CN115423323A - Security management method and device, electronic equipment and computer storage medium - Google Patents

Security management method and device, electronic equipment and computer storage medium Download PDF

Info

Publication number
CN115423323A
CN115423323A CN202211078104.5A CN202211078104A CN115423323A CN 115423323 A CN115423323 A CN 115423323A CN 202211078104 A CN202211078104 A CN 202211078104A CN 115423323 A CN115423323 A CN 115423323A
Authority
CN
China
Prior art keywords
information
historical
risk
order
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.)
Granted
Application number
CN202211078104.5A
Other languages
Chinese (zh)
Other versions
CN115423323B (en
Inventor
奚久洲
马谱皓
汪安辉
齐峰
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.)
Zhejiang Koubei Network Technology Co Ltd
Original Assignee
Zhejiang Koubei Network Technology 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 Zhejiang Koubei Network Technology Co Ltd filed Critical Zhejiang Koubei Network Technology Co Ltd
Priority to CN202211078104.5A priority Critical patent/CN115423323B/en
Publication of CN115423323A publication Critical patent/CN115423323A/en
Application granted granted Critical
Publication of CN115423323B publication Critical patent/CN115423323B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06Q10/00—Administration; Management
    • G06Q10/08—Logistics, e.g. warehousing, loading or distribution; Inventory or stock management
    • G06Q10/083—Shipping

Abstract

The application provides a security management method, a security management device, an electronic device and a computer storage medium, wherein the method comprises the following steps: acquiring information of an order to be distributed; extracting information of the order to be distributed, and obtaining a risk value of the order to be distributed by the target distribution personnel by using a risk prediction model of the target distribution personnel; and sending the information of the orders to be distributed to other distribution personnel under the condition that the risk value of the target distribution personnel for distributing the orders to be distributed is larger than the risk threshold value. Therefore, the security risk problem possibly occurring between the target distribution personnel and the merchants and the users can be effectively avoided.

