An intelligent risk control auditing strategy generation method based on big data analysis

Abstract

The application discloses an intelligent risk control auditing strategy generation method based on big data analysis, relates to the technical field of information risk control auditing, and is used for solving the problem of the descending matching degree of night recommendation information recommendation strategies, calling an information database to screen night recommendation information, setting monitoring time to count user traffic and click volume, generating access adhesion degree, collecting display time length to set recommendation weight, calculating adhesion benefit index and updating initial recommendation sequence, putting the updated sequence into a matching calibration mechanism, counting information jump-out rate and revisit rate, calling a default period matching threshold value and calculating a calibration proportion, correcting generated update threshold value, comparing the adhesion benefit index with the update threshold value to judge whether to perform next withdrawal processing, realizing dynamic recommendation optimization based on user behavior feedback, and improving the matching degree of night recommendation information recommendation strategies and user experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of information risk control auditing, and more particularly to an intelligent risk control auditing strategy generation method based on big data analysis. BACKGROUND

[0002] With the rapid development of Internet information services and intelligent recommendation systems, personalized recommendation technology based on user behavior characteristics is widely used in e-commerce platforms, information aggregation platforms and financial risk control scenarios. Existing recommendation strategies rely on static recommendation weights or fixed recommendation rules. In the case that the recommendation information delivery period and user behavior characteristics do not match well, the phenomenon of disconnection between recommended information and user demand is likely to occur.

[0003] The prior art has the following disadvantages:

[0004] Currently, there are differences in user access habits, click behavior and dwell time between night and day. The existing recommendation optimization method relies on static weights and single rules, lacks dynamic modeling and multi-dimensional analysis of night user behavior characteristics, and cannot timely adjust the night recommendation information sequence and the withdrawal of part of the night recommendation information, resulting in a decline in user experience and a decline in the matching degree of night recommendation information recommendation strategy. Therefore, there is a lack of abnormal identification and risk control of night recommendation information, and therefore, an intelligent risk control auditing strategy generation method based on big data analysis is proposed.

[0005] The above information disclosed in the background section is only intended to strengthen the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present application provide an intelligent risk control auditing strategy generation method based on big data analysis, which uses user behavior characteristic modeling, access adhesion degree calculation, adhesion benefit index evaluation and threshold dynamic calibration mechanism to solve the problems raised in the above background technology.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical scheme, an intelligent risk control auditing strategy generation method based on big data analysis, comprising the following steps:

[0008] Step S1: call the push data of each recommendation information in the information database, filter out the night recommendation information according to the push data, set the monitoring time, and count the user traffic and the number of clicks of the night recommendation information within the monitoring time;

[0009] Step S2: generate the access stickiness of the night recommendation information by integrating the user traffic and the click volume of the night recommendation information, collect the display duration of the night recommendation information and set the recommendation weight, calculate the stickiness benefit index of the night recommendation information in combination with the access stickiness, and update the initial recommendation sequence of each night recommendation information based on the stickiness benefit index;

[0010] Step S3: after updating the initial recommendation sequence, enter the matching calibration mechanism, count the information jump-out rate and information revisit rate, call the default time period matching threshold of the night recommendation information, and calculate the calibration proportion of the default time period matching threshold in combination with the information jump-out rate and information revisit rate;

[0011] Step S4: correct the default time period matching threshold by using the calibration proportion and generate the time period matching threshold, and compare the stickiness benefit index of each night recommendation information with the time period matching threshold to determine whether to perform the next withdrawal process.

[0012] In one preferred embodiment, in step S1, the push data of each recommendation information in the information database is called, the recommendation information with the delivery time period being the night period is obtained by searching and screening the push data, and is defined as the night recommendation information;

[0013] The user access quantity of the night recommendation information in the preset monitoring time is obtained as the user traffic through the user access log;

[0014] The click volume of the night recommendation information is obtained by counting the click quantity of the night recommendation information by the user in the preset monitoring time through the click behavior log.

