Game order optimization system based on user behaviors

Abstract

The invention relates to the technical field of behavior analysis, in particular to a game order optimization system based on user behaviors. According to the method and the device, the trend of the recharging behavior of the user is accurately judged, so that the rising and falling time periods of the recharging limit can be identified, and data support is provided for subsequent user behavior analysis. In combination with the game staying duration in the recharging trend interval, the preferred game category of the user in the recharging active period can be further revealed, so that order optimization can be dynamically adjusted according to real user interests. By utilizing the stability analysis of the staying behavior, whether the preference game of the user in the recharging active stage has continuity or not can be measured, so that the interference of short-term interest fluctuation on order sorting is effectively avoided. The jump-out behaviors during order browsing are aggregated and analyzed, so that the system can identify key positions causing user loss, and the display logic of order recommendation is optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of behavior analysis, and particularly to a game order optimization system based on user behavior. Background Art

[0002] The technical field of behavior analysis includes the collection, modeling, analysis, and application of user behavior data in a specific environment. The core content of this technical field is to record and organize user behavior data, excavate user behavior patterns and behavior preferences, so as to realize functions such as behavior prediction, user portrait construction, and personalized recommendation.

[0003] Among them, the game order optimization system based on user behavior refers to collecting and analyzing data such as the browsing behavior, click behavior, purchase behavior, and preference settings of game users on the game platform, establishing a user behavior data set, and formulating corresponding game order recommendation or sorting schemes according to specific behavior characteristics such as user activity, consumption habits, and game preferences in this data set.

[0004] In the existing technology during the order optimization process, it mainly relies on users' browsing, clicking, and purchase records, and it is difficult to accurately capture the dynamic changes of users' interests, resulting in the recommended content being prone to falling into an inherent pattern. Since the user recharge behavior trend is not combined, the system cannot timely identify the consumption activity of users during a specific time period, making the recommended content may be out of touch with the current needs of users and reducing the conversion rate. The attention to the game stay behavior is relatively low, and the lack of in-depth analysis of the stay time fluctuations leads to a lag in the system's judgment of users' interests, and it is unable to effectively distinguish short-term interest fluctuations from long-term preferences. The order recommendation logic is single, and it fails to optimize by combining the bounce behavior when users browse the list, making some recommended content may have obvious user rejection characteristics, increasing the proportion of ineffective recommendations. Due to the lack of targeted adjustment, the order sorting method cannot adapt to the real needs of different user groups, affecting the accuracy of the recommendation system and the user experience. Summary of the Invention

[0005] The purpose of the present invention is to solve the disadvantages existing in the prior art, and to propose a game order optimization system based on user behavior.

[0006] To achieve the above object, the present invention adopts the following technical solutions: The game order optimization system based on user behavior includes: The recharge trend discrimination module collects the game recharge amounts of users at specified time nodes within a set time window, marks the rising trend section and the falling section of the recharge amount therefrom, and sets recharge trend section labels; The stay duration sorting module extracts the game session stay durations of the user group in the rising trend section marked by the recharge trend section label during the corresponding time period, and obtains a recharge stay fluctuation corresponding data set; The label stability recognition module screens the dataset corresponding to the recharge stay fluctuations, extracts the preferred game labels corresponding to the user within the recharge rising section after screening, determines whether the stay behavior of the user for the specified game label remains stable within a continuous time period, and obtains the label stability analysis result; The jump-out clustering analysis module extracts the label stability analysis result, records the position serial number where each jump-out behavior occurs during the process of the unstable user group browsing the order list, counts the frequency of the jump-out positions of the label-unstable user group under each label, and screens the label jump-out clustering interval; The sorting priority adjustment module obtains the label jump-out clustering interval, adjusts the sorting order of the orders in the corresponding order recommendation list, and generates an optimized label clustering order sequence.

[0007] As a further solution of the present invention, the recharge trend section label includes an upward trend section, a downward section, and a recharge slope change interval. The dataset corresponding to the recharge stay fluctuations includes the stay duration fluctuation range, the recharge trend change type, and the user stay behavior characteristics. The label stability analysis result includes an active fluctuation coefficient, a stability threshold comparison result, and a user behavior stability determination. The label jump-out clustering interval includes the distribution of jump-out position serial numbers, the corresponding jump-out frequency of the order label, and the label jump-out behavior characteristics. The optimized label clustering order sequence includes the adjusted order sorting, the click-through conversion rate comparison result, and the optimized structure of the recommendation list.