Description

Security management method and device, electronic equipment and computer storage medium
Technical Field
The present application relates to the field of computer technologies, and in particular, to a security management method and apparatus, an electronic device, and a computer storage medium.
Background
With the rapid development of the internet, people can do shopping, ordering, sending and receiving express and other activities on the internet, and meanwhile, the delivery personnel can take a delivery link between a merchant and a customer. However, in the distribution process, distribution personnel often struggle with merchants or customers due to factors such as slow shipment of merchants, improper distribution time, remote distribution location and the like, and may cause a public security problem in a serious case.
Disclosure of Invention
The embodiment of the application provides a security management method, a security management device, electronic equipment and a computer storage medium, so as to avoid the risk problem possibly occurring between target distribution personnel and merchants and users.
In a first aspect, an embodiment of the present application provides a security management method, where the method includes:
acquiring information of an order to be distributed;
extracting the information of the order to be distributed, and obtaining a risk value of the order to be distributed by the target distribution personnel by using a risk prediction model of the target distribution personnel; the risk prediction model of the target delivery personnel is obtained by training based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel; the information of each historical order comprises voice characteristic information;
under the condition that the risk value of the target delivery personnel delivering the orders to be distributed is larger than a risk threshold value, sending the information of the orders to be distributed to other delivery personnel; and the risk value of the other delivery personnel delivering the order to be distributed is less than or equal to the risk threshold value.
In a second aspect, an embodiment of the present application provides a method for building a risk prediction model, where the method includes:
acquiring a preset risk prediction initial model and acquiring information of historical orders of a plurality of known public security complaint results distributed by target distribution personnel; the information of each historical order comprises voice characteristic information;
and training the risk prediction initial model based on the voice characteristic information in the information of the plurality of historical orders of the plurality of known public security complaint results distributed by the target distribution personnel to obtain a risk prediction model.
In a third aspect, an embodiment of the present application provides a security management apparatus, where the apparatus includes:
the acquisition module is used for acquiring information of the order to be distributed;
the obtaining module is used for extracting the information of the order to be distributed and obtaining a risk value of the order to be distributed by the target distribution personnel by utilizing a risk prediction model of the target distribution personnel; the risk prediction model of the target delivery personnel is obtained by training based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel; the information of each historical order comprises voice characteristic information;
the distribution module is used for sending the information of the order to be distributed to other distribution personnel under the condition that the risk value of the target distribution personnel for distributing the order to be distributed is larger than a risk threshold value; and the risk value of the other delivery personnel delivering the orders to be distributed is less than or equal to the risk threshold.
In a fourth aspect, an embodiment of the present application provides an apparatus for building a risk prediction model, where the apparatus includes:
the system comprises an acquisition module, a risk prediction module and a risk prediction module, wherein the acquisition module is used for acquiring a preset risk prediction initial model and acquiring information of a plurality of historical orders of known public security complaint results which are delivered by target delivery personnel; the information of each historical order comprises voice characteristic information;
and the obtaining module is used for training the risk prediction initial model based on the voice characteristic information in the information of a plurality of historical orders of a plurality of known peace complaint results distributed by the target distribution personnel to obtain a risk prediction model.
In a fifth aspect, embodiments of the present application provide a computer storage medium storing a plurality of instructions adapted to be loaded by a processor and to perform the above-mentioned method steps.
In a sixth aspect, an embodiment of the present application provides an electronic device, which may include: a processor and a memory;
wherein the memory stores a computer program adapted to be loaded by the processor and to perform the above-mentioned method steps.
The embodiment of the application can be implemented by acquiring information of the order to be distributed; extracting information of the order to be distributed, and obtaining a risk value of the order to be distributed by the target distribution personnel by using a risk prediction model of the target distribution personnel; and under the condition that the risk value of the target delivery personnel delivering the order to be distributed is larger than the risk threshold value, sending the information of the order to be distributed to other delivery personnel. According to the risk prediction method and device, the risk prediction model can be trained according to the voice characteristic information in the information of the plurality of historical orders distributed by the target distributor, so that the problems that prediction risk values are not accurate and the like caused by the fact that emotions of the target distributor, the user and a merchant can not be accurately identified through the text information in the related technology are correspondingly reduced. In addition, when the risk value of the target delivery personnel is too large, the order to be delivered can be directly distributed to other delivery personnel with lower risk values, so that the possibly caused public security problem is avoided, and the good order of the society is effectively guaranteed.
Drawings
In order to more clearly illustrate the technical solutions in the present application, the drawings required for use in the embodiments will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present application, and it is obvious for those skilled in the art that other drawings can be obtained according to these drawings without creative efforts.
Fig. 1a is a schematic view of an application scenario of a related security management method;
FIG. 1b is a schematic diagram of a prediction process of a relevant risk prediction model;
fig. 2a is a schematic flow chart of a security management method provided in the present application;
FIG. 2b is a schematic diagram of a prediction process of a risk prediction model provided in the present application;
fig. 3a is a schematic flowchart of another security management method provided in the present application;
FIG. 3b is a schematic diagram of a prediction process of another risk prediction model provided herein;
fig. 4a is a schematic flowchart of another security management method provided in the present application;
fig. 4b is a schematic view of an application scenario of another security management method provided in the present application;
FIG. 5a is a schematic flow chart of a method for establishing a risk prediction model according to the present application;
FIG. 5b is a schematic diagram of the internal structure of another risk prediction model provided in the present application;
FIG. 5c is a schematic diagram of the internal structure of a differential evaluation prediction model provided in the present application;
fig. 6 is a schematic structural diagram of a security management apparatus according to an embodiment of the present application;
fig. 7 is a schematic structural diagram of an apparatus for building a risk prediction model according to an embodiment of the present application;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present disclosure;
fig. 9 is a schematic structural diagram of another electronic device according to an embodiment of the present application. .
Detailed Description
When the following description refers to the accompanying drawings, like numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary examples do not represent all implementations consistent with the present application. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the application, as detailed in the appended claims.
In the description of the present application, it is to be understood that the terms "first," "second," and the like are used for descriptive purposes only and are not to be construed as indicating or implying relative importance. The specific meaning of the above terms in the present application can be understood in a specific case by those of ordinary skill in the art. Further, in the description of the present application, "a plurality" means two or more unless otherwise specified. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship.
For service industries such as express delivery, take-away and the like, security risk prevention and control are important means for guaranteeing the benefits of industry participants. The related public security risk analysis method mainly starts from the public security risk attribute of a delivery person (such as a rider), and evaluates the public security risk threat of the delivery person to a current user by analyzing the long-term behavior pattern of the delivery person.
Fig. 1a is a schematic diagram illustrating an application scenario of a related security management method. As shown in fig. 1a, the application scenario may include: restaurant 11, take out 12, rider 13, customer 14, no peace complaint between customer and rider 15, and a peace complaint between customer and rider 16. It can be understood that after the restaurant 11 delivers the packaged takeout 12 to the rider 13, the rider 13 sends the packaged takeout 12 to the position of the customer 14 according to the order information, and after the customer 14 receives the takeout 12, if no public complaint is generated, the Application program (APP) of the takeout platform can display that the public complaint result 15 between the customer and the rider is good, that is, no public complaint exists; if the peace complaint is generated, the result 16 of the peace complaint between the customer and the rider can be displayed through the APP to indicate that the peace complaint exists, namely, an alarm event caused by the fact that the rider attacks the customer and the rider takes an action, and further relevant information of the alarm event is stored in a database of a server of the takeout platform APP.
Fig. 1b schematically shows a prediction process of the relevant risk prediction model. The preset model of the rider x can train the prediction model based on the order information of n historical users of the rider and the real public security complaint results of the order information. Specifically, the prediction model may be adjusted by comparing the public security complaint result corresponding to the predicted risk value of the order information of each historical user with the real public security complaint result until the public security complaint result corresponding to the predicted risk value is the same as the real public security complaint result, and then the current order of the user m is input to the prediction model to obtain the prediction result of the order. If the predicted result is that there is a peace complaint when the order of the user m is sent to the rider x, it can be considered that the order is sent to other riders whose predicted result is that there is no peace complaint, so as to avoid the possible peace risk.
However, the above security management method may cause the prediction model to be completed based on the public security complaint results of the historical users, and is often limited to the guarantee for platform users (such as ordering staff), and lacks the ability to guarantee for distribution staff and merchants. Therefore, a solution that can comprehensively analyze security risks of all parties in the fulfillment process and further can fully guarantee good relationship and security among distribution personnel, users and merchants is urgently needed.
Fig. 2a is a schematic flow chart of a security management method according to an embodiment of the present application. As shown in fig. 2a, the security management method may include the steps of:
step 201, obtaining information of an order to be distributed.
The to-be-distributed order is an order which is not distributed to the distribution personnel, and the information of the to-be-distributed order may include: order number, merchant name, commodity picture, commodity purchase time, commodity amount, receiving address, delivery time and other information.
For example, after a set of items is purchased at restaurant A on the user litter take out platform, the server of the take out platform generates order information for the user litter that has not been assigned to the deliverer.
And step 202, extracting information of the order to be distributed, and obtaining a risk value of the order to be distributed by the target distribution personnel by using a risk prediction model of the target distribution personnel.
The risk prediction model of the target delivery personnel in the embodiment of the application can be obtained by training based on information of historical orders of a plurality of known public security complaint results delivered by the target delivery personnel. Wherein the information of each historical order may include voice characteristic information.
It is to be understood that the target delivery person in the embodiment of the present application indicates a delivery person to which the server intends to distribute the order of the user. The risk prediction model of the target delivery person represents a model for predicting the magnitude of the risk that the target delivery person may deliver the order of the user.
Specifically, in the embodiment of the present application, the voice feature information in the information of the historical order may represent feature information extracted from a recording and/or a voice chat record of a call between a target delivery person and a merchant and/or between the target delivery person and a user corresponding to the historical order. For example, the voice feature information may be feature information extracted from a call recording generated by a target delivery person dialing a telephone to a merchant or a user through the delivery platform APP, or feature information extracted from a voice chat recording generated by the target delivery person and the merchant or the user in the delivery platform APP. It can be understood that the voice characteristic information corresponding to the historical order is acquired from the distribution platform, so that privacy of three parties such as target distribution personnel, merchants and users can be effectively protected, and data leakage is prevented.
Possibly, in some embodiments of the present application, it may be unnecessary to extract voice feature information, and to directly use the recording of the call and/or the voice chat between the target delivery person and the merchant and/or the target delivery person and the user.
Possibly, the embodiment of the application may acquire information of historical orders within a preset time period of the target delivery personnel from the database, for example, acquire information of orders delivered by the target delivery personnel in a period of approximately 3 months.
Refer to fig. 2b, which is a schematic diagram illustrating a prediction process of the risk prediction model according to the embodiment of the present application. According to the method and the device, the risk prediction model of the target delivery personnel can be trained by using the voice characteristic information in the information of the historical orders of n (n > 1) known peace complaint results, specifically, the generated risk prediction result and the known peace complaint results can be compared to train the risk prediction model, and after the training is finished, the risk value of the target delivery personnel when the target delivery personnel deliver the order m to be distributed can be predicted by using the risk prediction model of the target delivery personnel.
And step 203, comparing the magnitude between the risk value of the target delivery personnel output by the risk prediction model of the target delivery personnel and the risk threshold.
Specifically, the risk threshold in the embodiment of the present application indicates a preset value for distinguishing the risk result.
Step 2031, when the risk value of the target distribution personnel for distributing the order to be distributed is greater than the risk threshold, sending the information of the order to be distributed to other distribution personnel.
Specifically, the risk value of the other delivery personnel delivering the order to be dispensed needs to be less than or equal to the risk threshold.
For example, if the risk threshold can be 0.7, then a risk value greater than 0.7 can be understood as a risk problem that may occur when allocating an order to be allocated to target distribution personnel, such as a quarrel, abusive event, or a mutual attack, fighting-induced bleeding event between the rider and the merchant. It is desirable to allocate the order to be allocated to other distribution personnel with a risk value of less than 0.7 to reduce the likelihood of the above-mentioned malignancy.
It is understood that the risk values of other delivery personnel in the embodiment of the present application are determined in a manner similar to the risk values of the target delivery personnel. For example, when the risk value of the order to be distributed delivered by the rider a (target delivery person) output by the risk prediction model of the rider a is greater than the risk threshold value, the order to be distributed is input into the risk prediction model of the rider B (other delivery person), and if the output risk value of the order to be distributed delivered by the rider B is less than or equal to the risk threshold value, the order to be distributed can be distributed to the rider B.
Step 2032, when the risk value of the target delivery personnel delivering the order to be distributed is less than or equal to the risk threshold value, sending the information of the order to be distributed to the target delivery personnel.
It is to be appreciated that assigning the order to be allocated to the targeted delivery personnel may be considered not to create a security issue when the risk value of the targeted delivery personnel delivering the order to be allocated is less than or equal to the risk threshold.
In particular, from the schematic diagram of the prediction process of the risk prediction model shown in fig. 2b, it can be observed that: after the risk prediction model of the target delivery personnel predicts the risk value when the target delivery personnel deliver the order m to be distributed, if the obtained risk value of the order m to be distributed is greater than the risk threshold value, the order m to be distributed can be sent to other delivery personnel; and if the obtained risk value of the order m to be distributed is less than or equal to the risk threshold value, sending the order m to be distributed to the target delivery personnel.
For example, the user smalli generates a piece of information of an order to be allocated after a set of orders is issued by a restaurant a in a certain takeout platform APP, and a server of the takeout platform APP inputs the information of the order to be allocated into a risk prediction model of a nearby rider a to obtain a risk value of 0.8, wherein the risk value is greater than a risk threshold value of 0.7. Therefore, the server of the takeout platform APP needs to continuously determine the risk value of the next other rider, if it is determined that the next target rider is the rider B, the information of the order to be allocated can be input into the risk prediction model of the rider B to obtain the risk value of 0.3, and the risk value is smaller than the risk threshold value of 0.7, so that the server of the takeout platform APP can send the information of the order to be allocated to the mobile terminal of the rider B, and the information is distributed by the rider B.
According to the embodiment of the application, risk prediction can be performed on the target delivery personnel for delivering the to-be-distributed orders through a pre-established risk prediction model so as to determine a risk value possibly generated by the target delivery personnel for delivering the to-be-distributed orders. According to the risk prediction method and device, the risk prediction model can be trained according to the voice characteristic information in the information of the plurality of historical orders distributed by the target distributor, so that the problems that prediction risk values are not accurate and the like caused by the fact that emotions of the target distributor, the user and a merchant can not be accurately identified through the text information in the related technology are correspondingly reduced. In addition, when the risk value of the target delivery personnel is too large, the order to be delivered can be directly distributed to other delivery personnel with lower risk values, so that the possibly caused public security problem is avoided, and the good order of the society is effectively guaranteed.
Fig. 3a is a schematic flow chart of a security management method according to an embodiment of the present application. As shown in fig. 3a, the security management method may include the steps of:
step 301, obtaining information of historical orders of a plurality of known public security complaint results which are delivered by target delivery personnel.
Specifically, the information of each historical order in the embodiment of the present application may include: the method comprises the steps of obtaining characteristic information of target delivery personnel, characteristic information of historical orders, characteristic information of users corresponding to the historical orders, characteristic information of merchants corresponding to the historical orders and voice characteristic information corresponding to the historical orders.
In the embodiment of the present application, the characteristic information of the target delivery staff in the information of the historical order may include: attribute information of the target delivery person (for example, name of rider, native place, time length of work, number of picked-up sheets, etc.) and evaluation information of the target delivery person (for example, number of bad comments). The characteristic information of the historical order in the information of the historical order may include: the time of placing the order, the time of delivery, the meal fee, the delivery fee, the payment method and other information of the historical order. The characteristic information of the user corresponding to the historical order may include: the native place, the diet hobby, the character feature, the peak time of getting off the order and the like of the user. The voice feature information corresponding to the historical order may include: the method comprises the following steps of recording conversation between a target distributor and a merchant or a user, recording voice chatting between the target distributor and the merchant or the user in a distribution platform APP, and the like. The characteristic information of the merchant represents information that the merchant produces due to its own attributes, such as business hours, business years, geographic location, shipping efficiency, product type, rush hour, and the like.
Step 302, training a risk prediction model of the target delivery personnel based on information of a plurality of historical orders of known peace complaint results that the target delivery personnel completed delivery.
It is understood that the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the users corresponding to the historical orders, and the characteristic information of the merchants corresponding to the historical orders in the historical order information may all affect the delivery process, for example, the geographic location of the merchant is remote, and the actual location of the target delivery personnel is not clear although the target delivery personnel is nearby, so that the delivery time may be affected, thereby possibly causing unnecessary disputes between the target delivery personnel and the users.
Referring to fig. 3b, which is a schematic diagram of a prediction process of the risk prediction model in the embodiment of the present application, feature information of target distribution personnel, feature information of a historical order, feature information of a user corresponding to the historical order, and voice feature information corresponding to the historical order in information of n (n > 1) historical orders with known police complaint results may be used to train the risk prediction models of the feature information of the target distribution personnel and merchants corresponding to the historical order to obtain a highly accurate risk prediction probability of the peace, so that a risk value when the target distribution personnel distribute an order m to be distributed is obtained by using the trained risk prediction model of the target distribution personnel.
