency of contacting the user can be improved.
  • FIG. 2 is a schematic diagram of the execution process of an information processing method provided by an embodiment of the present invention. The process can be implemented with reference to the information processing method illustrated in FIG. 1.
  • the information processing method in the embodiment of the present invention may include a model training phase and a model prediction phase; in the model training phase, the feature generation device, target feature screening device, and model training The device can be trained to obtain the target model; accordingly, in the model prediction stage, by inputting the data of the user to be collected into the target model, the time to call the user to be collected can be predicted.
  • the functions of each device in the model training phase and the model prediction phase are introduced below.
  • the feature generation device can obtain at least one feature and the value of at least one feature corresponding to the M+N users based on the data of the M+N users ; Moreover, the feature generation device can also obtain the collection call results of M+N users, such as whether M+N users call collection succeeded or failed. If a user’s collection call succeeds, it can also obtain the call collection call to the user time.
  • the feature generation device may send the value of at least one feature corresponding to the M+N users and the result of the call of the M+N users to the target feature screening device.
  • the target feature screening device can select from at least one feature
  • the features that are strongly related to the success of the collection call and the features that are strongly related to the failure of the collection call can be filtered, and then the values of the features that are strongly related to the success of the collection call corresponding to the M+N users and the features strongly related to the failure of the collection call can be obtained
  • the value of is sent to the model training device.
  • the model training device can compare the model according to the value of the feature strongly related to the successful call of the collection of M+N users, the value of the feature strongly related to the failure of the call to collect the call, and the results of the call of the M+N user. Perform training to obtain the target model.
  • the target model can be based on the data of the user to be collected Predict the user to be collected, and obtain the time for calling the user to be collected.
  • the target model can determine the time to call the user to be collected by outputting the dial instruction value corresponding to the dialing time interval or the dial indicator value of not dialing.
  • the time to call the user to be collected can include not dialing to the user to be collected Or when to call the user to be collected.
  • the value of one or more target features corresponding to the user is obtained, and the method of contacting the user is obtained according to the value of one or more target features corresponding to the user.
  • the method of contacting the user includes the time for dialing the user; wherein the one or more target characteristics include at least one first target characteristic and at least one second target characteristic, and the first target characteristic
  • the second target feature is a feature that is strongly related to dialing success, and the second target feature is a feature that is strongly related to dialing failure.
  • the value of the first target feature and the value of the second target feature can be used to measure the contact effect
  • the dialing time with a better contact effect with the user can be determined; that is, in the embodiment of the present invention, through pre-analysis By obtaining the time for making a call to the user, a better contact effect can be obtained, so that the efficiency of contacting the user can be improved.
  • an embodiment of the present invention also provides an information processing device, and the specific content of the device can be implemented with reference to the above method.
  • Fig. 3 is an information processing device provided by an embodiment of the present invention, and the device includes:
  • the obtaining module 301 is configured to obtain the value of one or more target features corresponding to the user, the one or more target features including at least one first target feature and at least one second target feature, and the first target feature is A feature that is strongly related to dialing success, and the second target feature is a feature that is strongly related to dialing failure;
  • the processing module 302 is configured to determine a method of contacting the user according to the value of one or more target characteristics corresponding to the user; the method of contacting the user includes dialing the user time.
  • processing module 302 is specifically configured to:
  • the target model is obtained by the processing module in the following manner:
  • the one or more target characteristics are obtained by screening from the at least one characteristic
  • the dialing results of M users are used to indicate the time of successful dialing, and the dialing results of the N users are used to indicate dialing failure.
  • processing module 302 is specifically configured to:
  • the at least one feature For the first feature in the at least one feature, based on the number of users who dialed successfully and the number of users who failed to dial corresponding to the first feature, and the at least one feature other than the first feature The number of users who have successfully dialed and the number of users who have failed to dial corresponding to the feature is used to obtain the probability of dialing success corresponding to the first feature and/or the probability of dialing failure corresponding to the first feature; if the first feature If the corresponding probability of dialing success is greater than the first preset threshold, it is determined that the first feature is the first target feature; and/or, if the probability of dialing failure corresponding to the first feature is greater than the second preset threshold, then It is determined that the first feature is the second target feature.
  • the at least one characteristic includes any one or more of basic attribute characteristics, overdue repayment characteristics, dialing behavior characteristics, voice interaction characteristics, and risk performance characteristics.
  • processing module 302 is specifically configured to:
  • the dial indicator value belongs to the first type indicator value, it is determined that the time for dialing the user is the time point indicated by the dial indicator value, and if the dial indicator value is the second type indicator value, it is determined that it is incorrect
  • the user makes a call; the first type indicator value includes multiple dial indicator values obtained by dividing a preset dial time period according to a preset time period, and the second type indicator value is used to indicate that the dialing time period is not within the preset dial time period. Perform dialing.
  • the value of one or more target features corresponding to the user is obtained, and according to the value of one or more target features corresponding to the user, the value corresponding to the
  • the method for contacting the user includes the time for dialing the user; wherein the one or more target characteristics include at least one first target characteristic and at least one second target characteristic ,
  • the first target feature is a feature strongly related to dialing success, and the second target feature is a feature strongly related to dialing failure.
  • the value of the first target feature and the value of the second target feature can be used to measure the contact effect
  • the dialing time with a better contact effect with the user can be determined; that is, in the embodiment of the present invention, through pre-analysis By obtaining the time for making a call to the user, a better contact effect can be obtained, so that the efficiency of contacting the user can be improved.
  • the embodiments of the present invention also provide a computer-readable storage medium, including instructions, which when run on a computer, cause the computer to execute the information processing method described in any one of FIG. 1 or FIG. 2.
  • the embodiments of the present invention also provide a computer program product, which when running on a computer, causes the computer to execute the information processing method described in any one of FIG. 1 or FIG. 2.
  • an embodiment of the present invention provides a terminal device. As shown in FIG. 4, it includes at least one processor 1101 and a memory 1102 connected to the at least one processor.
  • the embodiment of the present invention does not limit the processor.
  • the specific connection medium between 1101 and the memory 1102 is, for example, the connection between the processor 1101 and the memory 1102 in FIG. 4 through a bus.
  • the bus can be divided into address bus, data bus, control bus, etc.
  • the memory 1102 stores instructions that can be executed by at least one processor 1101, and the at least one processor 1101 can execute the steps included in the foregoing information processing method by executing the instructions stored in the memory 1102.
  • the processor 1101 is the control center of the terminal device, which can use various interfaces and lines to connect various parts of the terminal device, and realize data by running or executing instructions stored in the memory 1102 and calling data stored in the memory 1102. deal with.
  • the processor 1101 may include one or more processing units, and the processor 1101 may integrate an application processor and a modem processor.
  • the application processor mainly processes the operating system, user interface, and application programs.
  • the adjustment processor mainly processes instructions issued by operation and maintenance personnel. It is understandable that the foregoing modem processor may not be integrated into the processor 1101.
  • the processor 1101 and the memory 1102 may be implemented on the same chip, and in some embodiments, they may also be implemented on separate chips.
  • the processor 1101 may be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application specific integrated circuit (ASIC), a field programmable gate array or other programmable logic devices, discrete gates or transistors Logic devices and discrete hardware components can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention.
  • the general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiment of the information processing method may be directly embodied as executed and completed by a hardware processor, or executed and completed by a combination of hardware and software modules in the processor.
  • the memory 1102 can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules.
  • the memory 1102 may include at least one type of storage medium, for example, it may include flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), magnetic memory, disk , CD, etc.
  • the memory 1102 is any other medium that can be used to carry or store desired program codes in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.
  • the memory 1102 in the embodiment of the present invention may also be a circuit or any other device capable of realizing a storage function, for storing program instructions and/or data.
  • an embodiment of the present invention provides a back-end device. As shown in FIG. 5, it includes at least one processor 1201 and a memory 1202 connected to the at least one processor.
  • the embodiment of the present invention does not limit the processing.
  • the connection between the processor 1201 and the memory 1202 in FIG. 5 is taken as an example.
  • the bus can be divided into address bus, data bus, control bus, etc.
  • the memory 1202 stores instructions that can be executed by at least one processor 1201, and the at least one processor 1201 can execute the steps included in the foregoing information processing method by executing the instructions stored in the memory 1202.
  • the processor 1201 is the control center of the back-end equipment, which can use various interfaces and lines to connect to various parts of the back-end equipment, and by running or executing instructions stored in the memory 1202 and calling data stored in the memory 1202, Realize data processing.
  • the processor 1201 may include one or more processing units, and the processor 1201 may integrate an application processor and a modem processor, where the application processor mainly processes operating systems, application programs, etc., and the modem processor Mainly analyze the received instructions and analyze the received results. It can be understood that the foregoing modem processor may not be integrated into the processor 1201.
  • the processor 1201 and the memory 1202 may be implemented on the same chip, and in some embodiments, they may also be implemented on separate chips.
  • the processor 1201 may be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application specific integrated circuit (ASIC), a field programmable gate array or other programmable logic devices, discrete gates or transistors Logic devices and discrete hardware components can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention.
  • the general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiment of the information processing method may be directly embodied as being executed and completed by a hardware processor, or executed and completed by a combination of hardware and software modules in the processor.
  • the memory 1202 as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules.
  • the memory 1202 may include at least one type of storage medium, for example, may include flash memory, hard disk, multimedia card, card-type memory, random access memory (Random Access Memory, RAM), static random access memory (Static Random Access Memory, SRAM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), magnetic memory, disk , CD, etc.
  • the memory 1202 is any other medium that can be used to carry or store desired program codes in the form of instructions or data structures and that can be accessed by a computer, but is not limited thereto.
  • the memory 1202 in the embodiment of the present invention may also be a circuit or any other device capable of realizing a storage function for storing program instructions and/or data.
  • the embodiments of the present invention may be provided as methods or computer program products. Therefore, the present invention may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.
  • a computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including the instruction device.
  • the device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram.
  • These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to produce computer-implemented processing, so as to execute on the computer or other programmable equipment.
  • the instructions provide steps for implementing functions specified in a flow or multiple flows in the flowchart and/or a block or multiple blocks in the block diagram.