[0008] As a further solution of the present invention, the specific steps for obtaining the recharge trend section label are as follows: The recharge data collection sub-module collects the game recharge amounts of the user at specified time nodes within a set time window, constructs a corresponding time series, uses the time nodes in the time series as the horizontal axis data, and the corresponding recharge amounts as the vertical axis data to construct a two-dimensional coordinate point set of time and recharge amounts; The recharge slope calculation sub-module is based on the two-dimensional coordinate point set of time and recharge amounts, and uses the formula: ; Calculate the recharge slope value of the th time node in the two-dimensional coordinate point set of time and recharge amounts ; Among them, represents the recharge amount of the th time node, represents the recharge amount of the previous time node , represents the time point of the th time node, represents the time point of the previous time node ; Based on the recharge slope value, the trend segment marking sub-module filters consecutive positive rate intervals as the upward trend segments of the recharge amount, and extracts the segments with negative rate and decline exceeding the recharge slope decline threshold as the downward segments, obtaining the recharge trend segment labels.

[0009] As a further solution of the present invention, the acquisition steps of the recharge stay fluctuation corresponding data set are specifically as follows: The game stay duration extraction sub-module extracts the game session stay durations of the user group within the upward trend segments marked in the recharge trend segment labels, calculates the single-session stay time based on the user login time and logout time, and accumulates the total stay duration of the corresponding users within the corresponding time period to obtain the user stay duration data; The stay duration fluctuation calculation sub-module calculates the stay duration fluctuation situation of the users within the recharge upward trend segment based on the user stay duration data, sets the upper and lower limits of the fluctuation interval, and obtains the stay duration fluctuation interval; The recharge stay trend correlation sub-module forms the corresponding relationship between the stay duration fluctuation of the users within the recharge upward trend segment and the recharge trend based on the stay duration fluctuation interval, and matches the user behavior patterns according to the data correlation to obtain the recharge stay fluctuation corresponding data set.

[0010] As a further solution of the present invention, the acquisition steps of the label stability analysis result are specifically as follows: The user screening sub-module screens the user group with a stay duration fluctuation interval greater than the stay fluctuation threshold in the recharge stay fluctuation corresponding data set, extracts the preferred game labels of the screened users within the recharge upward segment, and the number of stay sessions corresponding to each preferred game label of the users, obtaining the label stay session statistical data; The label activity calculation sub-module is based on the label stay session statistical data and uses the formula: ; Calculate the label activity fluctuation coefficient of the game label ; Among them, represents the user 's number of stay sessions under the corresponding label, represents the average number of stay sessions of all users under the corresponding label, represents the total stay duration of the corresponding label, represents the average stay duration of all labels, and respectively represent the maximum and minimum values of the number of stay sessions of the users under the corresponding label, and respectively represent the maximum and minimum values of the stay duration under the corresponding label; The tag stability determination submodule compares the tag activity fluctuation coefficient with a preset stability threshold to determine whether the user's stay behavior for the specified game tag remains stable within a continuous time period, and obtains a tag stability analysis result.

[0011] As a further solution of the present invention, the step of obtaining the tag jumping out of the aggregation interval is specifically: The active fluctuation screening submodule calls the tag stability analysis result based on the tag stability analysis result, extracts the game tags whose active fluctuation coefficient is lower than the stability threshold, identifies the unstable user group with the corresponding game tags, and obtains the jump position sequence of the unstable user group; The jump-out behavior recording submodule monitors the jump-out behavior of users when browsing the order list based on the game tag list of the unstable user group, records the position sequence number of each jump-out behavior, organizes the data by user dimension, and establishes the jump-out position sequence of the user group; The tag jump-out aggregation interval screening submodule is based on the user group jump-out position sequence and adopts the formula: ; Calculation Tags Concentration of bounce frequency , filter the tags with the most concentrated bounce behaviors and obtain the tag bounce concentration interval; in, Representative User In the tag The frequency of the jump out position below, Represented in the label The total number of users who bounced.

[0012] As a further solution of the present invention, the step of obtaining the optimized tag-aggregated order sequence is specifically as follows: The order conversion rate comparison submodule obtains the number of clicks and the number of transactions for each order based on the orders in the tag-out aggregation interval, calculates the click-through conversion rate, compares it with other orders in the order pool, and calculates the conversion rate ranking of the orders in the tag-out aggregation interval in the overall order pool to generate an order conversion rate comparison result; The order recommendation ranking adjustment submodule compares the relative positions of the orders in the conversion rate ranking based on the order conversion rate comparison results, rearranges the display order of the orders in the order recommendation list, adjusts the ranking of the orders in the recommendation list, and generates an optimized tag-aggregated order sequence.