Step 303, obtaining information of the order to be distributed.
Specifically, step 303 is identical to step 201, and is not described herein again.
And step 304, extracting the information of the order to be distributed, and obtaining the risk value of the order to be distributed by the target distribution personnel by using the risk prediction model of the target distribution personnel.
Specifically, step 304 is identical to step 202, and is not described herein again.
And step 305, comparing the magnitude between the risk value of the target delivery personnel output by the risk prediction model of the target delivery personnel and the risk threshold.
Specifically, step 305 is identical to step 203, and is not described herein again.
Step 3051, sending the information of the order to be distributed to other distribution personnel when the risk value of the target distribution personnel for distributing the order to be distributed is greater than the risk threshold value.
Specifically, step 3051 is identical to step 2031, and is not described herein again.
Step 3052, sending the information of the order to be distributed to the target distribution personnel when the risk value of the target distribution personnel for distributing the order to be distributed is smaller than or equal to the risk threshold.
Specifically, step 3052 is identical to step 2032, and is not described herein again.
Referring to fig. 3b, further, if the risk value of the order m to be allocated is greater than the risk threshold, the order m to be allocated may be sent to other delivery personnel; and if the obtained risk value of the order m to be distributed is smaller than or equal to the risk threshold value, sending the order m to be distributed to the target distribution personnel.
Therefore, the risk prediction model can be trained in a mode of fusing characteristic information of target delivery personnel, characteristic information of historical orders, characteristic information of users corresponding to the historical orders, characteristic information of merchants and voice characteristic information corresponding to the historical orders in the information of the historical orders of the rider. Accordingly, compared with the prior art in which only the user is concerned about the prediction probability obtained by the evaluation information of the historical order, the method and the device for obtaining the historical order are more accurate, and in addition, when the risk value of the target delivery personnel is overlarge, the order to be delivered can be directly distributed to other delivery personnel with lower risk values, so that the possibly caused security problem is avoided, and the good order of the society is effectively guaranteed.
Fig. 4a is a schematic flowchart of a security management method according to an embodiment of the present application. As shown in fig. 4a, the security management method may include the steps of:
step 401, obtaining information of a plurality of historical orders of known police complaint results which are delivered by the target delivery staff, and information of a plurality of historical orders of known attitude evaluation results which are delivered by the target delivery staff.
The evaluation results of a plurality of known attitudes of the target delivery personnel completing delivery may include: good results and poor results. Possibly, the attitude evaluation result may be an evaluation of the target delivery person by the merchant and an evaluation of the target delivery person by the user.
Referring to fig. 4b, an application scenario diagram of the security management method in the embodiment of the present application is shown. The application scenario may include: restaurant 41, take out 42, rider 43, customer 44, restaurant's evaluation of the rider 45, and customer's evaluation of the rider 46. It will be appreciated that after the restaurant 41 delivers the packaged takeaway 42 to the rider 43, the rider's service of taking food is evaluated by the APP of the takeaway platform, and the evaluation result is a poor evaluation 45, and then the rider 43 is dispatched to the location of the customer 44 according to the order information, and the customer 44 can evaluate the service of the rider 43 by the APP of the takeaway platform, and the evaluation result is a good evaluation 46.
Step 402, training a risk prediction model of the target delivery personnel based on information of a plurality of historical orders of known police complaint results that the target delivery personnel finish delivering and information of a plurality of historical orders of known attitude evaluation results that the target delivery personnel finish delivering.
It is understood that the risk value of the target delivery person delivering the order to be dispensed may also be increased when the number of bad comments made by the target delivery person within the past time preset time period is too large. For example, rider a has had a large fluctuation in mood over the past week and therefore received a large number of bad comments, which likely resulted in a subsequent order being disputed with the store or customer during distribution due to trivia, and a serious condition could attack the store or customer.
Therefore, in the embodiment of the present application, on the basis of the information of the historical orders of the known police complaint results that the standard distribution staff completes distribution, the information of the historical orders of the known attitude evaluation results is comprehensively considered to train the risk prediction model of the target distribution staff more effectively, so as to obtain a more accurate risk value of the target distribution staff for distributing the orders to be distributed.
Step 403, obtaining information of the order to be distributed.
Specifically, step 403 is identical to step 201, and is not described herein again.
And step 404, extracting the information of the order to be distributed, and obtaining a risk value of the order to be distributed by the target distribution personnel by using the risk prediction model of the target distribution personnel.
Specifically, step 404 is identical to step 202, and is not described herein again.
And step 405, comparing the risk value of the target delivery personnel output by the risk prediction model of the target delivery personnel with the risk threshold.
Specifically, step 405 is identical to step 203, and is not described herein again.
Step 4051, when the risk value of the target delivery person delivering the order to be delivered is greater than the risk threshold, sending the information of the order to be delivered to other delivery persons.
Specifically, step 4051 corresponds to step 2031, and is not described herein again.
Step 4052, when the risk value of the target delivery staff delivering the order to be distributed is less than or equal to the risk threshold, sending the information of the order to be distributed to the target delivery staff.
Specifically, step 4052 is identical to step 2032, and is not described here.
For example, in the embodiment of the present application, the risk prediction models of the feature information of the target delivery personnel, the feature information of the historical orders, the feature information of the users corresponding to the historical orders, and the voice feature information corresponding to the historical orders in the information of the historical orders of n (n > 1) known police complaint results and m (m > 1) known attitude evaluation results may be used to train the risk prediction models of the feature information of the target delivery personnel and the merchants corresponding to the historical orders to obtain a highly accurate risk prediction probability of the peace, so that the trained risk prediction model of the target delivery personnel is used to obtain a risk value when the target delivery personnel delivers an order m to be distributed. Further, if the obtained risk value of the order m to be distributed is greater than the risk threshold value, the order m to be distributed can be sent to other distribution personnel; and if the obtained risk value of the order m to be distributed is less than or equal to the risk threshold value, sending the order m to be distributed to the target delivery personnel.
Therefore, in the embodiment of the application, the risk prediction model of the target delivery personnel can be trained more effectively by comprehensively considering the information of the historical orders of a plurality of known police complaint results and the information of the historical orders of a plurality of known attitude evaluation results, which are delivered by the target delivery personnel, so that the output results of the public security risk prediction model of the target delivery personnel are trained more accurately to obtain the risk value with higher accuracy.
Fig. 5a is a schematic flow chart of a method for establishing a risk prediction model according to an embodiment of the present disclosure. As shown in fig. 5a, the security management method may include the steps of:
s501, a preset risk prediction initial model is obtained, and information of historical orders of a plurality of known public security complaint results, which are delivered by target delivery personnel, is obtained.
The information of each historical order can include voice characteristic information corresponding to the historical orders, which can be information such as call records of target distribution personnel and merchants or users, and voice chat records of the target distribution personnel and the merchants or the users in the distribution platform APP.
Possibly, in the embodiment of the present application, the voice feature information corresponding to the historical order within the preset time period of the target delivery person may be obtained from the database of the server, for example, the voice feature information corresponding to the order that the target delivery person completes delivery within the last half year is obtained.
S502, training the risk prediction initial model based on the voice characteristic information in the information of the plurality of historical orders of the plurality of known public security complaint results distributed by the target distribution personnel to obtain a risk prediction model.
For example, the risk prediction model of the target delivery person may be trained using the voice feature information in the information of the historical orders of n (n > 1) known peace complaint results, and specifically, the risk prediction result generated by the risk prediction model and the known peace complaint results may be trained by comparing them.
Further, the risk prediction initial model can be trained based on the voice feature information in the information of the historical orders of a plurality of known public security complaint results distributed by the target distribution personnel, so that the risk value of each historical order is obtained; determining a loss value of the risk prediction initial model based on the risk value of each historical order and the peace complaint result of each historical order; under the condition that the loss value is larger than the loss threshold value, adjusting parameters of the risk prediction initial model, and updating the risk prediction initial model; and performing voice characteristic information in the information of a plurality of historical orders based on a plurality of known peace complaint results which are delivered by the target delivery personnel, training the risk prediction initial model, and obtaining the risk value of each historical order until the loss value is less than or equal to the loss threshold value, thereby obtaining the risk prediction model.
It can be understood that the risk prediction initial model can be trained by comparing the loss value of the risk prediction initial model with the loss threshold value. For example, assuming that the loss value output by the risk prediction model is 0.95 and the preset loss threshold is 0.3, since the loss value output by the risk prediction initial model is 0.95> 0.3, the internal parameters of the risk prediction initial model need to be adjusted, and then the pair of risk prediction initial models are continuously trained until the loss value output by the risk prediction initial model is less than or equal to 0.3, so as to obtain the trained risk prediction model.
Further, in this embodiment, the information about each historical order may further include: the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the users corresponding to the historical orders and the characteristic information of the merchants corresponding to the historical orders.
Therefore, in the embodiment of the application, the risk prediction initial model may be trained based on feature information of the target distributor, feature information of the historical order, feature information of a user corresponding to the historical order, feature information of a merchant corresponding to the historical order, and voice feature information in the information of the historical orders of the plurality of known public security complaint results that the target distributor completes distribution, so as to obtain the risk value of each historical order.
Possibly, the method and the device for evaluating the attitude of the target distribution personnel can acquire the information of the historical orders of a plurality of known attitude evaluation results of the target distribution personnel completing distribution; training a risk prediction initial model based on the information of a plurality of historical orders of which the target delivery personnel finish delivery and have known attitude evaluation results, and obtaining the risk value of each historical order, wherein the information of the historical orders is the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of users corresponding to the historical orders, the characteristic information of merchants corresponding to the historical orders and the voice characteristic information corresponding to the historical orders.
Specifically, in the embodiment of the present application, the evaluation results of a plurality of known attitudes of the target delivery staff completing delivery may include: good results and poor results. The attitude evaluation result can be the evaluation result of the merchant to the target delivery personnel and the evaluation result of the user to the target delivery personnel. It will be appreciated that the risk value of the target delivery person delivering the order to be dispensed may also be increased if the target delivery person has gained too many bad reviews over a preset period of time in the past.
Specifically, the method and the device for determining the historical risk representation vector matrix of the target delivery personnel can determine the characteristic vector matrix of the target delivery personnel and the historical risk representation vector matrix of the target delivery personnel according to the characteristic information of the target delivery personnel in the historical order information; determining a characteristic vector matrix of a user and a historical risk characterization vector matrix of the user according to the characteristic information of the user corresponding to the historical order; determining a characteristic vector matrix of a merchant and a historical risk representation vector matrix of the merchant according to the characteristic information of the merchant corresponding to the historical order; determining a feature vector matrix of the historical order according to feature information of the historical order; determining an emotional state characterization vector matrix of a target delivery person according to voice characteristic information corresponding to a historical order; training the risk prediction initial model based on the feature vector matrix of the target person, the feature vector matrix of the user, the feature vector matrix of the merchant, the feature vector matrix of the historical order, the historical risk characterization vector matrix of the target delivery person, the historical risk characterization vector matrix of the merchant, the historical risk characterization vector matrix of the user, the emotional state characterization vector matrix of the target delivery person, and the evaluation result and the peace complaint result corresponding to the historical order corresponding to each historical order to obtain the risk value of each historical order.
Possibly, the embodiment of the present application may convert text features (for example, delivery duration, number of bad comments, number of public security complaints, and the like of the target delivery staff) included in the feature information of the target delivery staff in the historical order information into a feature vector matrix of the target delivery staff through Natural Language Processing (NLP). And converting text characteristics (such as eating habits, characters, common meal ordering time and the like of the customer) contained in the characteristic information of the user corresponding to the historical order into a characteristic vector matrix of the user through NLP. And converting text characteristics (such as business years, types of food, business hours, average meal times and the like of the merchants) contained in the characteristic information of the merchants corresponding to the historical orders into a characteristic vector matrix of the merchants through NLP. The text features (such as the generation time of the order, the amount of the meal, the delivery fee, the delivery time, the receiving address and the like) contained in the feature information of the historical order are converted into a feature vector matrix of the historical order through the NLP. Specifically, the text features contained in the feature information may be converted into a vector matrix form that can be identified by the risk prediction model in a Word Embedding (Word Embedding) or distributed vector (distributing) manner in the embodiment of the present application.
In this embodiment of the present application, the determining the eigenvector matrix of the target distribution staff and the historical risk characterization vector matrix of the target distribution staff according to the characteristic information of the target distribution staff in the historical order information may include: determining a characteristic vector matrix of target delivery personnel corresponding to the historical order according to the characteristic information of the target delivery personnel in the historical order information; and inputting the characteristic information of the target delivery personnel in the historical order information into a preset first depth residual error network to obtain a historical risk characterization vector matrix of the target delivery personnel corresponding to the historical order. In this embodiment of the present application, the determining the feature vector matrix of the user and the historical risk characterization vector matrix of the user according to the feature information of the user corresponding to the historical order may include: determining a characteristic vector matrix of a user corresponding to the historical order according to the characteristic information of the user corresponding to the historical order; and inputting the characteristic information of the user corresponding to the historical order into a preset second depth residual error network to obtain a historical risk characterization vector matrix of the user corresponding to the historical order. In this embodiment of the application, the determining the characteristic vector matrix of the merchant and the historical risk characterization vector matrix of the merchant according to the characteristic information of the merchant corresponding to the historical order may include: determining a feature vector matrix of a merchant corresponding to the historical order according to feature information of the merchant corresponding to the historical order; and inputting the characteristic information of the merchant corresponding to the historical order into a preset third depth residual error network to obtain a historical risk characterization vector matrix of the merchant corresponding to the historical order.
Wherein the first depth residual network may comprise an N-layer structure; the input information of the nth layer structure may include output information of the N-1 th layer structure and input information of the first layer structure; n is an integer greater than 2; each layer in the first depth residual error network is preset with historical risk information of users needing to be extracted. The second depth residual network may comprise an M-layer structure; the input information of the M-th layer structure includes output information of the M-1-th layer structure and input information of the first layer structure; m is an integer greater than 2; and presetting historical risk information of the user to be extracted by each layer in the second depth residual error network. The third depth residual network may comprise a P-layer structure; the input information of the P-th layer structure comprises output information of the P-1-th layer structure and input information of the first layer structure; p is an integer greater than 2; and presetting historical risk information of the merchants needing to be extracted by each layer in the third depth residual error network. Wherein, the corresponding numbers of N, M and P can be the same or different.
See fig. 5b for a schematic diagram of the internal structure of the risk prediction model according to the embodiment of the present application. The first, second, and third depth residual networks as in fig. 5b may comprise a two-layer residual structure Res & LN. For example, for a first layer of residual structure of the first depth residual network, the first layer of residual structure may be used to extract personality information of the target delivery personnel, and a second layer of residual structure may be used to extract meal average time and work experience information of the target delivery personnel. Specifically, the first layer of residual structure is Res & LN including two rectangular boxes, and the second layer of residual structure is Res & LN including three rectangular boxes, it can be understood that the rectangular box with the smaller gray value on the right side of the first layer of residual structure represents the input value obtained by encoding according to the feature information, and the rectangular box with the larger gray value on the left side of the first layer of residual structure represents the output value of the first layer of residual structure. The two small rectangular boxes on the right side of the second layer of residual error structure respectively represent an input value and an output value of the first layer of residual error structure, and it can be understood that the purpose of inputting the input value and the output value of the first layer of residual error structure into the second layer of residual error structure in the embodiment of the present application is to avoid loss of characteristic risk information, that is, by using the residual error structure, it is ensured that preset characteristic risk information is not lost layer by layer in a multi-layer depth network (for example, the first layer of residual error structure of the first layer of residual error network only concerns the character information of a target distributor, so that other information which can be further extracted, such as the meal delivery average time and work experience information of the target distributor, is lost), and the last output historical risk feature vector is inaccurate. Therefore, the input information of each layer of residual structure in the embodiment of the present application is the output information of the previous layer of residual structure plus the input information of the first layer of residual structure, i.e. the initial feature information.
Specifically, the voice feature information corresponding to the historical order in the embodiment of the present application may include sound information and noise information of the target delivery personnel. Further, in the embodiment of the present application, the determining the emotion state characterization vector matrix of the target delivery person according to the voice feature information corresponding to the historical order may include: converting a time domain signal corresponding to the voice information corresponding to the historical order into a frequency domain signal corresponding to the voice characteristic information corresponding to the historical order; performing cepstrum analysis on the frequency domain signal corresponding to the voice feature information corresponding to the historical order to determine a vector matrix corresponding to the voice information of the target delivery personnel in the voice feature information corresponding to the historical order; and determining an emotional state characterization vector matrix of the target delivery personnel based on the vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order.
Referring to fig. 5b, the voice feature information corresponding to the historical order may include: the voice information of the merchant, the voice information of the user, the voice information of the target delivery personnel, noise and the like. Specifically, the embodiment of the application can perform emotional analysis and rider risk condition identification on the voice characteristic information to determine dangerous emotional information such as abuse, attack, quarrel, harassment and the like which can be carried in a risk telephone.