Landscapes

  • Business, Economics & Management (AREA)
  • Accounting & Taxation (AREA)
  • Finance (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Marketing (AREA)
  • Economics (AREA)
  • Data Mining & Analysis (AREA)
  • Game Theory and Decision Science (AREA)
  • Technology Law (AREA)
  • Telephonic Communication Services (AREA)

Abstract

一种信息处理方法及装置，涉及金融科技技术领域，其中方法包括：获取用户对应的一个或多个目标特征的取值，其中，一个或多个目标特征包括与拨打成功强相关的特征和与拨打失败强相关的特征（101）；并根据用户对应的一个或多个目标特征的取值，得到对所述用户进行拨打的时间（102）。所述方法通过确定与拨打成功强相关的第一目标特征和与拨打失败强相关的第二目标特征，可以使用第一目标特征的取值和第二目标特征的取值衡量联系效果；也就是说，所述方法通过预先分析得到对用户进行拨打的时间，可以得到较好的联系效果，从而可以提高与用户进行联系的效率。

Description

一种信息处理方法及装置
相关申请的交叉引用
本申请要求在2019年06月11日提交中国专利局、申请号为201910501871.4、申请名称为“一种信息处理方法及装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及金融科技(Fintech)技术领域，尤其涉及一种信息处理方法及装置。
背景技术
随着计算机技术的发展，越来越多的技术应用在金融领域，传统金融行业也正在逐步向金融科技(Fintech)转变，然而，由于金融行业的安全性、实时性要求，使得金融科技领域对技术提出了更高的要求。以银行为例，银行中通常会涉及到多种类型的信贷业务，比如小微企业贷、控股企业贷、个体工商贷等。一般来说，若到达还款日期而用户还未还款，或者未到还款日期但用户按期还款的风险值较高，则银行可以向用户拨打催收电话，如此，既可以提醒用户按时还款，又可以获取用户的还款意愿，从而确定针对于该用户的后续催收计划。
人工催收方式是现有较为常用的一种催收方式，举例来说，若50名用户向银行申请了小微企业贷且超期未还款，则可以通过人工的方式设置这50名用户对应的联系方式(比如可以包括催收拨打的次数、介绍语、话术等)，进而可以通过运营商按照人工设置的联系方式向这50名用户拨打催收电话。采用该种方式，通常需要较多的人工干预，且可能并不能得到较好的催收效果，举例来说，若运营商向用户拨打电话的时间恰巧为用户的午休时间，则可能会导致用户完全接听不到电话，或者，在一些情况下，即使用户接通了电话，用户的还款意愿也较低。由此可知，采用上述方式可能会导致较高的拨打量伴随着较高的拒接率和较低的有效接通率，从而使得与用户进行联系的效率较低。
综上，目前亟需一种信息处理方法，用以提高与用户进行联系的效率。
发明内容
本发明实施例提供一种信息处理方法，用以提高向用户进行催收的效率。
第一方面，本发明实施例提供的一种信息处理方法，所述方法包括：
获取用户对应的一个或多个目标特征的取值，并根据所述用户对应的一个或多个目标特征的取值，确定向所述用户进行联系的方式；其中，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为拨打失败强相关的特征，所述向所述用户进行联系的方式包括对所述用户进行拨打的时间。
在上述设计中，通过确定与拨打成功强相关的第一目标特征和与拨打失败强相关的第二目标特征，可以使用第一目标特征的取值和第二目标特征的取值衡量联系效果，如此，通过使用用户的数据中第一目标特征的取值和第二目标特征的取值即可确定对用户联系效果较好的拨打时间；也就是说，上述设计通过预先分析得到对用户进行拨打的时间，可 以得到较好的联系效果，从而可以提高与用户进行联系的效率。
在一种可能的设计中，所述根据所述用户对应的一个或多个目标特征的取值，确定向所述用户进行联系的方式，包括：将所述用户对应的一个或多个目标特征的取值输入目标模型，得到向所述用户进行联系的方式。其中，所述目标模型是通过如下方式得到的：获取拨打成功的M个用户的数据和拨打失败的N个用户的数据，所述M个用户的数据和所述N个用户的数据中每个用户的数据包括至少一个特征的取值；从所述M个用户和所述N个用户中确定所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，并根据所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，从所述至少一个特征中筛选得到所述一个或多个目标特征；进一步地，使用所述M个用户和所述N个用户对应的所述一个或多个目标特征的取值以及所述M个用户和所述N个用户的拨打结果，训练得到所述目标模型；所述M个用户的拨打结果用于指示拨打成功的时间，所述N个用户的拨打结果用于指示拨打失败。
在上述设计中，通过使用拨打成功的用户的拨打时间和该用户对应的与拨打成功强相关的特征的取值训练得到目标模型，使得目标模型可以用于预测在哪个时间点对用户拨打电话可以获取到最好的联系效果，相应地，通过使用拨打失败的用户和该用户对应的与拨打失败强相关的特征的取值训练得到目标模型，使得目标模型可以用于预判是否可以无需对用户拨打电话(即预判即使对用户拨打电话该用户也不会接听或者该用户也不会还款)，由此可知，通过使用上述信息训练得到目标模型，可以使得目标模型具有较好的预测效果，从而可以提高联系效率，避免执行无用的拨打工作；且，通过使用模型训练和模型预测的方式确定向用户拨打电话的时间，可以无需人工干预，从而使得联系的准确度较高。
在一种可能的设计中，从所述至少一个特征中筛选得到所述一个或多个目标特征，包括：针对于所述至少一个特征中的第一特征，基于所述第一特征对应的拨打成功的用户的数量和拨打失败的用户的数量，以及所述至少一个特征中除所述第一特征以外的特征对应的拨打成功的用户的数量和拨打失败的用户的数量，得到所述第一特征对应的拨打成功的概率和/或所述第一特征对应的拨打失败的概率；进一步地，若所述第一特征对应的拨打成功的概率大于第一预设阈值，则确定所述第一特征为第一目标特征；和/或，若所述第一特征对应的拨打失败的概率大于第二预设阈值，则确定所述第一特征为第二目标特征。
在上述设计中，通过确定至少一个特征分别对应的拨打成功的概率，可以从至少一个特征中选取概率大于第一预设阈值的多个特征作为与拨打成功强相关的第一目标特征，相应地，通过确定至少一个特征分别对应的拨打失败的概率，可以从至少一个特征中选取概率大于第二预设阈值的多个特征作为与拨打失败强相关的第二目标特征，如此，第一目标特征和第二目标特征可以用于衡量联系效果的好坏；也就是说，通过如上方式确定第一目标特征与第二目标特征，可以从联系效果入手对拨打时间进行分析，从而可以提高与用户进行联系的准确率。
在一种可能的设计中，所述至少一个特征包括基本属性特征、逾期还款特征、拨打行为特征、语音交互特征、风险表现特征中的任意一项或任意多项。
在上述设计中，通过综合分析用户的各种特征，可以使得训练数据更加全面，从而提高训练得到的目标模型的准确性，使得目标模型的预测效果更好。
在一种可能的设计中，所述根据所述用户对应的一个或多个目标特征的取值，确定向所述用户进行联系的方式，包括：根据所述用户对应的一个或多个目标特征的取值，得到 拨打指示值，所述拨打指示值用于指示对所述用户执行拨打的时间或不执行拨打任务；相应地，若所述拨打指示值属于第一类型指示值，则确定对所述用户进行拨打的时间为所述拨打指示值指示的时间点，若所述拨打指示值为第二类型指示值，则确定不对所述用户进行拨打；所述第一类型指示值包括按照预设时长划分预设拨打时间段得到的多个拨打指示值，所述第二类型指示值用于指示在预设拨打时间段内不执行拨打。
在上述设计中，通过预先对预设拨打时间段进行划分得到多个拨打指示值，可以使用拨打指示值指示拨打的时间范围，一方面，通过目标模型预测拨打指示值，相比于使用目标模型预测具体的拨打时刻来说，降低了模型预测的复杂度，从而使得模型预测的效率更高；另一方面，通过将大量的拨打时刻划分为拨打时间区间，可以降低需要模型预测的数据量，进一步地提高模型预测的效率。
第二方面，本发明实施例提供的一种信息处理装置，所述装置包括：
获取模块，用于获取用户对应的一个或多个目标特征的取值，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为与拨打失败强相关的特征；
处理模块，用于根据所述用户对应的一个或多个目标特征的取值，确定与所述用户进行联系的方式，所述与所述用户进行联系的方式包括对所述用户进行拨打的时间。
在一种可能的设计中，所述处理模块具体用于：将所述用户对应的一个或多个目标特征的取值输入目标模型，得到对所述用户进行拨打的时间。其中，所述目标模型是所述处理模块通过如下方式得到的：获取拨打成功的M个用户的数据和拨打失败的N个用户的数据，所述M个用户的数据和所述N个用户的数据中每个用户的数据包括至少一个特征的取值；从所述M个用户和所述N个用户中确定所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，并根据所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，从所述至少一个特征中筛选得到所述一个或多个目标特征；进一步地，使用所述M个用户和所述N个用户对应的所述一个或多个目标特征的取值以及所述M个用户和所述N个用户的拨打结果，训练得到所述目标模型；所述M个用户的拨打结果用于指示拨打成功的时间，所述N个用户的拨打结果用于指示拨打失败。