[0013] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by accurately discriminating the trend of users' recharge behavior, it is possible to identify the time periods when the recharge amount increases and decreases, providing data support for subsequent user behavior analysis. Combining the game stay duration within the recharge trend interval can further reveal the preferred game categories of users during the recharge active period, enabling dynamic adjustment of order optimization based on real user interests. By analyzing the stability of stay behavior, it is possible to measure whether the preferred games of users during the recharge active stage are persistent, thus effectively avoiding the interference of short-term interest fluctuations on order ranking. Conducting clustering analysis on the bounce behavior when browsing orders enables the system to identify the key positions leading to user loss and optimize the display logic of order recommendations. Based on the clustering interval of bounce behavior, adjust the order ranking strategy to ensure that tags with a high bounce rate can be specifically optimized, thereby improving the click-through conversion rate. The overall logic ensures that order recommendations no longer rely solely on users' historical purchase behavior, but rather combine real-time recharge trends, game stay stability, and bounce behavior characteristics to achieve more refined order optimization. Through this system, it is possible to effectively reduce the number of recommendations for inefficient orders, increase users' acceptance of recommended content, and further optimize the conversion rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is the system flow chart of the present invention; Figure 2 is the flow chart of the recharge trend discrimination module of the present invention; Figure 3 is the flow chart of the stay duration sorting module of the present invention; Figure 4 is the flow chart of the label stability identification module of the present invention; Figure 5 is the flow chart of the bounce clustering analysis module of the present invention; Figure 6 is the flow chart of the sorting priority adjustment module of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0015] In order to make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0016] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0017] Please refer to Figure 1 , the game order optimization system based on user behavior includes: The recharge trend discrimination module collects the game recharge amounts of users at specified time nodes within a set time window and constructs a corresponding time series. Taking the time nodes in the time series as the horizontal axis data and the corresponding recharge amounts as the vertical axis data, it constructs a two-dimensional coordinate point set of time and recharge amount to show the mapping relationship between the recharge amount and time. It calculates the slope values between all adjacent time nodes in the two-dimensional coordinate point set of time and recharge amount, marks the increase amplitude of the continuously positive slope segments to form the recharge amount upward trend section, and the downward section where the slope value is negative and the decline exceeds the recharge slope decline threshold, and obtains the recharge trend section label; The stay duration sorting module extracts the game session stay durations of the user group in the upward trend section marked by the recharge trend section label within the corresponding time period, calculates the stay duration fluctuation range, forms the corresponding relationship between the stay duration fluctuation of the user in the recharge upward section and the recharge trend, and obtains the recharge stay fluctuation corresponding data set; The label stability recognition module screens the user group with a stay duration fluctuation range greater than the stay fluctuation threshold in the recharge stay fluctuation corresponding data set, extracts the corresponding preferred game labels of the screened users in the recharge upward section, counts the number of stay sessions of each preferred game label by the users in the recharge upward section, calculates the active fluctuation coefficient of the corresponding label according to the number of stay sessions, compares it with the preset stability threshold, and determines whether the stay behavior of the user for the specified game label remains stable within a continuous time period, and obtains the label stability analysis result; The bounce clustering analysis module extracts the game labels with an active fluctuation coefficient lower than the stability threshold and the unstable user groups with game labels in the label stability analysis result, records the position serial numbers of each bounce behavior during the process of the unstable user group browsing the order list, establishes an association between the bounce position serial numbers and the corresponding order labels, counts the frequencies of the bounce positions of the label-unstable user group under different labels, and screens the label bounce clustering interval; The sorting priority adjustment module obtains the click-through conversion rate of the orders in the label jump-out aggregation interval, compares it with the conversion rates of other orders in the order pool, adjusts the sorting order of the orders in the corresponding order recommendation list, and generates an optimized label aggregation order sequence; The recharge trend section label includes the upward trend section, the downward section, and the recharge slope change interval. The recharge stay fluctuation corresponding data set includes the stay duration fluctuation range, the recharge trend change type, and the user stay behavior characteristics. The label stability analysis result includes the active fluctuation coefficient, the stability threshold comparison result, and the user behavior stability determination. The label jump-out aggregation interval includes the jump-out position serial number distribution, the corresponding jump-out frequency of the order label, and the label jump-out behavior characteristics. The optimized label aggregation order sequence includes the adjusted order sorting, the click-through conversion rate comparison result, and the optimized structure of the recommendation list.