Specifically, in the embodiment of the application, the sound information of the target delivery staff in the voice feature information corresponding to the historical order may be directly obtained through the preset sound channel, but it is inevitable that the specified sound information directly obtained from the preset sound channel may carry noise information.
Possibly, in the field of sound processing, mel-Frequency Cepstral coefficients (MFCCs) are obtained based on human ear perception experiments, which consider the human ear as a specific filter that retains the low-Frequency part corresponding to the human vocal cord features and filters the high-Frequency part caused by noise. Therefore, the embodiment of the application can determine the voice information of the target delivery personnel through the MFCC.
Specifically, in the embodiment of the application, the time domain signal corresponding to the sound information needs to be first converted into the mel frequency domain signal for further separation, and then the mel frequency domain signal is subjected to cepstrum analysis to extract envelope data corresponding to the sound information of the target distribution staff. Specifically, the envelope data in the frequency domain information corresponding to the sound information of the target delivery personnel (i.e., the amplitude data in the spectrogram corresponding to the mel frequency domain signal) may be filtered to determine the sound information of the target delivery personnel in the voice feature information corresponding to the historical order. Possibly, the envelope data of the embodiment of the present application may be 13-dimensional feature, 157 time-series serialized data, that is, 13 × 157 vector matrix.
Possibly, the embodiment of the application can embed the envelope data into a voice emotion sub-network, and the voice emotion sub-network can be used for identifying information with large emotion fluctuation in the voice feature information of the target rider.
Specifically, in the embodiment of the present application, the determining an emotional state characterization vector matrix of the target delivery person based on the vector matrix corresponding to the sound information of the target delivery person in the voice feature information corresponding to the historical order may include: performing forward time sequence division on a vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order to obtain a first time sequence voice vector matrix; performing reverse time sequence division on a vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order to obtain a second time sequence voice vector matrix; weighting the first time sequence voice vector matrix and the second time sequence voice vector matrix to obtain a voice frequency strong correlation vector matrix; and determining an emotional state characterization vector matrix for the target delivery personnel based on the voice frequency strong correlation vector matrix and the vector threshold.
Referring to fig. 5b, the self-network of speech emotion can include a two-way Gated Recurrent Unit (GRU) time-series neural network, a multi-head attention mechanism, and a pooling layer. Specifically, the bidirectional GRU time-series neural network is to perform one-time segmentation on 157 time-series serialized data in envelope data according to a forward time series to obtain a first time-series voice vector matrix, perform one-time segmentation on a reverse time series to obtain a second time-series voice vector matrix, and perform weighting processing (i.e., a multi-head attention mechanism) on the first time-series voice vector matrix and the second time-series voice vector matrix to obtain new 13-dimensional feature and 157 time-series serialized data, i.e., a voice frequency strong correlation vector matrix.
Possibly, the accuracy of the historical risk characterization vector matrix of the target distribution personnel and the accuracy of the emotion state characterization vector matrix of the target distribution personnel can be improved by means of fusing the historical risk characterization vector matrix of the target distribution personnel and the emotion state characterization vector matrix of the target distribution personnel, and the accuracy of a risk prediction model of the target distribution personnel is further improved.
Specifically, the method and the device can perform inner product processing on the historical risk characterization vector matrix of the target distribution personnel and the emotional state characterization vector matrix of the target distribution personnel to obtain the weight corresponding to the historical risk characterization vector matrix of the target distribution personnel; determining historical risk fusion vectors of the target distribution personnel based on weights corresponding to the historical risk characterization vector matrixes of the target distribution personnel; performing inner product processing on the emotion state characterization vector matrix of the target distribution personnel and the historical risk characterization vector matrix of the target distribution personnel to obtain the weight corresponding to the emotion state characterization vector matrix of the target distribution personnel; determining an emotional state fusion vector of the target distribution personnel based on the weight corresponding to the emotional state characterization vector matrix of the target distribution personnel; training a risk prediction initial model based on a feature vector matrix of a target distributor, a feature vector matrix of a user, a feature vector matrix of a merchant, a feature vector matrix of a historical order, a historical risk fusion vector of the target distributor, a historical risk characterization vector matrix of the merchant, a historical risk characterization vector matrix of the user, an emotional state fusion vector of the target distributor, and an evaluation result and a peace complaint result corresponding to the historical order to obtain a risk value of each historical order.
Possibly, the embodiment of the application may adopt a feature mutual attention mechanism to enable the historical risk characterization vector matrix of the target distribution staff and the emotional state characterization vector of the target distribution staff to be fused with each other, that is, to perform inner product processing on the historical risk characterization vector matrix of the target distribution staff and the emotional state characterization vector matrix of the target distribution staff.
It should be noted that, in a case that dimensions of the historical risk characterization vector matrix of the target distributor and the emotional state characterization vector matrix of the target distributor are different, the dimensions of the historical risk characterization vector matrix and the emotional state characterization vector matrix of the target distributor need to be converted into the same through pooling action to perform inner product processing, for example, the emotional state characterization vector matrix of the target distributor output by the multi-head attention mechanism in fig. 5c is a 13 × 157 vector matrix, and if the historical risk characterization vector matrix of the target distributor is a 13-dimensional characterization vector, the emotional state characterization vector matrix of the target distributor needs to enter a pooling layer to perform pooling processing. Specifically, three different processing modes, namely maximum pooling (max), minimum pooling (min), and average pooling (mean pooling), can be selected for different dimensions in the 13 × 157 vector matrix, and the specifically selected mode needs to consider the meaning represented by the target dimension information, for example, if the second dimension in the emotional state representation vector matrix of the target distributor represents an open voice, the dimension needs to be processed by minimum pooling (min pooling).
Possibly, inner product processing A.B is carried out on the historical risk characterization vector matrix A of the target delivery personnel and the emotional state characterization vector matrix B of the target delivery personnel to obtain a weight a corresponding to the historical risk characterization vector matrix of the target delivery personnel, and furthermore, the weight a corresponding to the historical risk characterization vector matrix of the target delivery personnel is multiplied by the historical risk characterization vector matrix A of the target delivery personnel to obtain a historical risk fusion vector a multiplied by A of the target delivery personnel. And performing inner product processing B & A on the emotional state characterization vector matrix B of the target distribution personnel and the historical risk characterization vector matrix A of the target distribution personnel to obtain a weight B corresponding to the emotional state characterization vector matrix of the target distribution personnel, and further multiplying the weight B corresponding to the emotional state characterization vector matrix of the target distribution personnel by the emotional state characterization vector matrix B of the target distribution personnel to obtain an emotional state characterization vector matrix bxB of the target distribution personnel.
It can be understood that, in the embodiment of the present application, the weight a corresponding to the historical risk characterization vector matrix of the target delivery person is multiplied by the historical risk characterization vector matrix a of the target delivery person to obtain the historical risk fusion vector a × a of the target delivery person, which aims to weaken an objective vector in the historical risk characterization vector matrix (e.g., characteristics of eating and eating habits of the target delivery rider, delivering food to a user with the left hand or the right hand, and the like), and strengthen a subjective vector in the historical risk characterization vector matrix (e.g., characteristics of whether to deliver food to the home, a habit threat platform, a portal harassing client, and the like). The purpose of multiplying the weight B corresponding to the emotional state characterization vector matrix of the target delivery person by the emotional state characterization vector matrix B of the target delivery person to obtain the emotional state characterization vector matrix B × B of the target delivery person is to strengthen offensive features (e.g., features of quarrel, abuse, offensive language, etc.) in the emotional state characterization vector matrix of the target delivery person and weaken non-offensive features (e.g., features of laughing, disappointing, injuring, crying, etc.) in the emotional state characterization vector matrix of the target delivery person.
Specifically, the risk prediction initial model may be trained based on a feature vector matrix of a target distributor, a feature vector matrix of a user, a feature vector matrix of a merchant, a feature vector matrix of a historical order, a historical risk fusion vector of the target distributor, a historical risk characterization vector matrix of the merchant, a historical risk characterization vector matrix of the user, and an emotional state fusion vector of the target distributor, which correspond to each historical order in a plurality of historical orders of the target distributor, so as to obtain a risk value of each historical order.
Further, the loss value of the risk prediction initial model may be determined based on the risk value of each historical order, the public security complaint result of each historical order, the feature vector matrix of the target delivery personnel, the historical risk characterization vector matrix of the target delivery personnel, the feature vector matrix of the user, and the historical risk characterization vector matrix of the user.
It is understood that the risk prediction model of each delivery person in the embodiment of the present application is the same in the initial stage because there is no historical order information. For example, duffel and sheetlet are both newly registered riders on takeaway platform a, so the initial risk prediction model for rider duffel and the initial risk prediction model for rider sheetlet are the same. Possibly, after the rider duel and the rider duel respectively complete 100 delivery orders, since the initial risk prediction model of the rider duel and the initial risk prediction model of the rider duel are trained according to the 100 delivery orders that are respectively completed by the initial risk prediction model of the rider duel and the characteristic information of merchants, customers and the like corresponding to the 100 delivery orders, parameters of each neural network (for example, a depth residual error network, a bidirectional GRU timing neural network and the like) inside the initial risk prediction model of the rider duel and the initial risk prediction model of the rider duel can be changed, and therefore the trained risk prediction model of the rider duel and the trained risk prediction model of the rider duel can be different.
Specifically, the classification loss value of the risk prediction initial model can be determined based on the risk value of each historical order and the public security complaint result of each historical order; determining a characteristic information loss value of a risk prediction initial model based on the characteristic vector matrix of the target distribution personnel, the historical risk characterization vector matrix of the target distribution personnel, the characteristic vector matrix of the user and the historical risk characterization vector matrix of the user; and determining a loss value of the risk prediction initial model based on the classification loss value of the risk prediction initial model and the characteristic information loss value of the risk prediction initial model.
Specifically, the embodiment of the present application may determine the distance between the risk value of each historical order and the result of the public security complaint; and weighting the distances between the risk values of the plurality of historical orders and the corresponding public security complaint results to obtain the classification loss value of the risk prediction initial model. Determining a loss value of a target delivery person based on a characteristic vector matrix of the target delivery person corresponding to any one of a plurality of historical orders and a historical risk characterization vector matrix of the target delivery person; determining a loss value of the user based on a feature vector matrix of the user corresponding to each order in the plurality of historical orders and a historical risk characterization vector matrix of the user; and determining the characteristic information loss value of the risk prediction initial model based on the loss value of the target distribution personnel and the loss value of the user.
Possibly, the embodiment of the application may calculate the loss value L1= flooding (∑ [ CE (y', y)) of the risk prediction initial model by the following formula]+KLDiv u,r (encoder||decoder),b)
Where, Σ [ CE (y', y)]+KLDiv u,r (encoder, decoder) represents a loss function, and flooding represents an optimization function of the loss function. b represents the loss mean value of the risk prediction initial model after training part of historical orders, for example, all historical ordersB represents the average loss value obtained by adding the loss values corresponding to 2000 historical orders and dividing the sum by 2000 after 2000 historical orders are trained by the risk prediction initial model. Sigma [ CE (y', y)]Represents the categorical loss value of the historical order, where y' represents the risk value, y represents the known peace risk outcome, and CE represents the cross entropy, i.e., the distance between the two probabilities. KLdiv u,r The method comprises the steps that (encoder, decoder) represents a characteristic reconstruction loss item based on information divergence (KL divergence) of a user and a characteristic reconstruction loss item based on KL divergence of a target distributor, u represents the user, r represents the target distributor, the encoder represents a characteristic vector matrix of the user and a characteristic vector matrix of the target distributor, the decoder represents a historical risk characterization vector matrix of the user and a historical risk characterization vector matrix of the target distributor, and the KL divergence is calculated to determine the difference between the characteristic vector matrix and the historical risk characterization vector matrix.
Specifically, the optimization function flooding of the loss function is calculated in a manner of L1= | L-b | + b. Wherein L represents a loss function Σ [ CE (y', y)]+KLDiv u,r (encoder, decoder). It can be understood that the function of the optimization function flood of the loss function is to avoid the risk that the training process of the risk prediction initial model falls into the local optimal solution. Specifically, during the training process of the risk prediction initial model, the loss of the risk prediction initial model may be reduced very quickly by some special types of historical orders (for example, a certain historical order may correspond to situations that a merchant has a slow meal, a customer often makes a bad comment to a rider, a call voice in a recent historical order of a rider often exits, and the like, so that the risk prediction initial model may predict an accurate security risk result without adjusting internal parameters of the risk prediction initial model). Therefore, the risk prediction initial model may be made to have a gradient decreasing toward the loss gradient generated by the historical orders (i.e. toward the internal parameters of the risk prediction initial model corresponding to a small part of the historical orders of special types) in order to achieve a lower loss value, but the information of the historical orders of special types is obviously different from that of most of the information of all the historical orders, and thus, the risk prediction initial model may be made to have a gradient decreasing toward the loss gradient generated by the historical orders (i.e. toward the internal parameters of the risk prediction initial model corresponding to a small part of the historical orders of special types), and the information of the historical orders of special types may be obviously different from that of the historical orders of all the special typesThe gradient that can result from these few special types of historical orders falls off making the initial model of risk prediction unable to achieve a minimum loss value for all historical orders by continuing to train most of the remaining historical orders. Therefore, when the loss value output by a certain special type of historical order is smaller than the loss mean value b, optimization is needed, and the loss mean value b is added on the basis of the absolute value of the difference between the loss value output by the historical order and the loss mean value b so as to predict the risk of the training process of the initial model falling into the local optimal solution. It can be understood that, in the embodiment of the present application, when determining the loss value of the initial risk prediction model, considering that if only a historical order known to generate a public security problem is taken as a training set, the training set may face a problem of too little historical order data in the training set, so that the trained risk prediction model also has a problem of overfitting, the embodiment of the present application adds a loss value of feature information, that is, the embodiment of the present application also considers a feature comparison result of historical order information known to generate a poor evaluation, because if too many poor evaluations exist in the historical orders of the target distribution staff in a short time, the public security risk problem may also be caused.
For example, all 100 historical orders of little plum on the rider in the last month can be obtained, the distance between the risk value output by the risk prediction initial model of each historical order and the actual security result can be calculated by using CE (y ', y), for example, the 5 th historical order is input into the risk prediction initial model, the risk value output by the risk prediction initial model is 0.9, the actual security result is that the security problem exists (namely, an alarm event such as the occurrence of an intentional report or mutual attack of the user is given), the distance between the risk value and the actual security result is 0.1, and the distance is calculated by sigma [ CE (y', y)]The sum of the distances between the risk values and the actual police results for all 100 historical orders, i.e., the categorical loss values for the historical orders, is determined. In addition, in the embodiment of the present application, it is further required to calculate feature reconstruction loss items of KL divergences of all users corresponding to 100 historical orders (due to the existence of repeated users, the number of users may be less than 100), that is, difference values between the feature vector matrix of each user and the historical risk characterization vector matrix, and feature reconstruction loss items of small rider plum, that is, small riderThe difference value between the historical risk characterization vector matrix of lie and the historical risk characterization vector matrix of lie, and further, the difference value between the feature vector matrix of each user and the historical risk characterization vector matrix of each user and the difference value between the historical risk characterization vector matrix of little lie ridden and the historical risk characterization vector matrix of little lie need to be summed to obtain KLDiv u,r (encoder,decoder)。
It will be appreciated that if Σ [ CE (y', y)]+KLDiv u,r If the result of (encoder) is greater than the loss threshold of 0.3, the neural network parameters inside the risk prediction initial model need to be adjusted until Σ [ CE (y', y)]+KLDiv u,r The result of (encoder, decoder) is less than or equal to the loss threshold 0.3. If Σ [ CE (y', y)]+KLDiv u,r The result of (encoder, decoder) is too small, for example 0.002, which is smaller than the loss mean value b =0.005, so that the loss function needs to be adjusted by using the optimization function of the loss function, i.e. L1= |0.002-0.005| +0.005=0.008, so that the risk that the training process of the risk prediction initial model falls into the locally optimal solution can be avoided.
In one specific example, information of an order of customer a that has just been completed by a small rider (target delivery person) and a corresponding security risk result (no security issue) of the order are obtained, wherein the information of the order includes: the characteristic information of the customer a corresponding to the order, the characteristic information of the restaurant (merchant) corresponding to the order, the characteristic information of the small rider, the voice characteristic information of the call between the small rider and the customer a, the voice characteristic information of the call between the small rider and the restaurant, and the characteristic information of the order. Inputting the information of the order of the customer a into a risk prediction model of the small rider, specifically, referring to fig. 5c, it is necessary to generate an order feature vector matrix based on the feature information of the order, generate a feature vector matrix of the customer a based on the feature information of the customer a, and obtain a historical risk feature vector matrix of the customer a by using a second depth residual error network; and generating a characteristic vector matrix of the small piece of the rider based on the characteristic information of the small piece of the rider, acquiring a historical risk characteristic vector matrix of the small piece of the rider by using a first depth residual error network, generating a characteristic vector matrix of a restaurant based on the characteristic information of the restaurant, and acquiring a historical risk characteristic vector matrix of the restaurant by using a third depth residual error network. And the voice characteristic information of the call between the small piece of the rider and the customer A and the voice characteristic information of the call between the small piece of the rider and the restaurant are processed to obtain an emotional state representation vector matrix of the small piece of the rider.
Further, the historical risk characterization vector matrix and the emotional state characterization vector matrix of the small rider can be fused through a self-attention mechanism to obtain the historical risk fusion vector matrix and the emotional state fusion vector of the small rider respectively, and further, the emotional state fusion vector can be received through a drop layer (drop out). Further, the feature vector matrix of the customer a, the historical risk feature vector matrix of the customer a, the feature vector matrix of the rider slip, the historical risk fusion vector matrix of the rider slip, the emotional state fusion vector matrix of the rider slip, the feature vector matrix of the restaurant, the historical risk feature vector matrix of the restaurant, and the feature vector matrix of the order may be concatenated using a Concatenation network (Concatenation Connect) to obtain a concatenated vector matrix, and the activation function (e.g., an Elu activation function, referred to as Elu) refers to a nonlinear factor so that the input information input to the fully-connected network (Feed Forward) is a nonlinear function generated based on the information of the order of the customer a.