在一种可能的设计中，所述处理模块具体用于：针对于所述至少一个特征中的第一特征，基于所述第一特征对应的拨打成功的用户的数量和拨打失败的用户的数量，以及所述至少一个特征中除所述第一特征以外的特征对应的拨打成功的用户的数量和拨打失败的用户的数量，得到所述第一特征对应的拨打成功的概率和/或所述第一特征对应的拨打失败的概率；若所述第一特征对应的拨打成功的概率大于第一预设阈值，则确定所述第一特征为第一目标特征；和/或，若所述第一特征对应的拨打失败的概率大于第二预设阈值，则确定所述第一特征为第二目标特征。
在一种可能的设计中，所述至少一个特征包括基本属性特征、逾期还款特征、拨打行为特征、语音交互特征、风险表现特征中的任意一项或任意多项。
在一种可能的设计中，所述处理模块具体用于：根据所述用户对应的一个或多个目标特征的取值，得到拨打指示值，所述拨打指示值用于指示对所述用户执行拨打的时间或不执行拨打任务；进一步地，若所述拨打指示值属于第一类型指示值，则确定对所述用户进行拨打的时间为所述拨打指示值指示的时间点，若所述拨打指示值为第二类型指示值，则确定不对所述用户进行拨打；所述第一类型指示值包括按照预设时长划分预设拨打时间段 得到的多个拨打指示值，所述第二类型指示值用于指示在预设拨打时间段内不执行拨打。
第三方面，本发明实施例提供的一种计算机可读存储介质，包括指令，当其在计算机上运行时，使得计算机执行如上述第一方面任意所述的信息处理方法。
第四方面，本发明实施例提供的一种计算机程序产品，当其在计算机上运行时，使得计算机执行如上述第一方面任意所述的信息处理方法。
本发明的这些方面或其他方面在以下实施例的描述中会更加简明易懂。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简要介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域的普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例提供的一种信息处理方法对应的流程示意图；
图2为本发明实施例提供的一种信息处理方法的执行过程示意图；
图3为本发明实施例提供的一种信息处理装置的结构示意图；
图4为本发明实施例提供的一种终端设备的结构示意图；
图5为本发明实施例提供的一种后端设备的结构示意图。
具体实施方式
为了使本发明的目的、技术方案和优点更加清楚，下面将结合附图对本发明作进一步地详细描述，显然，所描述的实施例仅仅是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其它实施例，都属于本发明保护的范围。
金融科技(Fintech)是指将信息技术融入金融领域后，为金融领域带来的一种新的创新科技，通过使用先进的信息技术辅助实现金融作业、交易执行以及金融系统改进，可以提升金融系统的处理效率、业务规模，并可以降低成本和金融风险。一般来说，金融科技领域通常会涉及到大量的数据，比如信贷部门的贷款数据、销售部门的购卡数据、运维部门的维修数据等，如何采用科技的手段从大量的数据中挖掘出金融领域所需要的特征，一直是金融科技领域追求的目标。
以银行中的信贷部门为例，信贷部门的业务量通常是非常巨大的，比如，信贷部门的信贷数据库中可以存储着上万条甚至数十万条的历史贷款数据。在这种情况下，信贷部门的工作人员通常可以按照预设周期(比如一天、一个周、一个月)对这些历史贷款数据进行分析，若发现存在某一条贷款数据超过时限要求，则可以向该条信贷数据对应的用户拨打催收电话。然而，由于历史贷款数据的数据量非常大，导致待催款的用户较多，因此，采用何种方式对大量的待催款用户进行催收拨打并保证得到较好的催收效果，一直是金融科技领域需要解决的一个问题。
基于此，本发明实施例提供了一种信息处理方法，可以提高与用户进行联系的效率。
下面以催收拨打为例描述本发明实施例中的信息处理方法，可以理解地，该实施例仅是本发明中的信息处理方法的一种示例，并不对其构成限定；也就是说，本发明实施例中的信息处理方法也可以应用于其它领域，此处不作限定。
图1为本发明实施例提供的一种信息处理方法对应的流程示意图，该方法包括：
步骤101，获取待催收用户对应的一个或多个目标特征的取值。
此处，一个或多个目标特征可以包括至少一个第一目标特征和至少一个第二目标特征。其中，至少一个第一目标特征可以为与催收拨打成功强相关的特征，至少一个第二目标特征可以为与催收拨打失败强相关的特征。
具体实施中，若用户向信贷部门申请贷款，则信贷部门可以在向该用户批准贷款的同时将该用户的数据存储在第一信贷数据库中，第一信贷数据库可以用于存储向信贷部门进行贷款但暂未还款的用户的数据。其中，用户的数据可以包括用户的基本属性信息、贷款信息、信用信息等，基本属性信息可以包括用户的身高、血型、年龄、性别、家庭住址、联系方式、邮箱地址等，贷款信息可以包括用户A贷款的金额、贷款的时长、还款方式等，信用信息可以包括用户的历史贷款信息、其它部门对用户的评价信息等。表1为本发明实施例提供的一种可能的第一信贷数据库的示意表。
表1：一种可能的第一信贷数据库的示意
用户 年龄特征 性别特征 血型特征 信用特征 还款期限 贷款金额
用户A 1 35   O 优 2019.06.01 300万
用户A 2 25 男 B   2019.05.39 15万
用户A 3 40 男 AB 良 2019.07.01 1000万
如表作为一种示例，若用户的年龄为(20，35]之间的任一年龄，则年龄特征的取值可以为0，若用户的年龄(35，40]之间的任一年龄，则年龄特征的取值可以为1；若用户的性别为男，则性别特征的取值可以为1，若用户的性别为女，则性别特征的取值可以为0；若用户的信用为优，则信用特征的取值可以为1，若用户的信用为良，则信用特征的取值可以为0；若用户的血型为O，则血型特征的取值可以为1，若用户的血型为B，则血型特征的取值可以为2，若用户的血型为AB，则血型特征的取值可以为3。在该种示例下，第一信贷数据库中存储的用户的数据可以如表2所示。1所示，第一信贷数据库中可以存储有多个用户和多个用户的数据，其中，用户的基本属性信息和信用信息可以由多个特征的取值来表示。以用户A 3为例，用户A 3的年龄为40、性别为男、血型为AB型、信用为良，因此用户A 2的年龄特征的取值可以为40，性别特征的取值可以为男，血型特征的取值可以为AB，信用特征的取值可以为良；相应地，第一信贷数据库中还可以存储有多个用户贷款的金额和还款日期，由表1可知，用户A 3贷款的金额可以为1000万，还款日期可以为2019.07.01。
需要说明的是，表1仅是一种示例性的简单说明，其所列举的特征对应的值仅是为了便于说明方案，并不构成对方案的限定，可以理解地，特征的取值也可以以其它方式表示。
表2：一种可能的第一信贷数据库的示意
用户 年龄特征 性别特征 血型特征 信用特征 还款期限 贷款金额
用户A 1 0   1 1 2019.06.01 300万
用户A 2 0 1 2   2019.05.39 15万
用户A 3 1 1 3 0 2019.07.01 1000万
进一步地，若当前日期为2019.06.01，则信贷部门可以根据第一信贷数据库中用户A 1～用户A 3的还款期限，选择还款期限小于当前日期的用户A 1和还款期限等于当前日期的用户A 2作为待催收用户。针对于还款期限大于当前日期的用户A 3，信贷部门可以通过分析用户A 3的信用、还款期限、贷款金额评估用户A 3的还款风险，一般来说，用户的信用越低、还款期限越近、贷款金额越高，则用户的还款风险越高；根据表1或表2所示，由于用户A 3的贷款金额较大，还款期限与当前日期相差较近，且信用一般，因此，信贷部门也可以将用户A 3作为待催收用户。
在一个示例中，在确定待催收用户后，若发现第一信贷数据库中存储的待催收用户的数据不全面(比如缺少待催收用户的联系方式)，或者待催收用户的数据存在错误(比如待催收用户的手机号码已停机)，则可以从银行的数据库中获取待催收用户的数据。一般来说，银行的数据库中可以存储有银行中的各个部门与用户执行交易的过程中所统计的用户的数据，因此，贷款部门可以从银行的数据库中获取待催收用户A 1的性别特征和待催收用户A 2的信用特征，并可以对第一信贷数据库中待催收用户A 1和待催收用户A 2的信息进行补充或修正。如此，通过获取全面的待催收用户的数据，可以刻画真实的待催收用户的画像，使得基于全面的信息对待催收用户进行预测的预测结果更为准确。
下面以待催收用户A 3为例描述确定向待催收用户A 3进行催收拨打的时间的具体实现过程，确定向待催收用户A 1和/或待催收用户A 2进行催收拨打的时间的过程可以参照实现，具体不再赘述。
具体实施中，若与催收拨打成功强关联的特征包括年龄特征和信用特征，与催收拨打失败强关联的特征包括性别特征和信用特征，则可以获取待催收用户A 3对应的年龄特征、性别特征和信用特征的取值，进而根据待催收用户A 3对应的年龄特征、性别特征和信用特征的取值确定向待催收用户A 3进行催收拨打的时间，以使得在该催收拨打时间对应的时间点向用户A 3进行催收拨打电话时，能够得到较好的催收结果。
步骤102，根据待催收用户对应的一个或多个目标特征的取值，得到对待催收用户进行催收拨打的时间。
本发明实施例中，根据待催收用户A 3对应的一个或多个目标特征的取值得到对待催收用户A 3进行催收拨打的时间的过程可以为通过人工实现的，或者也可以为服务器自动实现的，具体不作限定。本发明实施例的下列实施例中具体描述采用服务器自动实现的方式得到对待催收用户A 3进行催收拨打的时间，该种方式可以无需人工操作，并可以提高催收拨打的效率。