[0018] Please refer to Figure 2 , and the specific steps for obtaining the recharge trend section label are as follows: The recharge data collection sub-module collects the game recharge amounts of users at specified time nodes within the set time window, constructs the corresponding time series, uses the time nodes in the time series as the horizontal axis data, and the corresponding recharge amounts as the vertical axis data to construct a two-dimensional coordinate point set of time and recharge amount; First, set the specific range of the time window, such as 24 hours, 7 days, or 30 days, and select a suitable time span according to the recharge mode of the game and the user activity. For a certain mobile game, the daily active peak of its players usually appears from 8 pm to 11 pm. Therefore, a 24-hour window can be selected, and the recharge amount is recorded in hours. Set an example. Suppose the recharge data of a certain player within 24 hours is shown in Table 1:

[0019] On this basis, construct a time series, use the time nodes in the time series as the horizontal axis data, and the corresponding recharge amounts as the vertical axis data to draw a two-dimensional coordinate point set of time and recharge amount, and show the mapping relationship between the recharge amount and time. In the visualization process, select the recharge data of the sample player group to form a two-dimensional coordinate point set. For example, the recharge data points of player A are (0, 10), (1, 5), (2, 0), (3, 20) respectively. All the data points form a recharge change curve and provide a data basis for subsequent trend analysis to obtain a two-dimensional coordinate point set of time and recharge amount.

[0020] The recharge slope calculation sub-module is based on the two-dimensional coordinate point set of time and recharge amount, and uses the formula: ; Calculate the recharge slope value of the th time node in the two-dimensional coordinate point set of time and recharge amount; Among them, represents the recharge amount at the th time node, represents the recharge amount at the previous time node , represents the time point of the th time node, represents the time point of the previous time node. The recharge fluctuation adjustment factor is , and the time interval smoothing factor is . The main purpose of setting the recharge fluctuation adjustment factor and the time interval smoothing factor is to measure the recharge slope between time nodes more accurately and stably, and to avoid extreme or distorted calculation results caused by sharp fluctuations in recharge amounts or uneven time intervals. .

[0021] Suppose a player's recharge amounts at the 3rd hour and the 4th hour are 20 yuan and 50 yuan respectively, that is , and the time interval is 1 hour, that is . Calculate its recharge slope: ; This value indicates that the recharge slope from the 3rd hour to the 4th hour is 11.53, reflecting the growth rate of the recharge amount within this time interval. After calculating the recharge slope values for all adjacent time nodes, the recharge slope values between time nodes are obtained.

[0022] Based on the recharge slope values, the trend section marking sub-module screens out continuous positive rate intervals as the upward trend sections of the recharge amount, and extracts sections with negative rates and a decline exceeding the recharge slope decline threshold as the downward sections, obtaining recharge trend section labels; Screen out continuous positive slope intervals as the upward trend sections of the recharge amount. For sections where the slopes between adjacent time nodes are all positive and continuously increasing, they are defined as upward trend sections. For example, in the above example, and are both positive values, and , so this interval belongs to the upward trend section; at the same time, for sections with negative slope values and a decline exceeding the recharge slope decline threshold, they are defined as downward sections. For example, the calculation results of the recharge slope for a certain time period are shown in Table 2:

[0023] Assume that the preset recharge slope decline threshold is 10. When and the decline , then it is determined that this interval is a significant downward section, obtaining the recharge trend section label.

[0024] Please refer to Figure 3, the steps for obtaining the dataset corresponding to the recharge stay fluctuation are specifically as follows: The game stay duration extraction sub-module extracts the game session stay durations of the user group within the rising trend section marked in the recharge trend section label, calculates the single-session stay time based on the user's login time and logout time, and accumulates the total stay duration of the corresponding user within the corresponding time period to obtain the user stay duration data; The game session data of the user group corresponding to the rising trend section marked in the recharge trend section label needs to be extracted first, and the login time and logout time of each user within this time period need to be clearly recorded, and the start time of each game session is set and the end time , and calculate the stay duration of a single session as , in order to ensure data accuracy, for the situation of reconnecting due to disconnection or logging in again after a short disconnection, the consecutive sessions occurring within a short time (such as within 5 minutes) need to be merged. The following is an example data setting: Table 3 Game Session Duration Record Table

[0025] As shown in Table 3, the single-game stay duration of each user has been calculated and stored. On this basis, calculate the total stay duration of the user within the entire rising trend section. If the user has multiple sessions within this time period, sum up the stay times of all sessions. For example, if user U001 has three sessions within a rising trend section, lasting 1820 seconds, 2400 seconds, and 1150 seconds respectively, then the total stay duration is calculated as follows: , and the user stay duration data is obtained.