Specifically, the role of the fully connected network is to receive the non-linear function output risk values that customer A's order may produce. The fully-connected Network may be a feed-Forward Neural Network (FNN), which may employ a unidirectional multi-layer structure. Where each layer contains a number of neurons. In which each neuron can receive the signal of a neuron in the previous layer and generate an output to the next layer. The layer 0 is an input layer, the last layer is an output layer, and a plurality of hidden layers can be included between the input layer and the output layer.
Further, if the risk value 0.5 of the order of the small rider distribution customer a output by the full connection network is smaller than the risk probability threshold value 0.7, it can be considered that the order of the customer a does not have a public security problem, the prediction result is compared with the real result of the order of the customer a, it can be determined that the internal parameters of the risk prediction model of the small rider do not need to be adjusted, and the risk value of the order to be distributed of the small rider can be predicted by using the current risk prediction model of the small rider.
Therefore, 8 pieces of feature information of merchants, including feature information of multiple historical orders, and the like, can be fused and spliced through technical means such as a deep residual error network, a self-attention mechanism, a multi-head attention mechanism and the like, and then input into an activation function and a full connection network to output a risk value, so that risk behavior characteristics and real-time voice emotions of target distribution personnel, users and merchants in a service scene can be comprehensively analyzed, possible public security risks can be evaluated, and finally, the target distribution personnel, the users and the merchants with the risks can be isolated in a scheduling intervention mode, so that rights and interests of all parties and safety can be guaranteed.
In addition, referring to fig. 5c, in the embodiment of the present application, parameters of the activation function and parameters of the fully-connected network may also be adjusted, for example, a weight of a parameter corresponding to attitude evaluation in the activation function and a weight of a parameter corresponding to attitude evaluation in the fully-connected network are enhanced, so as to output a poor evaluation probability that the target delivery staff delivers the order to be distributed.
The foregoing description of specific embodiments has been presented for purposes of illustration and description. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
Fig. 6 is a schematic structural diagram of a security management apparatus provided in the present application. The device is used for a server and executes the security management method of any one of the above embodiments of the present specification. As shown in fig. 6, the security management apparatus may include:
the obtaining module 61 is used for obtaining information of the order to be distributed;
an obtaining module 62, configured to extract information of the order to be allocated, and obtain a risk value of the target distribution staff distributing the order to be allocated by using a risk prediction model of the target distribution staff; the risk prediction model of the target delivery personnel is obtained by training based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel; the information of each historical order comprises voice characteristic information;
the distribution module 63 is configured to send information of the to-be-distributed orders to other distribution staff when a risk value of the target distribution staff distributing the to-be-distributed orders is greater than a risk threshold; and the risk value of the other delivery personnel delivering the order to be distributed is less than or equal to the risk threshold value.
According to the embodiment of the application, risk prediction can be performed on the orders to be distributed delivered by the target delivery personnel through the pre-established risk prediction model so as to determine the risk value possibly generated when the target delivery personnel delivers the orders to be distributed. According to the risk prediction method and device, the risk prediction model can be trained according to the voice characteristic information in the information of the plurality of historical orders delivered by the target delivery personnel, so that the problems that the predicted risk value is not accurate and the like caused by the fact that the emotions of the target delivery personnel, the users and the merchants cannot be accurately identified through the text information in the related technology are correspondingly solved. In addition, when the risk value of the target delivery personnel is too large, the order to be delivered can be directly distributed to other delivery personnel with lower risk values, so that the possibly caused public security problem is avoided, and the good order of the society is effectively guaranteed.
In some embodiments, before the obtaining module 62, the apparatus further includes:
the first acquisition module is used for acquiring information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel; wherein the information of each historical order further comprises: the characteristic information of the target delivery personnel, the characteristic information of the historical order, the characteristic information of the user corresponding to the historical order and the characteristic information of the merchant corresponding to the historical order;
the risk prediction module is used for training a risk prediction model of the target delivery personnel based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel.
In some embodiments, before the obtaining module 62, the apparatus further includes:
a second obtaining module, configured to obtain information of historical orders of multiple known public security complaint results that are delivered by the target delivery staff, and information of historical orders of multiple known attitude evaluation results that are delivered by the target delivery staff;
and the second training module is used for training the risk prediction model of the target delivery personnel based on the information of the historical orders of a plurality of known public security complaint results distributed by the target delivery personnel and the information of the historical orders of a plurality of known attitude evaluation results distributed by the target delivery personnel.
Fig. 7 is a schematic structural diagram of an apparatus for building a risk prediction model provided in the present application. The device is used for a server and executes the method for establishing the risk prediction model according to any one of the above embodiments of the specification. As shown in fig. 7, the means for establishing a risk prediction model may include:
the obtaining module 71 is configured to obtain a preset risk prediction initial model, and obtain information of a plurality of historical orders of known public security complaint results, which are delivered by target delivery staff; the information of each historical order comprises voice characteristic information;
an obtaining module 72, configured to train the risk prediction initial model based on voice feature information in the information of the multiple historical orders of the multiple known public security complaint results that the target delivery staff completes delivery, so as to obtain a risk prediction model.
In some embodiments, the obtaining module 72 includes:
the first obtaining submodule is used for training the risk prediction initial model based on voice characteristic information in the information of the historical orders of a plurality of known public security complaint results, which are delivered by the target delivery personnel, so as to obtain the risk value of each historical order;
the first determining submodule is used for determining a loss value of the risk prediction initial model based on the risk value of each historical order and the public security complaint result of each historical order;
the updating submodule is used for adjusting parameters of the risk prediction initial model and updating the risk prediction initial model under the condition that the loss value is greater than a loss threshold value;
and a second obtaining module, configured to execute the speech feature information in the information of the multiple historical orders based on the multiple known public security complaint results that the target delivery staff completes delivery again, train the risk prediction initial model, and obtain a risk value of each historical order until the loss value is less than or equal to the loss threshold value, so as to obtain the risk prediction model.
In some embodiments, the information for each of the historical orders further comprises: the characteristic information of the target delivery personnel, the characteristic information of the historical order, the characteristic information of the user corresponding to the historical order and the characteristic information of the merchant corresponding to the historical order;
the first obtaining submodule is specifically configured to:
and training the risk prediction initial model based on the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the users corresponding to the historical orders, the characteristic information of the merchants corresponding to the historical orders and the voice characteristic information in the information of the historical orders of which the target delivery personnel finish delivering the known peace complaint results, so as to obtain the risk value of each historical order.
In some embodiments, before the first deriving submodule, the apparatus further comprises:
the first obtaining sub-module is used for obtaining information of historical orders of a plurality of known attitude evaluation results of the target distribution personnel completing distribution;
the first obtaining submodule is specifically configured to:
training the risk prediction initial model based on the characteristic information of the target delivery personnel in the information of the plurality of historical orders of known police complaint results which are delivered by the target delivery personnel and the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the user corresponding to the historical orders, the characteristic information of the merchant corresponding to the historical orders and the voice characteristic information in the information of the historical orders of the plurality of known attitude evaluation results which are delivered by the target delivery personnel, and obtaining the risk value of each historical order.
In some embodiments, the first deriving submodule includes:
the first determining unit is used for determining a characteristic vector matrix of the target delivery personnel and a historical risk characterization vector matrix of the target delivery personnel according to the characteristic information of the target delivery personnel in the historical order information;
the second determining unit is used for determining a characteristic vector matrix of the user and a historical risk characterization vector matrix of the user according to the characteristic information of the user corresponding to the historical order;
a third determining unit, configured to determine, according to feature information of a merchant corresponding to the historical order, a feature vector matrix of the merchant and a historical risk characterization vector matrix of the merchant;
a fourth determining unit, configured to determine a feature vector matrix of the historical order according to feature information of the historical order;
a fifth determining unit, configured to determine, according to the voice feature information corresponding to the historical order, an emotional state characterization vector matrix of the target delivery person;
a first obtaining unit, configured to train the risk prediction initial model based on a feature vector matrix of the target person, a feature vector matrix of the user, a feature vector matrix of the merchant, a feature vector matrix of the historical order, a historical risk characterization vector matrix of the target distributor, a historical risk characterization vector matrix of the merchant, a historical risk characterization vector matrix of the user, an emotional state characterization vector matrix of the target distributor, and an evaluation result and a peace complaint result corresponding to the historical order, which correspond to each historical order, to obtain a risk value of each historical order.
In some embodiments, the first determining unit includes:
the first determining subunit is configured to determine, according to the feature information of the target delivery person in the historical order information, a feature vector matrix of the target delivery person corresponding to the historical order;
the first obtaining subunit is configured to input feature information of the target distribution staff in the historical order information into a preset first depth residual error network to obtain a historical risk characterization vector matrix of the target distribution staff corresponding to the historical order;
the second determination unit includes:
the second determining subunit is configured to determine, according to the feature information of the user corresponding to the historical order, a feature vector matrix of the user corresponding to the historical order;
the second obtaining subunit is configured to input feature information of the user corresponding to the historical order into a preset second depth residual error network to obtain a historical risk characterization vector matrix of the user corresponding to the historical order;
the third determining unit includes:
the third determining subunit is configured to determine, according to the feature information of the merchant corresponding to the historical order, a feature vector matrix of the merchant corresponding to the historical order;
and the third obtaining subunit is configured to input the feature information of the merchant corresponding to the historical order into a preset third depth residual error network to obtain a historical risk characterization vector matrix of the merchant corresponding to the historical order.
In some embodiments, the voice feature information corresponding to the historical order includes: sound information and noise information of the target distribution personnel;
the fifth determination unit includes:
the conversion subunit is configured to convert the time domain signal corresponding to the voice feature information corresponding to the historical order into a frequency domain signal corresponding to the voice feature information corresponding to the historical order;
the cepstrum analysis subunit is configured to perform cepstrum analysis on the frequency domain signal corresponding to the voice feature information corresponding to the historical order to determine a vector matrix corresponding to the sound information of the target distribution staff in the voice feature information corresponding to the historical order;
and the fourth determining subunit is configured to determine, based on a vector matrix corresponding to the sound information of the target delivery person in the voice feature information corresponding to the historical order, an emotional state characterization vector matrix of the target delivery person.
In some embodiments, the fourth determining subunit is specifically configured to:
performing forward time sequence division on a vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order to obtain a first time sequence voice vector matrix; performing reverse time sequence division on a vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order to obtain a second time sequence voice vector matrix; weighting the first time sequence voice vector matrix and the second time sequence voice vector matrix to obtain a voice frequency strong correlation vector matrix; and determining an emotional state characterization vector matrix of the target distribution personnel based on the voice frequency strong correlation vector matrix and a vector threshold value.
In some embodiments, the first obtaining unit includes:
a fifth obtaining subunit, configured to perform inner product processing on the historical risk characterization vector matrix of the target distribution staff and the emotional state characterization vector matrix of the target distribution staff, so as to obtain a weight corresponding to the historical risk characterization vector matrix of the target distribution staff;
a fifth determining subunit, configured to determine a historical risk fusion vector of the target delivery staff based on a weight corresponding to the historical risk characterization vector matrix of the target delivery staff;
a sixth obtaining subunit, configured to perform inner product processing on the emotional state characterization vector matrix of the target distribution staff and the historical risk characterization vector matrix of the target distribution staff, so as to obtain a weight corresponding to the emotional state characterization vector matrix of the target distribution staff;
a sixth determining subunit, configured to determine an emotional state fusion vector of the target distribution staff based on a weight corresponding to the emotional state characterization vector matrix of the target distribution staff;
and a seventh obtaining subunit, configured to train the risk prediction initial model based on the eigenvector matrix of the target distributor, the eigenvector matrix of the user, the eigenvector matrix of the merchant, the eigenvector matrix of the historical order, the historical risk fusion vector of the target distributor, the historical risk characterization vector matrix of the merchant, the historical risk characterization vector matrix of the user, the emotional state fusion vector of the target distributor, and the evaluation result and the peace complaint result corresponding to the historical order, so as to obtain a risk value of each historical order.
In some embodiments, the seventh derivation subunit is specifically configured to:
training the risk prediction initial model based on a feature vector matrix of the target delivery personnel, a feature vector matrix of the user, a feature vector matrix of the merchant, a feature vector matrix of the historical order, a historical risk fusion vector of the target delivery personnel, a historical risk characterization vector matrix of the merchant, a historical risk characterization vector matrix of the user and an emotional state fusion vector of the target delivery personnel, which correspond to each historical order of the plurality of historical orders of the target delivery personnel, to obtain a risk value of each historical order;
the first determining submodule is specifically configured to:
and determining a loss value of the risk prediction initial model based on the risk value of each historical order, the public security complaint result of each historical order, the eigenvector matrix of the target delivery personnel, the historical risk characterization vector matrix of the target delivery personnel, the eigenvector matrix of the user and the historical risk characterization vector matrix of the user.
In some embodiments, the first determining submodule is specifically configured to:
determining a classification loss value of the risk prediction initial model based on the risk value of each historical order and the public security complaint result of each historical order; determining a characteristic information loss value of the risk prediction initial model based on the characteristic vector matrix of the target delivery personnel, the historical risk characterization vector matrix of the target delivery personnel, the characteristic vector matrix of the user and the historical risk characterization vector matrix of the user; determining a loss value of the initial risk prediction model based on the classification loss value of the initial risk prediction model and the characteristic information loss value of the initial risk prediction model.
In some embodiments, the first determining submodule is specifically configured to:
determining a distance between the risk value of each historical order and the result of the public complaint; and weighting the distances between the risk values of the plurality of historical orders and the corresponding public security complaint results to obtain the classification loss value of the risk prediction initial model.
In some embodiments, the first determining submodule is specifically configured to: determining a loss value of the target delivery personnel based on a characteristic vector matrix of the target delivery personnel corresponding to any one of the plurality of historical orders and a historical risk characterization vector matrix of the target delivery personnel; determining a loss value of the user based on a feature vector matrix of the user corresponding to each order in a plurality of historical orders and a historical risk characterization vector matrix of the user; and determining a characteristic information loss value of the risk prediction initial model based on the loss value of the target distribution personnel and the loss value of the user.
It should be noted that, when the apparatus for building a risk prediction model provided in the foregoing embodiment executes the method for building a risk prediction model, only the division of the functional modules is taken as an example, and in practical applications, the function distribution may be completed by different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules, so as to complete all or part of the functions described above. In addition, the device for establishing the risk prediction model and the method for establishing the risk prediction model provided by the embodiments belong to the same concept, and the embodiment of the method for establishing the risk prediction model embodies the implementation process in detail and is not repeated herein.
The above application serial numbers are merely for description and do not represent the merits of the embodiments.
Please refer to fig. 8, which provides a schematic structural diagram of an electronic device according to the present application. As shown in fig. 8, the electronic device 80 may include: at least one processor 81, at least one network interface 84, a user interface 83, a memory 85, at least one communication bus 82.
Wherein a communication bus 82 is used to enable the connection communication between these components.
The user interface 83 may include a Display screen (Display) and a Camera (Camera), and the optional user interface 83 may also include a standard wired interface and a wireless interface.
The network interface 84 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface), among others.
Processor 81 may include one or more processing cores, among others. The processor 81 connects various components throughout the electronic device 80 using various interfaces and lines to perform various functions of the electronic device 80 and process data by executing or executing instructions, programs, code sets, or instruction sets stored in the memory 85 and invoking data stored in the memory 85. Alternatively, the processor 81 may be implemented in at least one hardware form of Digital Signal Processing (DSP), field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). Processor 81 may integrate one or a combination of Central Processing Unit (CPU), graphics Processing Unit (GPU), modem, and the like. Wherein, the CPU mainly processes an operating system, a user interface, an application program and the like; the GPU is used for rendering and drawing the content required to be displayed by the display screen; the modem is used to handle wireless communications. It is understood that the modem may not be integrated into the processor 81, but may be implemented by a single chip.
The Memory 85 may include a Random Access Memory (RAM) or a Read-Only Memory (Read-Only Memory). Optionally, the memory 85 includes a non-transitory computer-readable medium. The memory 85 may be used to store instructions, programs, code, sets of codes, or sets of instructions. The memory 85 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the above-described method embodiments, and the like; the storage data area may store data and the like referred to in the above respective method embodiments. The memory 85 may alternatively be at least one memory device located remotely from the processor 81. As shown in fig. 8, the memory 85, which is a kind of computer storage medium, may include therein an operating system, a network communication module, a user interface module, and a security management application program.
In the electronic device 80 shown in fig. 8, the user interface 83 is mainly used as an interface for providing input for a user, and acquiring data input by the user; and processor 81 may be configured to invoke the security management application stored in memory 85 and perform the following operations:
acquiring information of an order to be distributed;
extracting the information of the order to be distributed, and obtaining a risk value of the order to be distributed by the target distribution personnel by using a risk prediction model of the target distribution personnel; the risk prediction model of the target delivery personnel is obtained by training based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel; the information of each historical order comprises voice characteristic information corresponding to the historical orders;
under the condition that the risk value of the target delivery personnel delivering the order to be distributed is larger than a risk threshold value, sending the information of the order to be distributed to other delivery personnel; and the risk value of the other delivery personnel delivering the order to be distributed is less than or equal to the risk threshold value. In some embodiments, the processor 81 further performs, before performing extracting the information of the to-be-distributed order and obtaining a risk value of the to-be-distributed order distributed by the target distribution personnel by using a risk prediction model of the target distribution personnel, a step of:
obtaining information of historical orders of a plurality of known public security complaint results of which the target delivery personnel finish delivery; wherein the information of each historical order further comprises: the characteristic information of the target delivery personnel, the characteristic information of the historical order, the characteristic information of the user corresponding to the historical order and the characteristic information of the merchant corresponding to the historical order;
training a risk prediction model of the target delivery personnel based on information of historical orders of a plurality of known peace complaint results that the target delivery personnel completed delivery.
In some embodiments, before performing the extracting of the information about the to-be-distributed order and obtaining the risk value of the target delivery person for delivering the to-be-distributed order by using the risk prediction model of the target delivery person, the processor 81 further performs:
acquiring information of a plurality of historical orders of known public security complaint results distributed by the target distribution personnel and information of a plurality of historical orders of known attitude evaluation results distributed by the target distribution personnel;
and training a risk prediction model of the target delivery personnel based on the information of the historical orders of the plurality of known public security complaint results distributed by the target delivery personnel and the information of the historical orders of the plurality of known attitude evaluation results distributed by the target delivery personnel.
Please refer to fig. 9, which provides a schematic structural diagram of an electronic device according to the present application. As shown in fig. 9, the electronic device 90 may include: at least one processor 91, at least one network interface 94, a user interface 93, a memory 95, at least one communication bus 92.
Wherein a communication bus 92 is used to enable the connection communication between these components.
The user interface 93 may include a Display screen (Display) and a Camera (Camera), and the optional user interface 93 may also include a standard wired interface and a wireless interface.
The network interface 94 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface), among others.
Processor 91 may include one or more processing cores, among others. The processor 91 connects various parts throughout the electronic device 90 using various interfaces and lines to perform various functions of the electronic device 90 and process data by executing or executing instructions, programs, code sets, or instruction sets stored in the memory 95 and invoking data stored in the memory 95. Alternatively, the processor 91 may be implemented in at least one hardware form of Digital Signal Processing (DSP), field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor 91 may integrate one or more of a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), a modem, and the like. Wherein, the CPU mainly processes an operating system, a user interface, an application program and the like; the GPU is used for rendering and drawing the content required to be displayed by the display screen; the modem is used to handle wireless communications. It is understood that the above modem may not be integrated into the processor 91, but may be implemented by a single chip.
The Memory 95 may include a Random Access Memory (RAM) or a Read-Only Memory (Read-Only Memory). Optionally, the memory 95 includes a non-transitory computer-readable medium. The memory 95 may be used to store instructions, programs, code, sets of codes, or sets of instructions. The memory 95 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing the above-mentioned method embodiments, and the like; the storage data area may store data and the like referred to in the above respective method embodiments. The memory 95 may optionally be at least one memory device located remotely from the processor 91. As shown in fig. 9, memory 95, which is a type of computer storage medium, may include an operating system, a network communication module, a user interface module, and a build risk prediction application.
In the electronic device 90 shown in fig. 9, the user interface 93 is mainly used as an interface for providing input for a user, and acquiring data input by the user; and the processor 91 may be configured to invoke the risk prediction modeling application stored in the memory 95 and specifically perform the following operations:
acquiring a preset risk prediction initial model and acquiring information of historical orders of a plurality of known public security complaint results distributed by target distribution personnel; the information of each historical order comprises voice characteristic information corresponding to the historical order;
and training the risk prediction initial model based on the voice characteristic information in the information of the plurality of historical orders of the plurality of known public security complaint results distributed by the target distribution personnel to obtain a risk prediction model.
In some embodiments, the processor 91, when executing the voice feature information in the information of the plurality of historical orders based on the plurality of known public security complaint results that are delivered by the target delivery staff, training the risk prediction initial model, and obtaining a risk prediction model, specifically executes:
training the risk prediction initial model based on voice characteristic information in the information of the plurality of historical orders of known public security complaint results which are delivered by the target delivery personnel, so as to obtain a risk value of each historical order;
determining a loss value of the risk prediction initial model based on the risk value of each historical order and the peace complaint result of each historical order;
under the condition that the loss value is larger than a loss threshold value, adjusting parameters of the risk prediction initial model, and updating the risk prediction initial model;
and executing the voice characteristic information in the information of the plurality of historical orders based on the known peace complaint results of the target delivery personnel completing the delivery again, training the risk prediction initial model to obtain the risk value of each historical order, and obtaining the risk prediction model until the loss value is less than or equal to the loss threshold value.
In some embodiments, the information for each of the historical orders further comprises: the characteristic information of the target delivery personnel, the characteristic information of the historical order, the characteristic information of the user corresponding to the historical order and the characteristic information of the merchant corresponding to the historical order;
when the processor 91 executes the speech feature information in the information of the historical orders based on the multiple known public security complaint results that the target delivery staff completes delivery, trains the risk prediction initial model, and obtains the risk value of each historical order, specifically executes:
and training the risk prediction initial model based on the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the users corresponding to the historical orders, the characteristic information of the merchants corresponding to the historical orders and the voice characteristic information in the information of the historical orders of which the target delivery personnel finish delivering the known peace complaint results, so as to obtain the risk value of each historical order.
In some embodiments, before executing the feature information of the target delivery person, the feature information of the historical order, the feature information of the user corresponding to the historical order, the feature information of the merchant corresponding to the historical order, and the voice feature information in the information of the historical orders based on the plurality of known public security complaint results that the target delivery person completes delivery, and training the risk prediction initial model to obtain the risk value of each historical order, the processor 91 further executes:
obtaining information of historical orders of a plurality of known attitude evaluation results of the target distribution personnel completing distribution;
the method for training the risk prediction initial model based on the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the users corresponding to the historical orders, the characteristic information of the merchants corresponding to the historical orders and the voice characteristic information in the information of the plurality of historical orders of which the target delivery personnel finish delivery and have known public security complaint results comprises the following steps:
training the risk prediction initial model based on the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the users corresponding to the historical orders, the characteristic information of the merchants corresponding to the historical orders and the voice characteristic information in the information of the historical orders of the plurality of known attitude evaluation results of the target delivery personnel, wherein the information of the historical orders of the target delivery personnel is delivered, and the risk value of each historical order is obtained.
In some embodiments, the processor 91, in executing the feature information of the target delivery staff, the feature information of the historical order, the feature information of the user corresponding to the historical order, the feature information of the merchant corresponding to the historical order, and the voice feature information in the information of the historical order based on the plurality of known public security complaint results that the target delivery staff completes delivery, and the plurality of known attitude assessment results that the target delivery staff completes delivery, trains the risk prediction initial model to obtain the risk value of each historical order, specifically executes:
determining a characteristic vector matrix of the target delivery personnel and a historical risk characterization vector matrix of the target delivery personnel according to the characteristic information of the target delivery personnel in the historical order information;
determining a characteristic vector matrix of the user and a historical risk characterization vector matrix of the user according to the characteristic information of the user corresponding to the historical order;
determining a characteristic vector matrix of a merchant and a historical risk characterization vector matrix of the merchant according to the characteristic information of the merchant corresponding to the historical order;
determining a feature vector matrix of the historical order according to the feature information of the historical order;
determining an emotional state characterization vector matrix of the target delivery personnel according to the voice characteristic information corresponding to the historical order;
training the risk prediction initial model based on the feature vector matrix of the target person, the feature vector matrix of the user, the feature vector matrix of the merchant, the feature vector matrix of the historical order, the historical risk characterization vector matrix of the target delivery person, the historical risk characterization vector matrix of the merchant, the historical risk characterization vector matrix of the user, the emotional state characterization vector matrix of the target delivery person, and the evaluation result and the peace complaint result corresponding to the historical order corresponding to each historical order to obtain the risk value of each historical order.
In some embodiments, when the determining the feature vector matrix of the target delivery person and the historical risk characterization vector matrix of the target delivery person according to the feature information of the target delivery person in the historical order information is performed, the processor 91 specifically performs:
determining a characteristic vector matrix of the target delivery personnel corresponding to the historical order according to the characteristic information of the target delivery personnel in the historical order information;
inputting the characteristic information of the target delivery personnel in the historical order information into a preset first depth residual error network to obtain a historical risk characterization vector matrix of the target delivery personnel corresponding to the historical order;
the determining, by the processor 91, the feature vector matrix of the user and the historical risk characterization vector matrix of the user according to the feature information of the user corresponding to the historical order includes:
determining a feature vector matrix of a user corresponding to the historical order according to feature information of the user corresponding to the historical order;
inputting the characteristic information of the user corresponding to the historical order into a preset second depth residual error network to obtain a historical risk characterization vector matrix of the user corresponding to the historical order;
the determining, by the processor 91, the characteristic vector matrix of the merchant and the historical risk characterization vector matrix of the merchant according to the characteristic information of the merchant corresponding to the historical order includes:
determining a feature vector matrix of a merchant corresponding to the historical order according to feature information of the merchant corresponding to the historical order;
inputting the characteristic information of the merchant corresponding to the historical order into a preset third depth residual error network to obtain a historical risk characterization vector matrix of the merchant corresponding to the historical order.
In some embodiments, the voice feature information corresponding to the historical order includes: sound information and noise information of the target distribution personnel;
when the processor 91 executes the determination of the emotion state characterization vector matrix of the target delivery person according to the voice feature information corresponding to the historical order, specifically executes:
converting a time domain signal corresponding to the voice characteristic information corresponding to the historical order into a frequency domain signal corresponding to the voice characteristic information corresponding to the historical order;
performing cepstrum analysis on the frequency domain signal corresponding to the voice feature information corresponding to the historical order to determine a vector matrix corresponding to the voice information of the target delivery personnel in the voice feature information corresponding to the historical order;
and determining an emotional state characterization vector matrix of the target delivery personnel based on a vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order.
In some embodiments, when the processor 91 determines the emotional state characterization vector matrix of the target delivery person based on the vector matrix corresponding to the sound information of the target delivery person in the voice feature information corresponding to the historical order, specifically performs:
performing forward time sequence division on a vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order to obtain a first time sequence voice vector matrix;
performing reverse time sequence division on a vector matrix corresponding to the sound information of the target delivery personnel in the voice feature information corresponding to the historical order to obtain a second time sequence voice vector matrix;
weighting the first time sequence voice vector matrix and the second time sequence voice vector matrix to obtain a voice frequency strong correlation vector matrix;
and determining an emotional state characterization vector matrix of the target delivery personnel based on the voice frequency strong correlation vector matrix and a vector threshold value.
In some embodiments, the processor 91, when executing the feature vector matrix of the target person, the feature vector matrix of the user, the feature vector matrix of the merchant, the feature vector matrix of the historical order, the historical risk characterization vector matrix of the target delivery person, the historical risk characterization vector matrix of the merchant, the historical risk characterization vector matrix of the user, the emotional state characterization vector matrix of the target delivery person, and the evaluation result and the police complaint result corresponding to the historical order corresponding to each historical order, training the risk prediction initial model to obtain the risk value of each historical order, specifically executes:
performing inner product processing on the historical risk characterization vector matrix of the target distribution personnel and the emotional state characterization vector matrix of the target distribution personnel to obtain the weight corresponding to the historical risk characterization vector matrix of the target distribution personnel;
determining a historical risk fusion vector of the target delivery personnel based on the weight corresponding to the historical risk characterization vector matrix of the target delivery personnel;
performing inner product processing on the emotional state characterization vector matrix of the target distribution personnel and the historical risk characterization vector matrix of the target distribution personnel to obtain the weight corresponding to the emotional state characterization vector matrix of the target distribution personnel;
determining an emotional state fusion vector of the target distribution personnel based on the weight corresponding to the emotional state characterization vector matrix of the target distribution personnel;
training the risk prediction initial model based on the feature vector matrix of the target delivery personnel, the feature vector matrix of the user, the feature vector matrix of the merchant, the feature vector matrix of the historical order, the historical risk fusion vector of the target delivery personnel, the historical risk characterization vector matrix of the merchant, the historical risk characterization vector matrix of the user, the emotional state fusion vector of the target delivery personnel, and the evaluation result and the public security complaint result corresponding to the historical order to obtain the risk value of each historical order.
In some embodiments, when executing the training of the risk prediction initial model based on the feature vector matrix of the target distributor, the feature vector matrix of the user, the feature vector matrix of the merchant, the feature vector matrix of the historical order, the historical risk fusion vector of the target distributor, the historical risk characterization vector matrix of the merchant, the historical risk characterization vector matrix of the user, the emotional state fusion vector of the target distributor, and the evaluation result and the peace complaint result corresponding to the historical order, to obtain the risk value of each historical order, the processor 91 specifically executes:
training the risk prediction initial model based on a feature vector matrix of the target delivery personnel, a feature vector matrix of the user, a feature vector matrix of the merchant, a feature vector matrix of the historical order, a historical risk fusion vector of the target delivery personnel, a historical risk characterization vector matrix of the merchant, a historical risk characterization vector matrix of the user and an emotional state fusion vector of the target delivery personnel, which correspond to each historical order in the plurality of historical orders of the target delivery personnel, to obtain a risk value of each historical order;
in some embodiments, when the determining the loss value of the initial risk prediction model based on the risk value of each historical order and the result of the public security complaint of each historical order, the processor 91 specifically performs:
and determining a loss value of the risk prediction initial model based on the risk value of each historical order, the peace complaint result of each historical order, the eigenvector matrix of the target delivery personnel, the historical risk characterization vector matrix of the target delivery personnel, the eigenvector matrix of the user and the historical risk characterization vector matrix of the user.
In some embodiments, the processor 91 specifically performs, when executing the determining of the loss value of the risk prediction initial model based on the risk value of each historical order, the result of the public complaint of each historical order, the feature vector matrix of the target delivery person, the historical risk characterization vector matrix of the target delivery person, the feature vector matrix of the user, and the historical risk characterization vector matrix of the user:
determining a classification loss value of the risk prediction initial model based on the risk value of each historical order and the peace complaint result of each historical order;
determining a characteristic information loss value of the risk prediction initial model based on the characteristic vector matrix of the target delivery personnel, the historical risk characterization vector matrix of the target delivery personnel, the characteristic vector matrix of the user and the historical risk characterization vector matrix of the user;
determining a loss value of the initial risk prediction model based on the classification loss value of the initial risk prediction model and the characteristic information loss value of the initial risk prediction model.
In some embodiments, the processor 91, when executing the step of determining the classification loss value of the initial risk prediction model based on the risk value of each historical order and the result of the public security complaint of each historical order, specifically executes:
determining a distance between the risk value of each historical order and the result of the public complaint;
and weighting the distances between the risk values of the plurality of historical orders and the corresponding public security complaint results to obtain the classification loss value of the risk prediction initial model.
In some embodiments, the processor 91, when executing the determining of the characteristic information loss value of the risk prediction initial model based on the characteristic vector matrix of the target delivery person, the historical risk characterization vector matrix of the target delivery person, the characteristic vector matrix of the user, and the historical risk characterization vector matrix of the user, specifically executes:
determining a loss value of the target delivery personnel based on a characteristic vector matrix of the target delivery personnel corresponding to any one of the plurality of historical orders and a historical risk characterization vector matrix of the target delivery personnel;
determining a loss value of the user based on a feature vector matrix of the user corresponding to each order in a plurality of historical orders and a historical risk characterization vector matrix of the user;
and determining a characteristic information loss value of the risk prediction initial model based on the loss value of the target distribution personnel and the loss value of the user.
The present application also provides a computer-readable storage medium having stored therein instructions, which when executed on a computer or processor, cause the computer or processor to perform one or more of the steps in the embodiments of fig. 2a, 3a, 4a, and 5a described above. The respective constituent modules of the security management apparatus may be stored in the computer-readable storage medium if they are implemented in the form of software functional units and sold or used as independent products.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the processes or functions described in the present application are generated in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in or transmitted over a computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that incorporates one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., digital Versatile Disk (DVD)), or a semiconductor medium (e.g., solid State Disk (SSD)), among others.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by instructing relevant hardware by a computer program, and the program may be stored in a computer-readable storage medium, and when executed, may include the processes of the embodiments of the methods described above. And the aforementioned storage medium includes: various media capable of storing program codes, such as a Read Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, and an optical disk. The technical features in the present examples and embodiments may be arbitrarily combined without conflict.
The above-described embodiments are only preferred embodiments of the present application, and are not intended to limit the scope of the present application, and various modifications and improvements made to the technical solutions of the present application by those skilled in the art without departing from the design spirit of the present application should fall within the protection scope defined by the claims of the present application.