在一种可能的实现方式中，可以将待催收用户A 3对应的一个或多个目标特征的取值输入目标模型，从而得到对待催收用户A 3进行催收拨打的时间。在一个示例中，目标模型可以根据待催收用户A 3对应的一个或多个目标特征的取值，预测得到拨打指示值，若拨打指示值属于第一类型指示值，则可以在拨打指示值所指示的时间点对待催收用户A 3进行催收拨打，若拨打指示值为第二类型指示值，则可以不对待催收用户A 3进行催收拨打；其中，第一类型指示值可以包括按照预设时长划分预设拨打时间段得到的多个拨打指示值，第二类型指示值可以用于指示在预设拨打时间段内不执行催收拨打。
举例来说，若预设拨打时间段为8:00～20:00，预设时长为15min，则按照预设时长划分预设拨打时间段可以得到第一～第四十八时间区间，第一时间区间可以为8:00～8:15，第二时间区间可以为8:15～8:30，……，第四十七时间区间可以为19:30～19:45，第四十八时间区 间可以为19：45～20:00。相应地，第一类型指示值可以包括与第一～第四十八时间区间分别对应的第一～第四十八拨打指示值，比如，第一拨打指示值为1，第二拨打指示值为2，……，第四十七拨打指示值为47，第四十八拨打指示值为48；第二类型指示值可以为不同于第一类型指示值包括的48个指示值中的任一指示值，比如0。如此，若目标模型预测得到的拨打指示值为3，则说明对待催收用户A 3进行催收拨打的最佳时间为8:30～8:45中的任一时刻；若目标模型预测得到的拨打指示值为40，则说明对待催收用户A 3进行催收拨打的最佳时间为18:00～18:15中的任一时刻；若目标模型预测得到的拨打指示值为0，则说明无需对待催收用户A 3进行催收拨打。
本发明实施例中，通过预先对预设拨打时间段进行划分得到多个拨打指示值，可以使用拨打指示值指示拨打的时间范围，一方面，通过目标模型预测拨打指示值，相比于使用目标模型预测具体的拨打时刻来说，可以降低了模型预测的复杂度，从而使得模型预测的效率更高；另一方面，通过将大量的拨打时刻划分为拨打时间区间，可以降低需要模型预测的数据量，进一步地提高模型预测的效率。
需要说明的是，本发明实施例提供的信息处理方法可以包括第一阶段(即目标模型的训练过程)和第二阶段(即目标模型的预测过程)，上述实施例中具体描述了使用目标模型进行预测的过程，下面重点描述训练得到目标模型的过程。
本发明实施例中，信贷部门可以将申请贷款的用户中已还款的用户的数据和/或进行过催收拨打但暂未还款的用户的数据存储在第二信贷数据库中，如此，第二信贷数据库中可以存储有主动还款的用户的数据、已执行过催收拨打并还款的用户的数据、已执行过催收拨打但未还款的用户的数据。具体实施中，可以从第二信贷数据库中获取已进行过催收拨打的多个用户的数据，并可以从1000个用户中确定出催收拨打成功的用户和催收拨打失败的用户。此处，催收拨打成功可以是指在向用户进行催收拨打时用户接通了电话并按时还款或者按照电话中承诺的时间还款，或者用户接通过了电话并在拨打催收电话的当天完成了还款等；相应地，催收拨打失败可以是指在向用户进行催收拨打时用户未接通电话，或者用户接通了电话但中途挂断，或者用户接通了电话但并在按时还款或未在拨打催收电话的当天完成了还款或未按照电话中承诺的时间还款等。需要说明的是，催收拨打成功和催收拨打失败还可以包括其它示例，具体不作限定。
在一个示例中，第二信贷数据库中每个用户的数据可以包括该用户对应的至少一个特征的取值，至少一个特征可以包括基本属性特征、逾期还款特征、拨打行为特征、语音交互特征、风险表现特征中的任意一项或任意多项。其中，基本属性特征可以如表1或表2所示；逾期还款特征可以用于表征用户的逾期还款行为，比如，用户逾期还款的时间、逾期还款的金额、按时还款的配合程度等；拨打行为特征可以用于表征用户接收催收拨打电话的情况，比如向用户进行催收拨打的时间、在该时间进行催收拨打时用户是否接听电话等；语音交互特征可以用于表征用户接听催收拨打电话的表现，比如用户在催收拨打电话中表现的还款意愿、还款情绪等；风险表现特征可以用于表征用户在接听催收拨打电话后的行为，比如可以根据用户在催收拨打电话后是否进行当天还款以及用户在接收催收拨打电话时的态度等评估用户的风险值，该风险值可以用于表征催收拨打电话的有效程度。
本发明实施例中，通过综合分析用户的各种特征，可以使得训练数据更加全面，从而提高训练得到的目标模型的准确性，使得目标模型的预测效果更好。
举例来说，若从第二信贷数据库中获取了1000个用户的数据，1000个用户中存在600 个拨打成功的用户和400个拨打失败的用户，则可以从催收拨打成功的600个用户中确定出至少一个特征分别对应的催收拨打成功的用户的数量，并可以从催收拨打失败的400个用户中确定出至少一个特征分别对应的催收拨打失败的用户的数量；进而可以根据至少一个特征分别对应的催收拨打成功的用户的数量与600个用户确定至少一个特征分别对应的催收拨打成功的频率，并可以根据至少一个特征分别对应的催收拨打失败的用户的数量与600个用户确至少一个特征分别对应的催收拨打失败的频率，从而可以得到至少一个特征与催收拨打结果对应关系表。表3为本发明实施例提供的一种至少一个特征与催收拨打结果的对应关系示意表。
表3:一种至少一个特征与催收拨打结果的对应关系示意
  特征a 1 特征a 2 特征a 3 特征a 4 特征a 5
拨打成功 0.8 0.3 0.7 0.2 0.4
拨打失败   0.1 0.3 0.8 0.4
以特征a 1为例，如表3所示，特征a 1对应的催收拨打成功的用户的频率为0.8，则说明催收拨打成功的600个用户中存在480个用户的数据具有特征a 1；相应地，特征a 1对应的催收拨打失败的用户的频率为0，则说明催收拨打失败的400个用户中每个用户的数据均不具有特征a 1。
一般来说，若某一特征对应的催收拨打成功的用户的频率越大，说明该特征与催收拨打成功的关联性越强，催收拨打成功的多个用户存在该特征的概率也较大；如此，可以将该特征作为与催收拨打成功强关联的第一目标特征；相应地，若某一特征对应的催收拨打失败的用户的频率越大，说明该特征与催收拨打失败的关联性越强，催收拨打失败的多个用户存在该特征的概率也较大；如此，可以将该特征作为与催收拨打失败强关联的第二目标特征。
综上，在一个示例中，本发明实施例可以采用如下方法从至少一个特征中筛选得到与催收拨打成功强关联的第一目标特征和与催收拨打失败强关联的第二目标特征：
以特征a 2为例，具体实施中，可以根据表3确定出催收拨打成功的600个用户中存在特征a 2的用户的数量N 11(即180)、催收拨打成功的600个用户中不存在特征a 2的用户的数量N 10(即420)、催收拨打失败的400个用户中存在特征a 2的用户的数量N 01(即60)、催收拨打失败的400个用户中不存在特征a 2的用户的数量N 00(即340)。进一步地，可以分别计算得到N 11对应的期望值E 11、N 10对应的期望值E 10、N 01对应的期望值E 01、N 00对应的期望值E 00；以N 11对应的期望值为例，N 11对应的期望值可以为N 11、催收拨打成功的用户的概率(即催收拨打成功的600个用户与1000个用户的比例)、特征a 2的概率(即1000个用户中存在特征a 2的用户的数量与不存在特征a 2的用户的数量的比例)的乘积。
在确定N 11、E 11、N 10、E 10、N 01、E 01、N 00和E 00之后，可以使用N 11和N 11对应的期望值E 11、N 10对应的期望值E 10、N 01对应的期望值E 01、N 00对应的期望值E 00分别计算得到特征a 2对应的第一卡方值和/或特征a 2对应的第二卡方值；其中，特征a 2对应的第一卡方值可以用于预测“催收拨打成功与特征a 2相关”(即特征a 2对应的催收拨打成功的概率)这一假设为真的概率，特征a 2对应的第二卡方值可以用于预测“催收拨打失败与特征a 2相关”(即特征a 2对应的催收拨打失败的概率)这一假设为真的概率。
以特征a 2对应的第一卡方值为例，具体地说，特征a 2对应的第一卡方值可以满足如下 条件：
Figure PCTCN2020094746-appb-000001
其中，t可以用于表征为催收拨打成功或失败，若催收拨打成功，则t可以为1，若催收拨打失败，则t可以为0；c可以用于表征特征a 2是否存在，若特征a 2存在，则c可以为1，若特征a 2不存在，则c可以为0。
进一步地，在根据特征a 2对应的第一卡方值预测得到特征a 2对应的催收拨打成功的概率后，若确定特征a 2对应的催收拨打成功的概率大于第一预设阈值，说明特征a 2与催收拨打成功强相关，从而可以将特征a 2作为一个第一目标特征；和/或，在根据特征a 2对应的第二卡方值预测得到特征a 2对应的催收拨打失败的概率后，若确定特征a 2对应的催收拨打失败的概率大于第二预设阈值，说明特征a 2与催收拨打失败强相关，从而可以将特征a 2作为一个第二目标特征。其中，第一预设阈值与第二预设阈值可以由本领域技术人员根据经验进行设置，第一预设阈值与第二预设阈值可以相同，也可以不同，具体不作限定。
本发明实施例中，通过确定至少一个特征分别对应的催收拨打成功的概率，可以从至少一个特征中选取概率大于第一预设阈值的多个特征作为与催收拨打成功强相关的第一目标特征，相应地，通过确定至少一个特征分别对应的催收拨打失败的概率，可以从至少一个特征中选取概率大于第二预设阈值的多个特征作为与催收拨打失败强相关的第二目标特征，如此，第一目标特征和第二目标特征可以用于衡量催收拨打效果的好坏；也就是说，通过如上方式确定第一目标特征与第二目标特征，可以从催收拨打效果入手对催收拨打时间进行分析，从而可以提高催熟拨打的准确率。
本发明实施例中，若根据上述过程得到10个第一目标特征和6个第二目标特征，则可以使用催收拨打成功的600个用户分别对应的10个第一目标特征的取值和催收拨打结果以及催收拨打失败的400个用户分别对应的6个第二目标特征的取值和催收拨打结果，训练得到目标模型。以用户R为例，若用户R催收拨打成功，则用户R的催收拨打结果可以为向用户R进行催收拨打的时间对应的时间区间的拨打指示值，比如，若10:03时向用户R进行催收拨打且催收拨打成功，则用户R的催收拨打结果可以为8，若20:00时向用户R进行催收拨打且催收拨打成功，则用户R的催收拨打结果可以为48；相应地，若用户R催收拨打成功，则用户R的催收拨打结果可以指示催收拨打失败的拨打指示值，比如0。