[0026] The stay duration fluctuation calculation sub-module calculates the stay duration fluctuation situation of the user within the recharge rising trend section based on the user stay duration data, sets the upper and lower limits of the fluctuation range, and obtains the stay duration fluctuation range; Set the upper and lower limits of the fluctuation range, using the formula: , ; Calculate the upper limit and the lower limit of the stay duration fluctuation range; where, represents the average stay duration of the user within this range, represents the standard deviation of the user's stay duration, is the interval adjustment factor, generally taking values from 1.5 to 2 to control the interval range.

[0027] First, calculate the average stay duration of all users and standard deviation to determine the fluctuation range, set the dwell time data of the sample user group as follows:

[0028] Calculate its mean value : .

[0029] Calculate the standard deviation : .

[0030] Set the interval adjustment factor , and calculate the upper and lower limits of the dwell time fluctuation interval: ; 、 This calculation shows that the change in the user's dwell time between 4671.2 seconds and 7136.8 seconds belongs to the normal fluctuation range, and the dwell time fluctuation interval is obtained.

[0031] The recharge dwell trend correlation sub-module forms the correspondence between the dwell time fluctuation and the recharge trend in the recharge upward trend section based on the dwell time fluctuation interval, and matches the user behavior pattern according to the data correlation to obtain the recharge dwell fluctuation corresponding data set; Call the dwell time fluctuation interval, analyze the change pattern of the dwell time of users in the recharge upward trend section, determine whether the dwell time shows a stable or large fluctuation trend, and calculate the volatility of the dwell time of each user within this interval : ; Among them, is the total dwell time of the user in the recharge upward trend section.

[0032] If , then the change in the user's dwell time is within the normal fluctuation range. If , then the dwell time of the user fluctuates greatly. For example, if the total dwell time of a user is 7250 seconds, then its volatility is calculated as follows: ; Since , it is determined that the dwell behavior of this user has a large fluctuation, and further analyze its impact on the recharge trend to obtain the recharge dwell fluctuation corresponding data set.

[0033] Please refer to Figure 4 , and the specific steps for obtaining the label stability analysis result are as follows: The user screening sub-module screens the user group with a dwell time fluctuation range greater than the dwell fluctuation threshold in the dataset corresponding to the recharge dwell fluctuation, extracts the preferred game tags of the screened users within the recharge ascending section, and the number of dwell sessions of the users corresponding to each preferred game tag, and obtains the tag dwell session statistical data; The dataset corresponding to the recharge dwell fluctuation contains the dwell time data of users within the recharge ascending trend section. To screen out users with a dwell time fluctuation range exceeding the preset dwell fluctuation threshold, it is first necessary to determine the calculation method of the dwell time fluctuation, obtain the dwell time data sequence of all users, calculate the fluctuation range of each user, and set an example data: Table 4 User Dwell Time Data

[0034] As shown in Table 4, the dwell time fluctuation range of each user has been calculated. The preset dwell fluctuation threshold is 1,000 seconds. Then U001 and U002 meet the screening conditions, and U003 is excluded. Among the screened user group, extract their preferred game tags within the recharge ascending trend section, that is, the game types where users mainly dwell during this period. For example, U001 mainly dwells in strategy games, and U002 mainly dwells in role-playing games. Calculate the number of dwell sessions of the users corresponding to this game tag. If U001 has 12 dwell sessions in strategy games and U002 has 15 dwell sessions in role-playing games, the statistical results are as follows: Table 5 Preferred Game Tags and Number of Dwell Sessions

[0035] Finally, obtain the tag dwell session statistical data.

[0036] The tag activity calculation sub-module, based on the tag dwell session statistical data, uses the formula: ; Calculate the tag activity fluctuation coefficient of the game tag ; Among them, represents the number of dwell sessions of user under this tag, represents the average number of dwell sessions of all users under this tag, represents the total dwell time of this tag, represents the average dwell time of all tags, and represent the maximum and minimum values of the number of dwell sessions of users under this tag respectively, and respectively represent the maximum and minimum values of the dwell time under this label, cross term is used to measure the fluctuation intensity of users on different labels, interval span term is used to evaluate the overall active change range.

[0037] If the number of user stay sessions for a certain strategy game label is 12, 10, 14, 13, 11 respectively, then calculate: .

[0038] At the same time, assume that the dwell times under this label are 5,400, 6,200, 5,900, 6,000, 5,800 respectively, then calculate: .

[0039] Calculate the difference term between the maximum and minimum values: , .

[0040] Calculate the cross term: .

[0041] Finally calculate : .

[0042] Finally obtain the label active fluctuation coefficient.

[0043] The label stability determination sub-module compares the label active fluctuation coefficient with a preset stability threshold to determine whether the dwell behavior of users on the specified game label remains stable within a continuous time period, and obtains the label stability analysis result; Call the label active fluctuation coefficient, compare it with the preset stability threshold, determine whether the dwell behavior of users on the specified game label remains stable within a continuous time period, and set the stability threshold (For example, in a certain game environment, the fluctuation coefficients of 70% of users are concentrated between 2,200 and 2,800, then this range can be used as a reference benchmark), if exceeds this threshold, it is determined that the label has a large active fluctuation during this time period. For example, the calculated in the above example satisfies , so the user behavior of this label fluctuates greatly. If the of a certain role-playing game label, then it satisfies , and it is judged as stable behavior, and finally the label stability analysis result is obtained.

[0044] Please refer to Figure 5 , the specific steps for obtaining the label out-of-cluster interval are as follows: Based on the label stability analysis results, the active fluctuation screening sub-module calls the label stability analysis results, extracts game labels with active fluctuation coefficients lower than the stability threshold, identifies unstable user groups with such game labels, and obtains the jump position sequence of the unstable user groups. Call the label stability analysis results, compare the active fluctuation coefficients of each game label, and determine whether it is lower than the stability threshold. The active fluctuation coefficient can be calculated from historical user interaction behaviors, usually calculated using the number of active times within a certain time window. For example, assume the number of active times of a certain game label in the past 7 days are: Then first calculate the mean of the active times as

[0045] Then calculate the variance The calculation formula for variance is: Substitute the data for calculation: .

[0046] Calculate the standard deviation : .

[0047] Calculate the fluctuation coefficient : .

[0048] Assume the stability threshold is set to 0.30. If , then the label is identified as unstable. In the above calculation, , so the label is classified as an unstable label. Then, perform the same calculation steps for all game labels, extract all eligible unstable game labels, and then screen the user groups with these game labels in the user behavior database, and establish their corresponding game label sets to generate a game label list of the unstable user groups. Based on the game label list of the unstable user groups, the jump behavior recording sub-module monitors the jump behavior of users when browsing the order list, records the position serial number of each jump behavior occurrence, sorts the data by user dimension, and establishes a user group jump position sequence. Extract the access trajectories of unstable users from the order browsing data, monitor their jump behavior in the order list. The jump behavior is usually defined as the situation where the user leaves the page without completing key operations (such as placing an order, favoriting, etc.). Record the jump position of the user in the order list. For example, if a user jumps directly to the home page after staying in the order list for 10 seconds, this behavior is recorded as jumping out after the 2nd order. Assume the recorded jump position sequence is , then sort the data by user ID to form a user group jump position sequence.

[0049] The label jump-out clustering interval screening sub-module, based on the user group jump-out position sequence, uses the formula: ; Calculate the jump-out frequency concentration of the label , screen the label with the most concentrated jump-out behavior, and obtain the label jump-out clustering interval; Among them, represents the frequency of the jump-out position of user under the label , represents the total number of users who have a jump-out under the label , represents taking the logarithm of the sum of the frequencies.

[0050] Suppose the jump-out position frequency data under a certain label is: User 1: , User 2: , User 3: , User 4: , User 5: , then the total number of users with a jump-out , calculate: ; ; .

[0051] This value represents the jump-out frequency concentration of the label , which is used to measure whether the jump-out behavior under this label is concentrated. The reference benchmark value is set to 6.5. If , it is considered that there is a jump-out clustering phenomenon for this label, and thus the label with the most concentrated jump-out behavior is screened out to obtain the label jump-out clustering interval.

[0052] The innovation of this formula lies in the combination of the sum of squares and logarithmic operations of the jump-out frequencies, enabling the jump-out trends under different labels to not only reflect the influence of the number of jump-outs but also emphasize the non-linear growth characteristics of high-frequency jump-out situations, thereby being able to more accurately identify the jump-out clustering interval.

[0053] Please refer to Figure 6 , and the specific steps for obtaining the optimized label clustering order sequence are as follows: The order conversion rate comparison sub-module, based on the orders within the label jump-out clustering interval, obtains the number of clicks and the number of transactions for each order, calculates the click-through conversion rate, compares it with other orders in the order pool, and statistically ranks the conversion rate of the orders in the label jump-out clustering interval in the overall order pool to generate the order conversion rate comparison result; First, extract the click records of all orders from the order database to obtain the click count and conversion count for each order. The click count refers to the total number of times a user clicks on the order in the order recommendation list, and the conversion count refers to the number of times a purchase is actually completed after clicking on the order. For example, if an order is clicked 120 times by users within a certain period and 18 of them are successfully converted into orders, then the click-through rate of this order is calculated as: , that is, 15%. Similarly, perform the same calculation on all orders outside the label clustering range to obtain the click-through rate data of all orders within the label range.

[0054] Next, extract the data of all orders from the order pool and count their click counts and conversion counts to calculate the conversion rate of all orders in the order pool. Assume that the order pool contains multiple orders, and their click counts and conversion counts are shown in the following table: Table 6 Order Click and Conversion Data in the Order Pool

[0055] Then, compare the click-through rate of orders outside the label clustering range with the conversion rate of other orders in the order pool, and calculate their relative rankings among all orders. Assume that the average conversion rate of orders in the order pool is 12.0%. If the click-through rate of an order outside a certain label clustering range is 15.0%, it is higher than the overall average and ranks among the top; if the click-through rate is 9.0%, it is lower than the average and ranks lower, generating a comparison result of order conversion rates.

[0056] Based on the order conversion rate comparison result, the order recommendation sorting adjustment sub-module compares the relative positions of orders in the conversion rate ranking, rearranges the display order of orders in the order recommendation list, adjusts the sorting of orders in the recommendation list, and generates an optimized label clustering order sequence; First, according to the click-through rate ranking of orders, mark orders with higher conversion rates as priority recommendation objects. At the same time, combined with the historical conversion trend, calculate the change amount of click and conversion trends of orders in historical data. For example, assume that the click counts of an order in the past three days are 100, 120, and 140 in sequence, and the conversion counts are 10, 15, and 18 in sequence. Then calculate the click growth rate : , that is, the click growth rate is 40%. Similarly, calculate the conversion growth rate : , that is, the conversion growth rate is 80%. Such orders have an obvious growth trend and should be sorted first.

[0057] Then, sort according to the click volume, conversion rate, and growth trend of all orders, determine the new order recommendation order, and generate a recommendation list according to the new sorting. The example is as follows: Table 7 Adjusted Recommended Order Sorting

[0058] By adjusting the recommended order of the orders, the orders with higher click-through conversion rates and better transaction trends are ranked higher, and an optimized label aggregation order sequence is generated according to the optimized order.

[0059] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution content of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A game order optimization system based on user behavior, characterized in that, The system includes: The recharge trend discrimination module collects the game recharge amounts of users at specified time nodes within a set time window, marks the upward trend section and downward trend section of the recharge amount, and sets recharge trend section labels. The stay duration sorting module extracts the game session stay durations of the user group in the upward trend section marked by the recharge trend section label within the corresponding time period, and obtains a recharge stay fluctuation corresponding data set. The label stability recognition module filters the recharge stay fluctuation corresponding data set, extracts the preferred game labels corresponding to the users after filtering in the recharge upward section, and determines whether the stay behavior of the users on the specified game label remains stable within a continuous time period, and obtains a label stability analysis result. The bounce clustering analysis module extracts the label stability analysis result, records the position serial numbers of each bounce behavior during the process of the unstable user group browsing the order list, counts the frequencies of the bounce positions of the label-unstable user group under each label, and filters the label bounce clustering interval. The sorting priority adjustment module obtains the label bounce clustering interval, adjusts the sorting order of the orders in the corresponding order recommendation list, and generates an optimized label clustering order sequence.

2. The game order optimization system based on user behavior according to claim 1, characterized in that, The recharge trend section labels include an upward trend section, a downward trend section, and a recharge slope change interval. The recharge stay fluctuation corresponding data set includes a stay duration fluctuation range, a recharge trend change type, and user stay behavior characteristics. The label stability analysis result includes an active fluctuation coefficient, a stability threshold comparison result, and a user behavior stability determination. The label bounce clustering interval includes a bounce position serial number distribution, an order label corresponding bounce frequency, and a label bounce behavior characteristic. The optimized label clustering order sequence includes an adjusted order sorting, a click-through rate comparison result, and an optimized structure of the recommendation list.

3. The game order optimization system based on user behavior according to claim 1, wherein The specific steps for obtaining the recharge trend section labels are as follows: The recharge data collection sub-module collects the game recharge amounts of users at specified time nodes within a set time window, constructs a corresponding time series, uses the time nodes in the time series as the horizontal axis data, and the corresponding recharge amounts as the vertical axis data to construct a two-dimensional coordinate point set of time and recharge amount. The recharge slope calculation sub-module is based on the two-dimensional coordinate point set of time and recharge amount, and uses the formula: ； Calculate the recharge slope value at the th time node in the two-dimensional coordinate point set of time and recharge amount ; Among them, represents the recharge amount at the th time node, represents the recharge amount at the previous time node , represents the time point at the th time node, represents the time point at the previous time node ; The trend section marking sub-module is based on the recharge slope value, filters the continuous positive rate interval as the recharge amount upward trend section, and extracts the section with a negative rate and a decline exceeding the recharge slope decline threshold as the downward trend section to obtain the recharge trend section label.

4. The game order optimization system based on user behavior according to claim 1, wherein The specific steps for obtaining the recharge stay fluctuation corresponding data set are as follows: The game stay duration extraction sub-module extracts the game session stay durations of the user group in the upward trend section marked by the recharge trend section label, calculates the single-session stay time based on the user login time and logout time, and accumulates the total stay duration of the corresponding user within the corresponding time period to obtain the user stay duration data. The stay duration fluctuation calculation sub-module is based on the user stay duration data, calculates the stay duration fluctuation of the user in the recharge upward trend section, sets the upper and lower limits of the fluctuation interval, and obtains the stay duration fluctuation interval. The recharge stay trend association submodule forms a corresponding relationship between the user's stay duration fluctuation and the recharge trend in the recharge rising trend section based on the stay duration fluctuation interval, and matches the user behavior pattern according to data correlation to obtain a recharge stay fluctuation corresponding data set.

5. The game order optimization system based on user behavior according to claim 1, characterized in that The steps for obtaining the label stability analysis results are specifically as follows: The user screening submodule screens the user group whose stay duration fluctuation interval is greater than the stay fluctuation threshold in the data set corresponding to the recharge stay fluctuation, extracts the preferred game tags of the screened users in the recharge rising section, and the number of stay sessions of the users corresponding to each preferred game tag, and obtains the tag stay session statistics; The tag activity calculation submodule uses the formula based on the tag stay session statistics: ； Calculation game label Label active fluctuation coefficient of ; Among them, represents the number of stay sessions of the user under the corresponding label, represents the average number of stay sessions of all users under the corresponding label, represents the total stay duration of the corresponding label, represents the average stay duration of all labels, and respectively represent the maximum and minimum values of the number of stay sessions of users under the corresponding label, and respectively represent the maximum and minimum values of the stay duration under the corresponding label; The tag stability determination submodule compares the tag activity fluctuation coefficient with a preset stability threshold to determine whether the user's stay behavior for the specified game tag remains stable within a continuous time period, and obtains a tag stability analysis result.

6. The game order optimization system based on user behavior according to claim 1, wherein The steps for obtaining the tag jumping out of the aggregation interval are specifically as follows: The active fluctuation screening submodule calls the tag stability analysis result based on the tag stability analysis result, extracts the game tags whose active fluctuation coefficient is lower than the stability threshold, identifies the unstable user group with the corresponding game tags, and obtains the jump position sequence of the unstable user group; The jump-out behavior recording submodule monitors the jump-out behavior of users when browsing the order list based on the game tag list of the unstable user group, records the position sequence number of each jump-out behavior, organizes the data by user dimension, and establishes the jump-out position sequence of the user group; The tag jump-out aggregation interval screening submodule is based on the user group jump-out position sequence and adopts the formula: ； Calculation label The bounce frequency concentration , screen the label with the most concentrated bounce behavior, and obtain the label bounce aggregation interval; Among them, represents the frequency of the bounce position under the label by the user and represents the total number of users who bounce under the label ​ 7. The game order optimization system based on user behavior according to claim 1, wherein The steps for obtaining the optimized tag-aggregated order sequence are specifically as follows: The order conversion rate comparison submodule obtains the number of clicks and the number of transactions for each order based on the orders in the tag-out aggregation interval, calculates the click-through conversion rate, compares it with other orders in the order pool, and calculates the conversion rate ranking of the orders in the tag-out aggregation interval in the overall order pool to generate an order conversion rate comparison result; The order recommendation ranking adjustment submodule compares the relative positions of the orders in the conversion rate ranking based on the order conversion rate comparison results, rearranges the display order of the orders in the order recommendation list, adjusts the ranking of the orders in the recommendation list, and generates an optimized tag-aggregated order sequence.

Images