Claims (10)

1. A method for security management, the method comprising:
acquiring information of an order to be distributed;
extracting the information of the order to be distributed, and obtaining a risk value of the order to be distributed by the target distribution personnel by using a risk prediction model of the target distribution personnel; the risk prediction model of the target delivery personnel is obtained by training based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel; the information of each historical order comprises voice characteristic information corresponding to the historical order;
under the condition that the risk value of the target delivery personnel delivering the order to be distributed is larger than a risk threshold value, sending the information of the order to be distributed to other delivery personnel; and the risk value of the other delivery personnel delivering the order to be distributed is less than or equal to the risk threshold value.
2. The method according to claim 1, wherein before extracting the information of the order to be distributed and obtaining the risk value of the target delivery personnel for delivering the order to be distributed by using the risk prediction model of the target delivery personnel, the method further comprises:
obtaining information of historical orders of a plurality of known public security complaint results of which the target delivery personnel finish delivery; wherein the information of each historical order further comprises: the characteristic information of the target delivery personnel, the characteristic information of the historical order, the characteristic information of the user corresponding to the historical order and the characteristic information of the merchant corresponding to the historical order are obtained;
training a risk prediction model of the target delivery personnel based on information of historical orders of a plurality of known peace complaint results that the target delivery personnel completed delivery.
3. The method of claim 2, wherein before the extracting the information of the order to be distributed and obtaining the risk value of the order to be distributed by the target distribution personnel by using the risk prediction model of the target distribution personnel, the method further comprises:
acquiring information of a plurality of historical orders of known public security complaint results distributed by the target distribution personnel and information of a plurality of historical orders of known attitude evaluation results distributed by the target distribution personnel;
training a risk prediction model of the target delivery personnel based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel and information of historical orders of a plurality of known attitude evaluation results which are delivered by the target delivery personnel.
4. A method of building a risk prediction model, the method comprising:
acquiring a preset risk prediction initial model and acquiring information of historical orders of a plurality of known public security complaint results distributed by target distribution personnel; the information of each historical order comprises voice characteristic information corresponding to the historical orders;
and training the risk prediction initial model based on the voice characteristic information in the information of the plurality of historical orders of the plurality of known public security complaint results distributed by the target distribution personnel to obtain a risk prediction model.
5. The method according to claim 4, wherein the training of the risk prediction initial model based on the voice feature information in the information of the plurality of historical orders of the plurality of known public security complaint results delivered by the target delivery staff to obtain a risk prediction model comprises:
training the risk prediction initial model based on voice characteristic information in the information of the plurality of historical orders of known public security complaint results which are delivered by the target delivery personnel, so as to obtain a risk value of each historical order;
determining a loss value of the risk prediction initial model based on the risk value of each historical order and the peace complaint result of each historical order;
under the condition that the loss value is larger than a loss threshold value, adjusting parameters of the risk prediction initial model, and updating the risk prediction initial model;
and executing the voice characteristic information in the information of the plurality of historical orders based on the plurality of known peace complaint results of the target delivery personnel completing delivery again, training the risk prediction initial model to obtain the risk value of each historical order, and obtaining the risk prediction model until the loss value is less than or equal to the loss threshold value.
6. The method of claim 5, wherein the information for each of the historical orders further comprises: the characteristic information of the target delivery personnel, the characteristic information of the historical order, the characteristic information of the user corresponding to the historical order and the characteristic information of the merchant corresponding to the historical order;
training the risk prediction initial model based on voice characteristic information in the information of the historical orders of the plurality of known peace complaint results distributed by the target distribution personnel to obtain the risk value of each historical order, wherein the method comprises the following steps:
and training the risk prediction initial model based on the characteristic information of the target delivery personnel, the characteristic information of the historical orders, the characteristic information of the users corresponding to the historical orders, the characteristic information of the merchants corresponding to the historical orders and the voice characteristic information in the information of the historical orders of which the target delivery personnel finish delivering a plurality of known public security complaint results, and obtaining the risk value of each historical order.
7. A security management apparatus, characterized in that the apparatus comprises:
the acquisition module is used for acquiring information of the order to be distributed;
the obtaining module is used for extracting the information of the order to be distributed and obtaining a risk value of the order to be distributed by the target distribution personnel by utilizing a risk prediction model of the target distribution personnel; the risk prediction model of the target delivery personnel is obtained by training based on information of historical orders of a plurality of known public security complaint results which are delivered by the target delivery personnel; the information of each historical order comprises voice characteristic information;
the distribution module is used for sending the information of the order to be distributed to other distribution personnel under the condition that the risk value of the target distribution personnel for distributing the order to be distributed is larger than a risk threshold value; and the risk value of the other delivery personnel delivering the order to be distributed is less than or equal to the risk threshold value.
8. An apparatus for modeling risk prediction, the apparatus comprising:
the system comprises an acquisition module, a risk prediction module and a risk prediction module, wherein the acquisition module is used for acquiring a preset risk prediction initial model and acquiring information of historical orders of a plurality of known public security complaint results which are delivered by target delivery personnel; the information of each historical order comprises voice characteristic information;
and the obtaining module is used for training the risk prediction initial model based on the voice characteristic information in the information of a plurality of historical orders of a plurality of known peace complaint results distributed by the target distribution personnel to obtain a risk prediction model.
9. A computer storage medium, characterized in that it stores a plurality of instructions adapted to be loaded by a processor and to perform the method steps according to any of claims 1-3 or 4-6.
10. An electronic device, comprising: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to perform the method steps according to any of claims 1-3 or 4-6.
CN202211078104.5A 2022-09-05 2022-09-05 Security management method, device, electronic equipment and computer storage medium Active CN115423323B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211078104.5A CN115423323B (en) 2022-09-05 2022-09-05 Security management method, device, electronic equipment and computer storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211078104.5A CN115423323B (en) 2022-09-05 2022-09-05 Security management method, device, electronic equipment and computer storage medium