具体实施中，可以根据催收拨打成功的600个用户分别对应的10个第一目标特征的取值和催收拨打结果以及催收拨打失败的400个用户分别对应的6个第二目标特征的取值和催收拨打结果构建特征矩阵。下面分别从几种可能的情形描述特征矩阵的结构。
情形一
在情形一中，10个第一目标特征与6个第二目标特征可以均不相同；如此，特征矩阵可以为一个1000行17列的矩阵。
其中，特征矩阵的前600行可以包括催收拨打成功的600个用户的信息(包括10个第一目标特征的取值和催收拨打结果)，特征矩阵的后400行可以包括催收拨打失败的400个用户的信息(包括6个第二目标特征的取值和催收拨打结果)；相应地，特征矩阵的前10列可以包括1000个用户对应的10个第一目标特征的取值，特征矩阵的第10～第16列可以包括1000个用户对应的6个第二目标特征的取值，特征矩阵的第17列可以包括1000个用户对应的催收拨打结果，每个用户的催收拨打结果可以为0～48之间的任一数值。
情形二
在情形一中，10个第一目标特征与6个第二目标特征可以存在i个特征相同，i>0；如此，特征矩阵可以为一个1000行17-i列的矩阵。
其中，特征矩阵的前600行可以包括催收拨打成功的600个用户的信息(包括10个第一目标特征的取值和催收拨打结果)，特征矩阵的后400行可以包括催收拨打失败的400个用户的信息(包括6个第二目标特征的取值和催收拨打结果)；相应地，特征矩阵的前10-i列可以包括1000个用户对应的10个第一目标特征中与6个第二目标特征不同的10-i个第一目标特征的取值，特征矩阵的第11-i～第10列可以包括1000个用户对应的10个第一目标特征中与6个第二目标特征相同的i个第一目标特征的取值，特征矩阵的第i～第16-i列可以包括1000个用户对应的6个第二目标特征的取值，特征矩阵的第17-i列可以包括1000个用户对应的催收拨打结果，每个用户的催收拨打结果可以为0～48之间的任一数值。
情形三
在情形三中，10个第一目标特征中的6个第一目标特征与6个第二目标特征相同；如此，特征矩阵可以为一个1000行11列的矩阵。
其中，特征矩阵的前600行可以包括催收拨打成功的600个用户的信息(包括10个第一目标特征的取值和催收拨打结果)，特征矩阵的后400行可以包括催收拨打失败的400个用户的信息(包括6个第二目标特征的取值和催收拨打结果)；相应地，特征矩阵的前4列可以包括1000个用户对应的10个第一目标特征中与6个第二目标特征不同的4个第一目标特征的取值，特征矩阵的第5～第10列可以包括1000个用户对应的10个第一目标特征中与6个第二目标特征相同的6个第一目标特征的取值，特征矩阵的第11列可以包括1000个用户对应的催收拨打结果，每个用户的催收拨打结果可以为0～48之间的任一数值。
进一步地，可以将特征矩阵输入深度学习平台，进而训练得到目标模型；其中，深度学习平台可以是指神经网络深度学习平台，比如长短期记忆网络(Long Short-Term Memory，LSTM)平台、如此，训练得到的目标模型可以为LSTM神经网络模型；或者，深度学习平台可以是指机器学习平台，比如差分整合移动平均自回归模型(Autoregressive Integrated Moving Average model，ARIMA)如此，训练得到的目标模型可以为线性模型、回归模型、概率图模型中的任意一个，具体不作限定。
以训练得到LSTM神经网络模型的过程为例，深度学习平台可以预先构建一个初始模型，该初始模型中可以包括输入门节点、输出门节点、遗忘门节点、处理节点等；具体实施中，在将特征矩阵输入深度学习平台后，初始模型中的输入门节点可以对特征矩阵中的特征进行分析，若确定某一特征为无用特征，则可以将该特征传送给遗忘门节点，若确定某一特征为有用特征，则可以将该特征传送给输出门节点；相应地，输出门节点可以对有用特征进行权重学习，比如可以使用预设激活函数对该权重的阈值进行调整，并可以根据调整的结果更新预设激活函数，遗忘门节点可以根据输出门节点的处理结果确定无用信息是否对出门节点的处理过程存在一定影响，若确定无用信息对出门节点的处理过程存在一定影响，则可以将无用信息传送给输出门节点，以使输出门节点更新模型，若确定无用信息对出门节点的处理过程不存在影响，则可以将无用信息丢弃。
本发明实施例中，输出门节点可以训练得到49个神经元节点，这49个神经元节点中存在48个神经元节点用于表征催收拨打成功对应的时间，剩余的1个神经元节点用于表征催收拨打失败。具体实施中，49个神经元节点可以分别与多分类器连接，多分类器可以用于 预测分别经过49个神经元节点的概率，进而可以从49个神经元节点中选择概率最大的神经元节点作为目标神经元节点，从而可以将目标神经元节点对应的拨打标识输出给用户。
举例来说，将用户A 1的数据输入目标模型后，多分类器可以预测得到48个拨打指示值分别对应的概率，若拨打指示值0对应的概率最大，则目标模型的输出可以为0，即目标模型预测得到“向用户A 1拨打催收电话时用户A 1不会接听催收电话”的信息，如此，可以无需向用户A 1拨打催收电话，从而提高催收拨打的效率，避免执行无用的催收拨打工作；相应地，若拨打指示值2对应的概率最大，则目标模型的输出可以为2，即目标模型预测得到“8:15～8:30时向用户A 1拨打催收电话的催收效果较好”，如此，可以在8:15～8:30中选择任意时刻向用户A 1拨打催收电话，从而可以无需人工确定拨打时间，提高催收拨打的准确度和效率。
本发明的上述实施例中，获取用户对应的一个或多个目标特征的取值，并根据所述用户对应的一个或多个目标特征的取值，得到与所述用户进行联系的方式，所述与所述用户进行联系的方式包括对所述用户进行拨打的时间；其中，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为拨打失败强相关的特征。本发明实施例中，通过确定与拨打成功强相关的第一目标特征和与拨打失败强相关的第二目标特征，可以使用第一目标特征的取值和第二目标特征的取值衡量联系效果，如此，通过使用用户的数据中第一目标特征的取值和第二目标特征的取值即可确定与用户的联系效果较好的拨打时间；也就是说，本发明实施例中通过预先分析得到对用户进行拨打的时间，可以得到较好的联系效果，从而可以提高与用户进行联系的效率。
下面以另一个角度描述本发明实施例提供的信息处理方法，图2为本发明实施例提供的一种信息处理方法的执行过程示意图，该过程可以参照图1所示意的信息处理方法进行实施。
具体实施中，如图2所示，本发明实施例中的信息处理方法可以包括模型训练阶段和模型预测阶段；在模型训练阶段中，通过依次连接的特征生成装置、目标特征筛选装置和模型训练装置可以训练得到目标模型；相应地，在模型预测阶段中，通过将待催收用户的数据输入目标模型，可以预测得到向待催收用户进行催收拨打的时间。下面分别介绍模型训练阶段和模型预测阶段中各个装置的功能。
模型训练阶段
若将已进行催收拨打的M+N个用户的数据输入特征生成装置，则特征生成装置可以根据M+N个用户的数据获取至少一个特征以及M+N个用户对应的至少一个特征的取值；且，特征生成装置还可以获取M+N个用户的催收拨打结果，比如M+N个用户催收拨打是成功还是失败，若某一用户催收拨打成功，则还可以获取向该用户催收拨打的时间。
进一步地，特征生成装置可以将M+N个用户对应的至少一个特征的取值和M+N个用户的催收拨打结果发送给目标特征筛选装置，如此，目标特征筛选装置可以从至少一个特征中筛选得到与催收拨打成功强相关的特征和与催收拨打失败强相关的特征，进而可以将M+N个用户对应的与催收拨打成功强相关的特征的取值和与催收拨打失败强相关的特征的取值发送给模型训练装置。
相应地，模型训练装置可以根据M+N个用户对应的与催收拨打成功强相关的特征的取值、与催收拨打失败强相关的特征的取值以及M+N个用户的催收拨打结果对模型进行训练， 得到目标模型。
模型预测阶段
若需要确定是否可以对待催收用户进行催收拨打，或者确定何时对待催收用户进行催收拨打的效果较好，则可以将待催收用户的数据输入目标模型；如此，目标模型可以根据待催收用户的数据对待催收用户进行预测，得到向待催收用户进行催收拨打的时间。其中，目标模型可以通过输出拨打时间区间对应的拨打指示值或不拨打的拨打指示值确定向待催收用户进行催收拨打的时间，向待催收用户进行催收拨打的时间可以包括不向待催收用户拨打或何时向待催收用户拨打。
需要说明的是，图2中各个装置执行的过程可以参照步骤101和步骤102的过程进行实现，此处不再赘述。
本发明的上述实施例中，获取用户对应的一个或多个目标特征的取值，并根据所述用户对应的一个或多个目标特征的取值，得到与所述用户进行联系的方式，所述与所述用户进行联系的方式包括对所述用户进行拨打的时间；其中，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为拨打失败强相关的特征。本发明实施例中，通过确定与拨打成功强相关的第一目标特征和与拨打失败强相关的第二目标特征，可以使用第一目标特征的取值和第二目标特征的取值衡量联系效果，如此，通过使用用户的数据中第一目标特征的取值和第二目标特征的取值即可确定与用户的联系效果较好的拨打时间；也就是说，本发明实施例中通过预先分析得到对用户进行拨打的时间，可以得到较好的联系效果，从而可以提高与用户进行联系的效率。
针对图1和图2所示意的方法流程，本发明实施例还提供一种信息处理装置，该装置的具体内容可以参照上述方法实施。
图3为本发明实施例提供的一种信息处理装置，所述装置包括：
获取模块301，用于获取用户对应的一个或多个目标特征的取值，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为与拨打失败强相关的特征；
处理模块302，用于根据所述用户对应的一个或多个目标特征的取值，确定向所述用户进行联系的方式；所述向所述用户进行联系的方式包括对所述用户进行拨打的时间。
可选地，所述处理模块302具体用于：
将所述用户对应的一个或多个目标特征的取值输入目标模型，得到向所述用户进行联系的方式；
其中，所述目标模型是所述处理模块通过如下方式得到的：
获取拨打成功的M个用户的数据和拨打失败的N个用户的数据，所述M个用户的数据和所述N个用户的数据中每个用户的数据包括至少一个特征的取值；
从所述M个用户和所述N个用户中确定所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，并根据所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，从所述至少一个特征中筛选得到所述一个或多个目标特征；
使用所述M个用户和所述N个用户对应的所述一个或多个目标特征的取值以及所述M个用户和所述N个用户的拨打结果，训练得到所述目标模型；所述M个用户的拨打结果用于指示拨打成功的时间，所述N个用户的拨打结果用于指示拨打失败。
可选地，所述处理模块302具体用于：
针对于所述至少一个特征中的第一特征，基于所述第一特征对应的拨打成功的用户的数量和拨打失败的用户的数量，以及所述至少一个特征中除所述第一特征以外的特征对应的拨打成功的用户的数量和拨打失败的用户的数量，得到所述第一特征对应的拨打成功的概率和/或所述第一特征对应的拨打失败的概率；若所述第一特征对应的拨打成功的概率大于第一预设阈值，则确定所述第一特征为第一目标特征；和/或，若所述第一特征对应的拨打失败的概率大于第二预设阈值，则确定所述第一特征为第二目标特征。
可选地，所述至少一个特征包括基本属性特征、逾期还款特征、拨打行为特征、语音交互特征、风险表现特征中的任意一项或任意多项。
可选地，所述处理模块302具体用于：
根据所述用户对应的一个或多个目标特征的取值，得到拨打指示值，所述拨打指示值用于指示对所述用户进行拨打的时间或不进行拨打；
若所述拨打指示值属于第一类型指示值，则确定对所述用户进行拨打的时间为所述拨打指示值指示的时间点，若所述拨打指示值为第二类型指示值，则确定不对所述用户进行拨打；所述第一类型指示值包括按照预设时长划分预设拨打时间段得到的多个拨打指示值，所述第二类型指示值用于指示在预设拨打时间段内不执行拨打。
从上述内容可以看出：本发明的上述实施例中，获取用户对应的一个或多个目标特征的取值，并根据所述用户对应的一个或多个目标特征的取值，得到与所述用户进行联系的方式，所述与所述用户进行联系的方式包括对所述用户进行拨打的时间；其中，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为拨打失败强相关的特征。本发明实施例中，通过确定与拨打成功强相关的第一目标特征和与拨打失败强相关的第二目标特征，可以使用第一目标特征的取值和第二目标特征的取值衡量联系效果，如此，通过使用用户的数据中第一目标特征的取值和第二目标特征的取值即可确定与用户的联系效果较好的拨打时间；也就是说，本发明实施例中通过预先分析得到对用户进行拨打的时间，可以得到较好的联系效果，从而可以提高与用户进行联系的效率。
基于同一发明构思，本发明实施例还提供了一种计算机可读存储介质，包括指令，当其在计算机上运行时，使得计算机执行如图1或图2任一项所述的信息处理方法。
基于同一发明构思，本发明实施例还提供了一种计算机程序产品，当其在计算机上运行时，使得计算机执行如图1或图2任一项所述的信息处理方法。
基于相同的技术构思，本发明实施例提供了一种终端设备，如图4所示，包括至少一个处理器1101，以及与至少一个处理器连接的存储器1102，本发明实施例中不限定处理器1101与存储器1102之间的具体连接介质，图4中处理器1101和存储器1102之间通过总线连接为例。总线可以分为地址总线、数据总线、控制总线等。
在本发明实施例中，存储器1102存储有可被至少一个处理器1101执行的指令，至少一个处理器1101通过执行存储器1102存储的指令，可以执行前述的信息处理方法中所包括的步骤。
其中，处理器1101是终端设备的控制中心，可以利用各种接口和线路连接终端设备的各个部分，通过运行或执行存储在存储器1102内的指令以及调用存储在存储器1102内的数据，从而实现数据处理。可选的，处理器1101可包括一个或多个处理单元，处理器1101 可集成应用处理器和调制解调处理器，其中，应用处理器主要处理操作系统、用户界面和应用程序等，调制解调处理器主要处理运维人员下发的指令。可以理解的是，上述调制解调处理器也可以不集成到处理器1101中。在一些实施例中，处理器1101和存储器1102可以在同一芯片上实现，在一些实施例中，它们也可以在独立的芯片上分别实现。
处理器1101可以是通用处理器，例如中央处理器(CPU)、数字信号处理器、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件，可以实现或者执行本发明实施例中公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者任何常规的处理器等。结合信息处理方法的实施例所公开的方法的步骤可以直接体现为硬件处理器执行完成，或者用处理器中的硬件及软件模块组合执行完成。
存储器1102作为一种非易失性计算机可读存储介质，可用于存储非易失性软件程序、非易失性计算机可执行程序以及模块。存储器1102可以包括至少一种类型的存储介质，例如可以包括闪存、硬盘、多媒体卡、卡型存储器、随机访问存储器(Random Access Memory，RAM)、静态随机访问存储器(Static Random Access Memory，SRAM)、可编程只读存储器(Programmable Read Only Memory，PROM)、只读存储器(Read Only Memory，ROM)、带电可擦除可编程只读存储器(Electrically Erasable Programmable Read-Only Memory，EEPROM)、磁性存储器、磁盘、光盘等等。存储器1102是能够用于携带或存储具有指令或数据结构形式的期望的程序代码并能够由计算机存取的任何其他介质，但不限于此。本发明实施例中的存储器1102还可以是电路或者其它任意能够实现存储功能的装置，用于存储程序指令和/或数据。
基于相同的技术构思，本发明实施例提供了一种后端设备，如图5所示，包括至少一个处理器1201，以及与至少一个处理器连接的存储器1202，本发明实施例中不限定处理器1201与存储器1202之间的具体连接介质，图5中处理器1201和存储器1202之间通过总线连接为例。总线可以分为地址总线、数据总线、控制总线等。
在本发明实施例中，存储器1202存储有可被至少一个处理器1201执行的指令，至少一个处理器1201通过执行存储器1202存储的指令，可以执行前述的信息处理方法中所包括的步骤。
其中，处理器1201是后端设备的控制中心，可以利用各种接口和线路连接后端设备的各个部分，通过运行或执行存储在存储器1202内的指令以及调用存储在存储器1202内的数据，从而实现数据处理。可选的，处理器1201可包括一个或多个处理单元，处理器1201可集成应用处理器和调制解调处理器，其中，应用处理器主要处理操作系统、应用程序等，调制解调处理器主要对接收到的指令进行解析以及对接收到的结果进行解析。可以理解的是，上述调制解调处理器也可以不集成到处理器1201中。在一些实施例中，处理器1201和存储器1202可以在同一芯片上实现，在一些实施例中，它们也可以在独立的芯片上分别实现。
处理器1201可以是通用处理器，例如中央处理器(CPU)、数字信号处理器、专用集成电路(Application Specific Integrated Circuit，ASIC)、现场可编程门阵列或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件，可以实现或者执行本发明实施例中公开的各方法、步骤及逻辑框图。通用处理器可以是微处理器或者任何常规的处理器等。结合信息处理方法的实施例所公开的方法的步骤可以直接体现为硬件处理器执行完成，或 者用处理器中的硬件及软件模块组合执行完成。
存储器1202作为一种非易失性计算机可读存储介质，可用于存储非易失性软件程序、非易失性计算机可执行程序以及模块。存储器1202可以包括至少一种类型的存储介质，例如可以包括闪存、硬盘、多媒体卡、卡型存储器、随机访问存储器(Random Access Memory，RAM)、静态随机访问存储器(Static Random Access Memory，SRAM)、可编程只读存储器(Programmable Read Only Memory，PROM)、只读存储器(Read Only Memory，ROM)、带电可擦除可编程只读存储器(Electrically Erasable Programmable Read-Only Memory，EEPROM)、磁性存储器、磁盘、光盘等等。存储器1202是能够用于携带或存储具有指令或数据结构形式的期望的程序代码并能够由计算机存取的任何其他介质，但不限于此。本发明实施例中的存储器1202还可以是电路或者其它任意能够实现存储功能的装置，用于存储程序指令和/或数据。
本领域内的技术人员应明白，本发明的实施例可提供为方法、或计算机程序产品。因此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
尽管已描述了本发明的优选实施例，但本领域内的技术人员一旦得知了基本创造性概念，则可对这些实施例作出另外的变更和修改。所以，所附权利要求意欲解释为包括优选实施例以及落入本发明范围的所有变更和修改。
显然，本领域的技术人员可以对本发明进行各种改动和变型而不脱离本发明的精神和范围。这样，倘若本发明的这些修改和变型属于本发明权利要求及其等同技术的范围之内，则本发明也意图包含这些改动和变型在内。

Claims (12)

  1. 一种信息处理方法，其特征在于，所述方法包括：
    获取用户对应的一个或多个目标特征的取值，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为与拨打失败强相关的特征；
    根据所述用户对应的一个或多个目标特征的取值，确定向所述用户进行联系的方式；所述向所述用户进行联系的方式包括对所述用户进行拨打的时间。
  2. 根据权利要求1所述的方法，其特征在于，所述根据所述用户对应的一个或多个目标特征的取值，确定与所述用户进行联系的方式，包括：
    将所述用户对应的一个或多个目标特征的取值输入目标模型，得到向所述用户进行联系的方式；
    其中，所述目标模型是通过如下方式得到的：
    获取拨打成功的M个用户的数据和拨打失败的N个用户的数据，所述M个用户的数据和所述N个用户的数据中每个用户的数据包括至少一个特征的取值；
    从所述M个用户和所述N个用户中确定所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，并根据所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，从所述至少一个特征中筛选得到所述一个或多个目标特征；
    使用所述M个用户和所述N个用户对应的所述一个或多个目标特征的取值以及所述M个用户和所述N个用户的拨打结果，训练得到所述目标模型；所述M个用户的拨打结果用于指示拨打成功的时间，所述N个用户的拨打结果用于指示拨打失败。
  3. 根据权利要求2所述的方法，其特征在于，从所述至少一个特征中筛选得到所述一个或多个目标特征，包括：
    针对于所述至少一个特征中的第一特征，基于所述第一特征对应的拨打成功的用户的数量和拨打失败的用户的数量，以及所述至少一个特征中除所述第一特征以外的特征对应的拨打成功的用户的数量和拨打失败的用户的数量，得到所述第一特征对应的拨打成功的概率和/或所述第一特征对应的拨打失败的概率；若所述第一特征对应的拨打成功的概率大于第一预设阈值，则确定所述第一特征为第一目标特征；和/或，若所述第一特征对应的拨打失败的概率大于第二预设阈值，则确定所述第一特征为第二目标特征。
  4. 根据权利要求2所述的方法，其特征在于，所述至少一个特征包括基本属性特征、逾期还款特征、拨打行为特征、语音交互特征、风险表现特征中的任意一项或任意多项。
  5. 根据权利要求1至4中任一项所述的方法，其特征在于，所述根据所述用户对应的一个或多个目标特征的取值，确定向所述用户进行联系的方式，包括：
    根据所述用户对应的一个或多个目标特征的取值，得到拨打指示值，所述拨打指示值用于指示对所述用户进行拨打的时间或不进行拨打；
    若所述拨打指示值属于第一类型指示值，则确定对所述用户进行拨打的时间为所述拨打指示值指示的时间点，若所述拨打指示值为第二类型指示值，则确定不对所述用户进行拨打；所述第一类型指示值包括按照预设时长划分预设拨打时间段得到的多个拨打指示值，所述第二类型指示值用于指示在预设拨打时间段内不执行拨打。
  6. 一种信息处理装置，其特征在于，所述装置包括：
    获取模块，用于获取用户对应的一个或多个目标特征的取值，所述一个或多个目标特征包括至少一个第一目标特征和至少一个第二目标特征，所述第一目标特征为与拨打成功强相关的特征，所述第二目标特征为与拨打失败强相关的特征；
    处理模块，用于根据所述用户对应的一个或多个目标特征的取值，确定向所述用户进行联系的方式；所述向所述用户进行联系的方式包括对所述用户进行拨打的时间。
  7. 根据权利要求6所述的装置，其特征在于，所述处理模块具体用于：
    将所述用户对应的一个或多个目标特征的取值输入目标模型，得到向所述用户进行联系的方式；
    其中，所述目标模型是所述处理模块通过如下方式得到的：
    获取拨打成功的M个用户的数据和拨打失败的N个用户的数据，所述M个用户的数据和所述N个用户的数据中每个用户的数据包括至少一个特征的取值；
    从所述M个用户和所述N个用户中确定所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，并根据所述至少一个特征分别对应的拨打成功的用户的数量和拨打失败的用户的数量，从所述至少一个特征中筛选得到所述一个或多个目标特征；
    使用所述M个用户和所述N个用户对应的所述一个或多个目标特征的取值以及所述M个用户和所述N个用户的拨打结果，训练得到所述目标模型；所述M个用户的拨打结果用于指示拨打成功的时间，所述N个用户的拨打结果用于指示拨打失败。
  8. 根据权利要求7所述的装置，其特征在于，所述处理模块具体用于：
    针对于所述至少一个特征中的第一特征，基于所述第一特征对应的拨打成功的用户的数量和拨打失败的用户的数量，以及所述至少一个特征中除所述第一特征以外的特征对应的拨打成功的用户的数量和拨打失败的用户的数量，得到所述第一特征对应的拨打成功的概率和/或所述第一特征对应的拨打失败的概率；若所述第一特征对应的拨打成功的概率大于第一预设阈值，则确定所述第一特征为第一目标特征；和/或，若所述第一特征对应的拨打失败的概率大于第二预设阈值，则确定所述第一特征为第二目标特征。
  9. 根据权利要求7所述的装置，其特征在于，所述至少一个特征包括基本属性特征、逾期还款特征、拨打行为特征、语音交互特征、风险表现特征中的任意一项或任意多项。
  10. 根据权利要求6至9中任一项所述的装置，其特征在于，所述处理模块具体用于：
    根据所述用户对应的一个或多个目标特征的取值，得到拨打指示值，所述拨打指示值用于指示对所述用户进行拨打的时间或不进行拨打；
    若所述拨打指示值属于第一类型指示值，则确定对所述用户进行拨打的时间为所述拨打指示值指示的时间点，若所述拨打指示值为第二类型指示值，则确定不对所述用户进行拨打；所述第一类型指示值包括按照预设时长划分预设拨打时间段得到的多个拨打指示值，所述第二类型指示值用于指示在预设拨打时间段内不执行拨打。
  11. 一种计算机可读存储介质，其特征在于，包括指令，当其在计算机上运行时，使得计算机执行如权利要求1至5任一项所述的方法。
  12. 一种计算机程序产品，其特征在于，当其在计算机上运行时，使得计算机执行如权利要求1至5任一项所述的方法。
PCT/CN2020/094746 2019-06-11 2020-06-05 一种信息处理方法及装置 Ceased WO2020248916A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910501871.4 2019-06-11
CN201910501871.4A CN110245985B (zh) 2019-06-11 2019-06-11 一种信息处理方法及装置

Publications (1)

Publication Number Publication Date
WO2020248916A1 true WO2020248916A1 (zh) 2020-12-17

Family

ID=67886620

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/094746 Ceased WO2020248916A1 (zh) 2019-06-11 2020-06-05 一种信息处理方法及装置

Country Status (2)

Country Link
CN (1) CN110245985B (zh)
WO (1) WO2020248916A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113989012A (zh) * 2021-09-17 2022-01-28 大箴(杭州)科技有限公司 不良资产的借款对象人群分类方法及装置、介质、设备

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110245985B (zh) * 2019-06-11 2023-09-08 深圳前海微众银行股份有限公司 一种信息处理方法及装置
CN113554227A (zh) * 2021-07-23 2021-10-26 中信银行股份有限公司 用户失联率预测模型的训练方法及用户失联率的预测方法
JP7421527B2 (ja) * 2021-09-27 2024-01-24 楽天グループ株式会社 情報処理装置、方法及びプログラム

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106339482A (zh) * 2016-08-31 2017-01-18 五八同城信息技术有限公司 确定商家最佳沟通时间的方法及装置
CN106952155A (zh) * 2017-03-08 2017-07-14 深圳前海纵腾金融科技服务有限公司 一种基于信用评分的催收方法及装置
US20170270569A1 (en) * 2007-04-09 2017-09-21 Yellowpages.Com Llc Systems and methods to provide connections via callback acceptance cross-reference to related applications
CN107528969A (zh) * 2017-07-27 2017-12-29 深圳市佰仟金融服务有限公司 电话拨打时间的管理方法、管理装置及终端设备
CN109559221A (zh) * 2018-11-20 2019-04-02 中国银行股份有限公司 基于用户数据的催收方法、装置和存储介质
CN110245985A (zh) * 2019-06-11 2019-09-17 深圳前海微众银行股份有限公司 一种信息处理方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8799150B2 (en) * 2009-09-30 2014-08-05 Scorelogix Llc System and method for predicting consumer credit risk using income risk based credit score
CN107909178B (zh) * 2017-08-31 2021-06-08 深圳壹账通智能科技有限公司 电子装置、失联修复率预测方法和计算机可读存储介质
CN108256691A (zh) * 2018-02-08 2018-07-06 成都智宝大数据科技有限公司 还款概率预测模型构建方法及装置
CN109558952A (zh) * 2018-11-27 2019-04-02 北京旷视科技有限公司 数据处理方法、系统、设备及存储介质
CN109858749A (zh) * 2018-12-26 2019-06-07 广东电网有限责任公司 一种基于客户征信的催费方法及系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170270569A1 (en) * 2007-04-09 2017-09-21 Yellowpages.Com Llc Systems and methods to provide connections via callback acceptance cross-reference to related applications
CN106339482A (zh) * 2016-08-31 2017-01-18 五八同城信息技术有限公司 确定商家最佳沟通时间的方法及装置
CN106952155A (zh) * 2017-03-08 2017-07-14 深圳前海纵腾金融科技服务有限公司 一种基于信用评分的催收方法及装置
CN107528969A (zh) * 2017-07-27 2017-12-29 深圳市佰仟金融服务有限公司 电话拨打时间的管理方法、管理装置及终端设备
CN109559221A (zh) * 2018-11-20 2019-04-02 中国银行股份有限公司 基于用户数据的催收方法、装置和存储介质
CN110245985A (zh) * 2019-06-11 2019-09-17 深圳前海微众银行股份有限公司 一种信息处理方法及装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113989012A (zh) * 2021-09-17 2022-01-28 大箴(杭州)科技有限公司 不良资产的借款对象人群分类方法及装置、介质、设备

Also Published As

Publication number Publication date
CN110245985B (zh) 2023-09-08
CN110245985A (zh) 2019-09-17

Similar Documents

Publication Publication Date Title
US20200210899A1 (en) Machine learning model training method and device, and electronic device
WO2020248916A1 (zh) 一种信息处理方法及装置
US20180308160A1 (en) Risk assessment method and system
US20180374098A1 (en) Modeling method and device for machine learning model
US8756071B2 (en) Methods and apparatus for queue-based cluster analysis
US10521748B2 (en) Retention risk determiner
CN112116245A (zh) 信贷风险评估方法、装置、计算机设备及存储介质
US20050080821A1 (en) System and method for managing collections accounts
CN111951097A (zh) 企业信用风险评估方法、装置、设备及存储介质
CN111738819A (zh) 表征数据筛选方法、装置和设备
CN107909178B (zh) 电子装置、失联修复率预测方法和计算机可读存储介质
US11037236B1 (en) Algorithm and models for creditworthiness based on user entered data within financial management application
CN111062518B (zh) 基于人工智能的处理催收业务的方法、装置及存储介质
TW202008251A (zh) 支付工具推薦方法、裝置、設備及電腦可讀儲存媒體
CN111061948B (zh) 一种用户标签推荐方法、装置、计算机设备及存储介质
CN111008896A (zh) 金融风险预警方法、装置、电子设备及存储介质
CN114116998A (zh) 答复语句生成方法、装置、计算机设备和存储介质
CN114090601B (zh) 一种数据筛选方法、装置、设备以及存储介质
CN109754290B (zh) 一种游戏数据的处理方法和装置
WO2019194696A1 (ru) Автоматизированная система разработки и управления моделями скоринга
CN115293783A (zh) 风险用户识别方法、装置、计算机设备和存储介质
US20210118049A1 (en) Systems and methods for intelligently optimizing a queue of actions in an interface using machine learning
KR20160003262A (ko) 연체가 있는 사업체의 확률 계산
CN116362895A (zh) 理财产品推荐方法、装置及存储介质
CN113255929B (zh) 异常用户可解释原因的获取方法和装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20823009

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20823009

Country of ref document: EP

Kind code of ref document: A1