[0015] In one preferred embodiment, in step S2, the ratio of the click volume of the night recommendation information to the user traffic is taken as the access stickiness of the night recommendation information;

[0016] The display duration of the night recommendation information is collected through the display behavior log, and the display duration refers to the cumulative time of the actual display of the night recommendation information on the terminal;

[0017] The recommendation weight is obtained by normalizing the display duration.

[0018] In one preferred embodiment, in step S2, the access stickiness and the display duration are weighted and summed to obtain the stickiness benefit index;

[0019] The initial recommendation sequence of the night recommendation information is called, and the initial recommendation sequence refers to the randomly generated arrangement sequence of the night recommendation information without considering the stickiness benefit index;

[0020] The night recommendation information in the initial recommendation sequence is reordered according to the stickiness benefit index from large to small to form an optimized recommendation sequence.

[0021] In a preferred embodiment, in step S3, the initial recommendation sequence is updated and then enters a matching calibration mechanism;

[0022] A preset statistical period is set, and the total number of visits is obtained by counting the click volume of the user accessing all the night recommendation information in the preset statistical period, and the stay duration of the user on the recommendation page corresponding to the night recommendation information is recorded;

[0023] If the stay duration is less than the preset stay duration threshold, the access behavior of the user accessing the night recommendation information is marked as a bounce event;

[0024] Otherwise, the access behavior is not marked;

[0025] The total number of bounces is obtained by counting the number of bounce events, and the information bounce rate is the ratio of the total number of bounces to the total number of visits.

[0026] In a preferred embodiment, in step S3, for each night recommendation information, the access user list is called through the access log database in a preset statistical period, and the number of visits of the user to the night recommendation information is counted;

[0027] If the number of visits of the user to the night recommendation information is greater than 1, the user is marked as a repeat click user;

[0028] Otherwise, the user is not marked.

[0029] In a preferred embodiment, in step S3, the number of repeat click users for each night recommendation information is accumulated to obtain the number of repeat click users;

[0030] The ratio of the number of repeat click users to the total number of users in the access user list of the night recommendation information is the repeat click user proportion;

[0031] The repeat click user proportion of each night recommendation information is added, and the ratio of the total number of night recommendation information to the total number of night recommendation information is the information revisit rate.

[0032] In a preferred embodiment, in step S3, the default period matching threshold of the night recommendation information is called through the recommendation strategy configuration library;

[0033] The calibration proportion of the default period matching threshold is calculated by combining the information bounce rate and the information revisit rate: , wherein, is the information revisit rate, is the information bounce rate, and is a preset adjustment factor, is the calibration proportion, is a preset calibration coefficient.

[0034] In a preferred embodiment, in step S4, the product of the proofreading ratio and the default time period matching threshold value is taken as the time period matching threshold value;

[0035] The time period matching threshold value is compared with the sticking benefit index of the night recommendation information to determine whether to perform the withdrawal processing:

[0036] If the sticking benefit index is lower than the time period matching threshold value, the night recommendation information is subjected to the withdrawal processing;

[0037] Otherwise, the night recommendation information is retained for display.

[0038] Technical effects and advantages of the present application:

[0039] The present application filters out the night recommendation information by calling the push data of each recommendation information in the information database, sets the monitoring time, counts the user flow and the click amount of the night recommendation information, generates the access sticking degree by comprehensively considering the user flow and the click amount, sets the recommendation weight by collecting the display time length, calculates the sticking benefit index, and updates the initial recommendation sequence according to the sticking benefit index, enters the updated recommendation sequence into the matching calibration mechanism, counts the information jump-out rate and the information revisit rate, calls the default time period matching threshold value, calculates the proofreading ratio by comprehensively considering the information jump-out rate and the information revisit rate, corrects the default time period matching threshold value according to the proofreading ratio to generate the updated threshold value, compares the sticking benefit index of each night recommendation information with the updated threshold value to determine whether to perform the withdrawal processing, and realizes the dynamic recommendation optimization based on the user behavior feedback, improves the matching degree of the night recommendation information recommendation strategy and the user experience. BRIEF DESCRIPTION OF DRAWINGS

[0040] Fig. 1 The implementation flowchart of the intelligent risk control audit strategy generation method based on big data analysis.

[0041] Fig. 2 The step schematic diagram of the intelligent risk control audit strategy generation method based on big data analysis. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0043] The present application filters out night recommendation information by calling the push data of each recommendation information in the information database, sets the monitoring time, counts the user flow and the click volume of the night recommendation information, generates the access adhesion degree by comprehensively considering the user flow and the click volume, sets the recommendation weight by collecting the display duration, calculates the adhesion benefit index, updates the initial recommendation sequence according to the adhesion benefit index, puts the updated recommendation sequence into the matching calibration mechanism, counts the information jump-out rate and the information revisit rate, calls the default period matching threshold, calculates the calibration proportion by comprehensively considering the information jump-out rate and the information revisit rate, corrects the default period matching threshold according to the calibration proportion to generate the updated threshold, compares the adhesion benefit index of each night recommendation information with the updated threshold to judge whether to perform the next withdrawal processing, and realizes the dynamic recommendation optimization based on the user behavior feedback.

[0044] Embodiment 1

[0045] As Figs. 1-2 shown, an intelligent risk control audit strategy generation method based on big data analysis includes the following steps:

[0046] Step S1: calling the push data of each recommendation information in the information database, filtering out the night recommendation information according to the push data, setting the monitoring time, counting the user flow and the click volume of the night recommendation information in the monitoring time;

[0047] Step S2: generating the access adhesion degree of the night recommendation information by comprehensively considering the user flow and the click volume of the night recommendation information, setting the recommendation weight by collecting the display duration of the night recommendation information, calculating the adhesion benefit index of the night recommendation information in combination with the access adhesion degree, and updating the initial recommendation sequence of each night recommendation information based on the adhesion benefit index;

[0048] Step S3: after updating the initial recommendation sequence, entering the matching calibration mechanism, counting the information jump-out rate and the information revisit rate, calling the default period matching threshold of the night recommendation information, and calculating the calibration proportion of the default period matching threshold by comprehensively considering the information jump-out rate and the information revisit rate;

[0049] Step S4: correcting the default period matching threshold by using the calibration proportion and generating the period matching threshold, comparing the adhesion benefit index of each night recommendation information with the period matching threshold to judge whether to perform the next withdrawal processing.

[0050] The specific implementation is as follows:

[0051] In step S1, the push data of each recommendation information in the information database is called, the push data includes the label information associated with the recommendation information, which is used to represent the delivery period, target user group and push behavior characteristics of the recommendation information, the recommendation information with the delivery period of night period is obtained by searching and filtering the push data, and is defined as night recommendation information.

[0052] It should be noted that the night period is a time interval according to the time interval division principle to limit the range of recommended information in the time dimension, which is used to limit the display time of the recommended information in the preset night interval, for example, 22:00 to 06:00 the next day is the night period, and the starting time and the ending time are dynamically adjusted according to the user behavior characteristics in the specific implementation.

[0053] After obtaining the night recommended information, a monitoring time is set, and the user traffic and the night recommended information click volume are counted in the monitoring time;

[0054] The user access quantity of the night recommended information in the monitoring time is obtained as the user traffic through the user access log, the user access quantity is the number of users who generate access behavior, and the judgment condition of the access behavior is that the user completes page loading, interface request or data interaction operation;

[0055] The night recommended information click volume is obtained by counting the number of clicks of the night recommended information by the user in the monitoring time through the click behavior log, the click behavior log records the interaction behavior of the user to the recommended information in the access process, including user identification, recommended information identification, click action type and timestamp, through the matching of the timestamp and the recommended information identification, the click record of the night recommended information in the monitoring time interval is filtered out, the multiple click behaviors of the same user on the same recommended information are de-duplicated, and the multiple repeated clicks in the same session are determined as one click behavior.

[0056] The recommended information with the delivery period being night is filtered out by calling the push data of the information database and retrieving the tag information, the period management of the recommended information is realized, the user traffic is counted based on the user access log, accurate input is provided for subsequent access adhesion degree calculation, the credibility of the adhesion benefit index is improved, and the rationality and matching degree of the recommended sequence update are enhanced.

[0057] It should be noted that the monitoring time refers to a time interval set for counting the user traffic and the night recommended information click volume after obtaining the night recommended information, which is set based on the preset rule or adaptively determined through the distribution characteristics of the historical user behavior data, for example, a certain period in the night period is selected as the monitoring time; the user access log refers to a behavior data set automatically recorded in the user access process, which is used to reflect the access behavior of the user in the monitoring time; the click behavior log refers to a behavior data set automatically generated when the user interacts with the recommended information, which is used to record the click response behavior of the user to the recommended information.

[0058] In step S2, based on the user traffic and the night recommended information click volume, the access adhesion degree of the night recommended information is calculated, which is used to quantify the attention degree and interaction adhesion characteristics of the user to the night recommended information in the monitoring time interval;

[0059] The higher the user traffic and the lower the click volume of the nighttime recommendation information within the monitoring time, the less attractive the nighttime recommendation information is to the user; conversely, the lower the user traffic and the higher the click volume of the nighttime recommendation information, the stronger the interactivity of the nighttime recommendation information.

[0060] The access adhesion degree is defined as the ratio of the click volume of the nighttime recommendation information to the user traffic, reflecting the relative attractiveness and interactive adhesion of the nighttime recommendation information in the user access behavior. The greater the access adhesion degree, the higher the adhesion degree of the nighttime recommendation information to the user, and vice versa.

[0061] The display duration of the nighttime recommendation information is collected through the display behavior log, which refers to the cumulative time of the nighttime recommendation information actually displayed on the terminal.

[0062] It should be noted that the display behavior log refers to the automatically generated and stored behavior data set when the recommendation information is pushed to the terminal and the actual display process is completed, which is used to record the exposure of the recommendation information on the user side.

[0063] Based on the display duration, the recommendation weight of the nighttime recommendation information is set, which is obtained by normalizing the display duration. The calculation formula is as follows:

[0064] ;

[0065] wherein, is the recommendation weight of the nighttime recommendation information, is the display duration of the nighttime recommendation information, is the maximum value of the display duration, and is the index value of the nighttime recommendation information, is the total number of nighttime recommendation information.

[0066] The greater the value of the recommendation weight, the longer the display exposure duration of the recommendation information on the user terminal; conversely, the smaller the value of the recommendation weight, the shorter the display exposure duration of the recommendation information on the user terminal.

[0067] After obtaining the access adhesion degree and the recommendation weight, the adhesion benefit index of the nighttime recommendation information is calculated, which is used to comprehensively evaluate the user response characteristics and display effect of the recommendation information. The calculation formula of the adhesion benefit index is as follows:

[0068] ;

[0069] wherein, is the adhesion benefit index of the nighttime recommendation information, is the access adhesion degree of the nighttime recommendation information, a recommendation weight of the nighttime recommendation information, and a weight coefficient of the access adhesion degree and the recommendation weight, satisfying .

[0070] The larger the value of the adhesion benefit index is, the higher the user attraction and display effectiveness of the nighttime recommendation information is; otherwise, the lower the effect of the nighttime recommendation information is.

[0071] Based on the adhesion benefit index, the initial recommendation sequence of each nighttime recommendation information is updated, and the initial recommendation sequence refers to the randomly generated nighttime recommendation information arrangement sequence before considering the adhesion benefit index.

[0072] The updating process reorders the nighttime recommendation information from large to small according to the adhesion benefit index, and the nighttime recommendation information with a higher adhesion benefit index is arranged in the front, and the nighttime recommendation information with a lower adhesion benefit index is arranged in the back.

[0073] The optimized recommendation sequence is formed through the updating process, which ensures that the display order of the nighttime recommendation information is more effectively matched with the user behavior, improves the overall recommendation effect, and provides a quantitative basis for subsequent risk control audit strategy generation.

[0074] In step S3, after updating the initial recommendation sequence, the matching calibration mechanism is entered, which refers to updating the default period matching threshold by calculating the information jump rate and the information revisit rate to generate a proofreading ratio to determine whether to process the nighttime recommendation information for withdrawal;

[0075] A preset statistical period is set, and the total number of accesses is obtained by counting the click volume of all nighttime recommendation information accessed by the user in the preset statistical period, and the user's stay duration on the recommendation page corresponding to the nighttime recommendation information is recorded.

[0076] If the stay duration is less than the preset stay duration threshold, the access behavior of the user accessing the nighttime recommendation information is marked as a jump-out event;

[0077] Otherwise, the access is not marked.

[0078] The total number of jumps-out is obtained by counting the number of jump-out events, and the ratio of the total number of jumps-outs to the total number of accesses is taken as the information jump rate.

[0079] It needs to be explained that the preset statistical period is used to limit the statistical time window of user behavior data, which can be set according to the night user active period and the push frequency of night recommendation information; the preset stay duration threshold is used to judge whether the user access to the recommendation information constitutes an effective access, which can be set according to the stay duration distribution of historical access data and the average reading time of content, for example, the historical access data shows that the average stay duration of the user is about 6 seconds, and the preset stay duration threshold can be set between 3 seconds and 5 seconds.

[0080] For each night recommendation information, the access user list is called through the access log database in the preset statistical period, and the number of accesses of the user to the night recommendation information is counted.

[0081] If the number of accesses of the user to the night recommendation information is greater than 1, the user is marked as a repeated click user; otherwise, the user is not marked;

[0082] The number of repeated click users for each night recommendation information is accumulated to obtain the number of repeated click users, and the ratio of the number of repeated click users to the total number of users in the access user list of the night recommendation information is taken as the proportion of repeated click users;

[0083] The proportion of repeated click users of each night recommendation information is accumulated, and the ratio of the total number of night recommendation information is taken as the information revisit rate.

[0084] The proportion of repeated click users is used to measure the proportion of repeated access behavior of a single night recommendation information in the access user group. The proportion of repeated click users of a single night recommendation information is used to represent the information revisit rate, and the result is easily affected by extreme data of individual information, resulting in distortion of the overall index. By accumulating the proportion of repeated click users of all night recommendation information and normalizing the division operation with the total number of night recommendation information, the average value of the proportion of repeated click users of each night recommendation information is calculated as the information revisit rate, thereby balancing the differences between different night recommendation information, and making the information revisit rate index objectively reflect the revisit characteristics of the overall night recommendation information.

[0085] The default period matching threshold of the night recommendation information is called through the recommendation strategy configuration library.

[0086] It needs to be explained that the access log database is used to store the interaction data of the user with the night recommendation information in the preset statistical period, and the access user list is called through the access log database according to the night recommendation information identifier; the recommendation strategy configuration library is a data storage unit for centralized management of night recommendation information related strategy parameters, which stores and calls the configuration related to the recommendation algorithm. In this embodiment, the default period matching threshold of the night recommendation information is obtained.

[0087] The default period matching threshold is matched with the correction ratio calculated by the comprehensive information bounce-off rate and information revisit rate: wherein, is the information revisit rate, is the information bounce-off rate, and is a preset adjustment factor, is the correction ratio, is a preset calibration coefficient.

[0088] The smaller the information bounce-off rate is, the higher the matching degree of the night recommendation information with the user interest is, and the greater the correction ratio is, the higher the default period matching threshold is, and the greater the information revisit rate is, the higher the repeated access frequency of the user to the night recommendation information is, and the greater the correction ratio is, the higher the default period matching threshold is;

[0089] It should be explained that the preset adjustment factor is used to control the weight of the information bounce-off rate and the information revisit rate in the calculation of the correction ratio, and the value range is 0 to 1 interval, which is set according to the historical recommendation effect and the user group characteristics, for example, when the information bounce-off rate counted in the continuous multiple monitoring periods is high, the preset adjustment factor of the information bounce-off rate is increased to enhance the down-regulation effect of the default period matching threshold; the preset calibration coefficient is used to control the sensitivity and amplitude of the correction ratio to the default period matching threshold, and the value range is 0 to 1 interval, which can be set according to the historical recommendation effect and the user behavior fluctuation degree, for example, when the user behavior in the historical data is stable, the value of the preset calibration coefficient can be increased.

[0090] The correction ratio is calculated by the comprehensive information bounce-off rate and information revisit rate, which can reflect the real interest change of the user to the recommendation information in the monitoring period, capture the interactive state of the user in real time, and improve the dynamic matching degree and timeliness of the night recommendation information and the user.

[0091] In step S4, the product of the correction ratio and the default period matching threshold is taken as the period matching threshold.

[0092] The period matching threshold is compared with the adhesion benefit index of the night recommendation information to determine whether to perform the withdrawal process:

[0093] If the adhesion benefit index is lower than the period matching threshold, the night recommendation information is processed for withdrawal.

[0094] Otherwise, the night recommendation information is retained for display.

[0095] When the correction ratio is greater than 1, the default period matching threshold is increased, thereby increasing the display range of the night recommendation information, and when the correction ratio is less than 1, the default period matching threshold is decreased, thereby reducing the display range of the night recommendation information.

[0096] When the adhesion benefit index is lower than the period matching threshold value, it indicates that the user adhesion of the corresponding night recommendation information is low, and the matching degree of the night recommendation information with the user is insufficient, so the corresponding night recommendation information is removed from the initial recommendation sequence; when the adhesion benefit index is higher than or equal to the period matching threshold value, it indicates that the user adhesion of the corresponding night recommendation information is high, and the corresponding night recommendation information is retained in the initial recommendation sequence for continuous display.

[0097] By applying the proofreading ratio to the default period matching threshold value and generating the period matching threshold value, the recommendation range of the night recommendation information can be adjusted, the matching between the recommendation result and the user behavior characteristics is ensured, the adhesion benefit index is compared with the corrected threshold value, and the removal processing is triggered, and the overall recommendation efficiency is improved.

[0098] Finally, it should be noted that in this document, the terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0099] Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including a" does not exclude the presence of another identical element in the process, method, article or device including the element.

[0100] In this document, the singular forms "a", "an" and "the" can also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms "include", "contain" or "have" and the like specify the presence of the stated features, integers, steps, operations, components, parts or combinations thereof, but do not exclude the presence or addition of one or more other features, integers, steps, operations, components, parts or combinations thereof. The possibility, while the term "and / or" used in this specification includes any and all combinations of the related listed items.

[0101] The various embodiments in the specification are described in a progressive manner, each embodiment focuses on the difference from other embodiments, and each embodiment can be combined as needed, and the same and similar parts refer to each other.

[0102] Those skilled in the art will appreciate that the foregoing description is by way of example only, and is not intended to limit the application solely thereto. Numerous modifications and adaptations will be apparent to those skilled in this art. Therefore, the true scope of the application is indicated by the appended claims, rather than by the foregoing description, and all modifications which come within the scope of the claims are intended to be embraced therein.

Claims

1. A method for generating intelligent risk control and audit strategies based on big data analysis, characterized in that: The method comprises the following steps: Step S1: calling the push data of each recommendation information in the information database, screening the night recommendation information according to the push data, setting the monitoring time, and counting the user flow and the click quantity of the night recommendation information in the monitoring time; Step S2: generating the access stickiness of the night recommendation information by comprehensively considering the user flow and the click quantity of the night recommendation information, collecting the display duration of the night recommendation information, setting the recommendation weight based on the display duration of the night recommendation information, calculating the stickiness benefit index of the night recommendation information by combining the recommendation weight with the access stickiness, and updating the initial recommendation sequence of each night recommendation information based on the stickiness benefit index; Step S3: after the initial recommendation sequence is updated, the matching calibration mechanism is inputted, the information jump rate and the information revisit rate are counted, the default time period matching threshold of the night recommendation information is called, and the calibration proportion of the default time period matching threshold is calculated by comprehensively considering the information jump rate and the information revisit rate; Step S4: the default time period matching threshold is corrected by using the calibration proportion to generate the time period matching threshold, and it is judged whether the night recommendation information is processed by comparing the stickiness benefit index of each night recommendation information with the time period matching threshold.

2. The intelligent risk control audit strategy generation method based on big data analysis according to claim 1, characterized in that: in step S1, the push data of each recommendation information in the information database is called, the recommendation information with the night period as the delivery time period is obtained by searching and screening the push data, and the recommendation information is defined as the night recommendation information; the user access quantity of the night recommendation information in the preset monitoring time is obtained as the user flow through the user access log; the click quantity of the night recommendation information in the preset monitoring time is obtained as the night recommendation information click quantity through the click behavior log.

3. The intelligent risk control audit strategy generation method based on big data analysis according to claim 2, characterized in that: in step S2, the ratio of the night recommendation information click quantity to the user flow is taken as the access stickiness of the night recommendation information; the display duration of the night recommendation information is collected through the display behavior log, and the display duration refers to the cumulative time of the actual display of the night recommendation information on the terminal; the recommendation weight is obtained by normalizing the display duration.

4. The intelligent risk control audit strategy generation method based on big data analysis according to claim 3, characterized in that: in step S2, the access stickiness and the recommendation weight are weighted and summed to obtain the stickiness benefit index; the initial recommendation sequence of the night recommendation information is called, and the initial recommendation sequence refers to the randomly generated night recommendation information arrangement sequence before considering the stickiness benefit index; the night recommendation information in the initial recommendation sequence is reordered according to the stickiness benefit index from large to small to form the optimized recommendation sequence.

5. The intelligent risk control audit strategy generation method based on big data analysis according to claim 1, characterized in that: in step S3, the matching calibration mechanism is inputted after the initial recommendation sequence is updated. A preset statistical period is set, and the total number of visits is obtained by counting the click volume of the user accessing all the night recommendation information in the preset statistical period, and the stay duration of the user on the recommendation page corresponding to the night recommendation information is recorded; If the stay duration is less than the preset stay duration threshold, the access behavior of the user accessing the night recommendation information is marked as a bounce event; Otherwise, the access behavior is not marked. The total number of bounces is obtained by counting the number of bounce events, and the information bounce rate is the ratio of the total number of bounces to the total number of visits.

6. The intelligent risk control audit strategy generation method based on big data analysis according to claim 5, characterized in that: In step S3, for each night recommendation information, the access user list is retrieved through the access log database within the preset statistical period, and the number of visits of the user to the night recommendation information is counted; If the number of visits of the user to the night recommendation information is greater than 1, the user is marked as a repeat click user; Otherwise, the user is not marked.

7. The intelligent risk control audit strategy generation method based on big data analysis according to claim 6, characterized in that: In step S3, the number of repeat click users for each night recommendation information is accumulated to obtain the number of repeat click users; The ratio of the number of repeat click users to the total number of users in the access user list of the night recommendation information is taken as the repeat click user proportion; The repeat click user proportion of each night recommendation information is added up, and the ratio of the sum to the total number of night recommendation information is taken as the information revisit rate.

8. The intelligent risk control audit strategy generation method based on big data analysis according to claim 7, characterized in that: In step S3, the default time period matching threshold of the night recommendation information is called through the recommendation strategy configuration library; The correction ratio of the default period matching threshold is calculated based on the information jump-out rate and the information revisit rate: wherein, is the information revisit rate, is the information jump-out rate, and is a preset adjustment factor, is the correction ratio, is a preset calibration coefficient.

9. The intelligent risk control audit strategy generation method based on big data analysis according to claim 1, characterized in that: In step S4, the product of the proofreading proportion and the default time period matching threshold is taken as the time period matching threshold; The time period matching threshold is compared with the stickiness benefit index of the night recommendation information to determine whether to perform the next withdrawal process: If the stickiness benefit index is lower than the time period matching threshold, the night recommendation information is subjected to the next withdrawal process; Otherwise, the night recommendation information is retained for display.

Images