Publications (2)

Publication Number Publication Date
CN115423323A true CN115423323A (en) 2022-12-02
CN115423323B CN115423323B (en) 2023-05-23

Family

ID=84203142

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211078104.5A Active CN115423323B (en) 2022-09-05 2022-09-05 Security management method, device, electronic equipment and computer storage medium

Country Status (1)

Country Link
CN (1) CN115423323B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116109116A (en) * 2023-04-13 2023-05-12 中建安装集团有限公司 Cloud computing-based comprehensive application program supervision control system and method

Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2287927A1 (en) * 1997-05-01 1998-11-05 William E. Datig Universal epistemological machine (a.k.a. android)
US20160005049A1 (en) * 2014-07-02 2016-01-07 Verizon Patent And Licensing Inc. Predicting a likelihood of customer service interactions
CN110097381A (en) * 2019-05-01 2019-08-06 青岛民航凯亚系统集成有限公司 A kind of complaint automatic identification early warning system and method applied to airport service
CN110119847A (en) * 2019-05-14 2019-08-13 拉扎斯网络科技（上海）有限公司 A kind of prediction technique, device, storage medium and electronic equipment dispensing duration
CN110489951A (en) * 2019-07-08 2019-11-22 招联消费金融有限公司 Method, apparatus, computer equipment and the storage medium of risk identification
CN111612200A (en) * 2019-02-25 2020-09-01 北京嘀嘀无限科技发展有限公司 Order security prediction method, device, server and storage medium
CN111625413A (en) * 2020-04-23 2020-09-04 平安科技（深圳）有限公司 Index abnormality analysis method, index abnormality analysis device and storage medium
CN112581297A (en) * 2020-12-18 2021-03-30 中国平安人寿保险股份有限公司 Information pushing method and device based on artificial intelligence and computer equipment
CN113421022A (en) * 2021-07-14 2021-09-21 武汉字节空间科技有限公司 Order business processing method and equipment based on big data and computer storage medium
CN113889099A (en) * 2021-10-08 2022-01-04 上海金仕达软件科技有限公司 Voice recognition method and system
US11276127B1 (en) * 2021-03-04 2022-03-15 Timothy Dirk Stevens Recommending matches using machine learning
CN114596111A (en) * 2022-03-03 2022-06-07 浙江吉利控股集团有限公司 Risk identification model generation method, device, equipment and storage medium
CN114971446A (en) * 2021-02-22 2022-08-30 顺丰科技有限公司 Method and device for constructing loss risk prediction model and computer equipment
CN114997945A (en) * 2022-05-18 2022-09-02 拉扎斯网络科技（上海）有限公司 Order state display method and device, electronic equipment and computer storage medium

Patent Citations (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2287927A1 (en) * 1997-05-01 1998-11-05 William E. Datig Universal epistemological machine (a.k.a. android)
US20160005049A1 (en) * 2014-07-02 2016-01-07 Verizon Patent And Licensing Inc. Predicting a likelihood of customer service interactions
CN111612200A (en) * 2019-02-25 2020-09-01 北京嘀嘀无限科技发展有限公司 Order security prediction method, device, server and storage medium
CN110097381A (en) * 2019-05-01 2019-08-06 青岛民航凯亚系统集成有限公司 A kind of complaint automatic identification early warning system and method applied to airport service
CN110119847A (en) * 2019-05-14 2019-08-13 拉扎斯网络科技（上海）有限公司 A kind of prediction technique, device, storage medium and electronic equipment dispensing duration
CN110489951A (en) * 2019-07-08 2019-11-22 招联消费金融有限公司 Method, apparatus, computer equipment and the storage medium of risk identification
CN111625413A (en) * 2020-04-23 2020-09-04 平安科技（深圳）有限公司 Index abnormality analysis method, index abnormality analysis device and storage medium
CN112581297A (en) * 2020-12-18 2021-03-30 中国平安人寿保险股份有限公司 Information pushing method and device based on artificial intelligence and computer equipment
CN114971446A (en) * 2021-02-22 2022-08-30 顺丰科技有限公司 Method and device for constructing loss risk prediction model and computer equipment
US11276127B1 (en) * 2021-03-04 2022-03-15 Timothy Dirk Stevens Recommending matches using machine learning
CN113421022A (en) * 2021-07-14 2021-09-21 武汉字节空间科技有限公司 Order business processing method and equipment based on big data and computer storage medium
CN113889099A (en) * 2021-10-08 2022-01-04 上海金仕达软件科技有限公司 Voice recognition method and system
CN114596111A (en) * 2022-03-03 2022-06-07 浙江吉利控股集团有限公司 Risk identification model generation method, device, equipment and storage medium
CN114997945A (en) * 2022-05-18 2022-09-02 拉扎斯网络科技（上海）有限公司 Order state display method and device, electronic equipment and computer storage medium

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
GRAVES ALEX 等: "Towards end-to-end speech recognition with recurrent neural networks" *
张镇鹏: "基于网络口碑的B2C企业服务质量问题分析及对策研究" *
李奇龙: "基于对话的多模态融合情绪识别方法研究" *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116109116A (en) * 2023-04-13 2023-05-12 中建安装集团有限公司 Cloud computing-based comprehensive application program supervision control system and method

Also Published As

Publication number Publication date
CN115423323B (en) 2023-05-23

Similar Documents

Publication Publication Date Title
US10262195B2 (en) Predictive and responsive video analytics system and methods
CN107071193A (en) The method and apparatus of interactive answering system accessing user
US10652655B1 (en) Cognitive volume and speech frequency levels adjustment
CN110309299B (en) Communication anti-fraud method, device, computer readable medium and electronic equipment
US11153242B2 (en) Automated prevention of sending objectionable content through electronic communications
CN109960761A (en) Information recommendation method, device, equipment and computer readable storage medium
JP2015534149A (en) A machine learning model for hierarchically based arrays
CN109285011A (en) A kind of smart client return visit method, device and equipment
US20190019507A1 (en) Dynamic personalized multi-turn interaction of cognitive models
CN115423323A (en) Security management method and device, electronic equipment and computer storage medium
US10678821B2 (en) Evaluating theses using tree structures
CN112581226A (en) Product recommendation method and system
CN107465797A (en) Terminal device incoming information display methods and device
TWM578858U (en) Cross-channel artificial intelligence dialogue platform
US20220253787A1 (en) Assessing project quality using confidence analysis of project communications
US20200311648A1 (en) Rating system for law enforcement
CN113742593A (en) Method and device for pushing information
US11669759B2 (en) Entity resource recommendation system based on interaction vectorization
CN113112326A (en) User identification method, method for displaying data to user and related device
US11967311B1 (en) Recipient customized generative voice modeling
TWI739067B (en) Cross-channel artificial intelligence dialogue platform and operation method thereof
US20230352003A1 (en) Systems and methods to improve trust in conversations
AU2018101340A4 (en) Robotic Property Conveyancing Service
TWI769653B (en) Cross-channel artificial intelligence dialogue platform capable of reassembling voice segment and its operation method
TWI767498B (en) Cross-channel artificial intelligence dialogue platform integrating machine learning and operation method thereof

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant