A personalized content operation and maintenance auxiliary system and method based on data analysis

Abstract

The present invention relates to the field of operation and maintenance assistance, and discloses a personalized content operation and maintenance assistance system and method based on data analysis, the method comprising: obtaining platform data of each user in real time, the platform data comprising the number of logins, online time, content type viewed and corresponding frequency, and the length of stay for viewing each content; inputting the number of logins and online time of each user in a unit time period into a trained neural network model, and outputting the activity of each user; calculating the stability index of the current user according to the activity, content type viewed and corresponding frequency of each user in a unit time period, judging the stability of the content type preferred by each user according to the comparison result of the stability index and a preset stability interval; selecting different recommendation strategies according to the stability of the content type preferred by each user, evaluating the quality of the content pushed after the recommendation strategy is executed, and adjusting the recommendation strategy of each user in real time according to the evaluation result.

Description

Technical Field

[0001] The present invention relates to the field of operation and maintenance assistance, and in particular to a personalized content operation and maintenance assistance system and method based on data analysis. Background Art

[0002] Operation and maintenance are essential during the operation of each platform. Operation and maintenance refers to the daily maintenance and management of systems, applications or services to ensure that they can run stably, safely and reliably.

[0003] In order to improve user satisfaction, each platform will set up automatic content recommendations. However, when actually making content recommendations, the same method is used to recommend content to users with different levels of activity. As a result, a large amount of unnecessary computing power is wasted on users with low activity levels. As a result, users who really need content recommendations cannot access the content they are really interested in, which not only affects user satisfaction but also wastes a lot of computing resources. Summary of the invention

[0004] The purpose of the present invention is to provide a personalized content operation and maintenance auxiliary system and method based on data analysis to solve the above technical problems.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A personalized content operation and maintenance auxiliary system and method based on data analysis, comprising:

[0007] S1: Acquire platform data of each user in real time, including but not limited to the number of logins, online time, content types viewed and corresponding frequency, and duration of viewing each content;

[0008] S2: Input the number of logins and online time of each user in a unit time period into the trained neural network model, and output the activity of each user;

[0009] S3: Calculate the stability index of the current user based on the activity level, content type viewed and corresponding frequency of each user in a unit time period, and compare the stability index with the preset stability interval to determine the stability of the content type preferred by each user;

[0010] S4: Select different recommendation strategies based on the stability of each user's preferred content type, evaluate the quality of the content pushed after the recommendation strategy is executed, and adjust the recommendation strategy for each user in real time based on the evaluation results.

[0011] As a further technical solution, the method for obtaining the stability index is:

[0012] By formula:

[0013] Calculate and obtain the stability index P of the current user;

[0014] In the formula, A i is the number of content types viewed in a unit time period, A0 is the predicted number of content types viewed in a unit time period, and B i is the duration of viewing each content, B0 is the predicted viewing duration based on historical data, m is the total number of contents viewed in a unit time period, and υ i1 ,υ i2 is the preset proportional coefficient, C i is the viewing frequency of each type of content in a unit time period, C0 is the predicted viewing frequency of each type of content in a unit time period, W is the activity of the current user, and K is the coefficient of variation.

[0015] As a further technical solution, the stability process of judging the content type preferred by each user is as follows:

[0016] Compare the stability index P with the preset stability interval [P0, P1];

[0017] If P∈[P0, P1], then the stability of the content type preferred by the current user is judged to be normal;

[0018] If P>P1, it is judged that the stability of the content type preferred by the current user is poor;

[0019] If P∈[0, P0], it is judged that the stability of the content type preferred by the current user is good.

[0020] As a further technical solution, the coefficient of variation

[0021]

[0022] In the formula, ε is a dynamic value; t0 and t1 are the starting point and the ending point of the unit time period, respectively. ij (t) is the time-varying curve of the number of content types viewed by the current user in a unit time period, C ij (t) is the time-varying curve of the frequency of viewing each type of content by the current user in a unit time period, q is the number of unit time periods, j1 , j2 is the conversion coefficient, which is selected based on the experimental data fitting.

[0023] As a further technical solution, the process of obtaining the dynamic value is:

[0024] By formula: Calculate and obtain the dynamic value ε of the current user;

[0025] in, It is the online time of the current user after each login. is the average online duration of the current user after each login, N is the number of logins in a unit time period, N0(t) is the curve of the number of logins in a unit time period changing with time, and Γ is the dimensionless coefficient.

[0026] As a further technical solution, the process of establishing the recommendation strategy is as follows:

[0027] For users who prefer less stable content types, randomly push each type of content;

[0028] For users who prefer a stable content type, the formula is: Calculate the recommended coefficient g for each type i , m is the number of contents corresponding to each type in a unit time period, according to the recommendation coefficient g i After sorting in descending order, select the content corresponding to the first X types for push;

[0029] For users who prefer stable content types, we will sort the content types from high to low based on the frequency of content types viewed in a unit time period, and push the content of the type that ranks first;

[0030] As a further technical solution, the process of evaluating the quality of content pushed after the recommendation strategy is executed is as follows:

[0031] Randomly obtain the time-varying curve B of the current user's stay time for viewing each content within Y preset time periods i (t) and the predicted change curve of the reading time over time B0(t);

[0032] By formula: Calculate and obtain the push quality coefficient f;

[0033] The push quality coefficient f and the warning threshold f obtained based on historical data th Compare;

[0034] If f ≥ f th , then it is judged that the quality of the content pushed after the current recommendation strategy is executed is poor and needs to be adjusted; if f<f th , it is judged that the quality of the content pushed after the current recommendation strategy is executed is good, and the execution continues.

[0035] As a further technical solution, the process of adjusting the recommendation strategy is:

[0036] Take the current time as the starting point and the end point after a unit time interval, and perform the recommendation coefficient for each type again.

[0037] If exists Then the current type is eliminated, and the last sorted X+1, X+2, ..., X+n are replaced in sequence; Δg0 is a preset value, which is formulated based on experimental data;

[0038] Otherwise, follow Re-push the content after sorting in descending order.

[0039] A personalized content operation and maintenance assistance system based on data analysis, wherein the assistance system is applicable to the above-mentioned personalized content operation and maintenance assistance method based on data analysis.

[0040] Beneficial effects of the present invention:

[0041] First, the platform data of each user is obtained, and then the number of logins and online time of each user in a unit time period are extracted and input into the trained neural network model to output the activity of each user. Then, the stability index of the current user is calculated according to the activity of each user in a unit time period, the type of content viewed and the corresponding frequency. According to the comparison result of the stability index and the preset stability interval, the stability of the content type preferred by each user is judged. Through the stability analysis of the user's preferred content type, the user can be screened; among them, for users with good stability of preferred content type, such users tend to prefer a single type of content, so only a single type of content is pushed, and there is no need to perform real-time analysis and calculation during each online process, which can effectively save computing power; for users with poor stability of preferred content type, it means that the user's favorite content type is uncertain or the user randomly browses and has no actual reading needs, so random content recommendation is performed, which also reduces computing power investment; for users with normal stability of preferred content type, such users are high-quality users of the platform, so personalized content types are recommended to such users to improve the user's satisfaction with the content browsing experience, and the quality of the content pushed after the recommendation strategy is executed is evaluated, and the recommendation strategy of each user is adjusted in real time according to the evaluation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The present invention will be further described below in conjunction with the accompanying drawings.

[0043] Figure 1 It is a diagram of the method steps of the present invention. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] See also Figure 1 As shown, the present invention is a personalized content operation and maintenance auxiliary system and method based on data analysis, including:

[0046] S1: Acquire platform data of each user in real time, including but not limited to the number of logins, online time, content types viewed and corresponding frequency, and duration of viewing each content;

[0047] S2: Input the number of logins and online time of each user in a unit time period into the trained neural network model, and output the activity of each user;

[0048] S3: Calculate the stability index of the current user based on the activity level, content type viewed and corresponding frequency of each user in a unit time period, and compare the stability index with the preset stability interval to determine the stability of the content type preferred by each user;

[0049] S4: Select different recommendation strategies based on the stability of each user's preferred content type, evaluate the quality of the content pushed after the recommendation strategy is executed, and adjust the recommendation strategy for each user in real time based on the evaluation results.

[0050] In this embodiment, the platform data of each user is first obtained, and then the number of logins and online time of each user in a unit time period are extracted and input into the trained neural network model to output the activity of each user. Then, the stability index of the current user is calculated based on the activity of each user in a unit time period, the type of content viewed, and the corresponding frequency. The stability of the content type that each user tends to prefer is determined based on the comparison result of the stability index and the preset stability interval. Through the stability analysis of the content type that the user tends to prefer, the users can be screened.

[0051] Among them, for users who prefer stable content types, they tend to prefer a single type of content, so only a single type of content can be pushed, without the need for real-time analysis and calculation during each online process, which can effectively save computing power;

[0052] For users who prefer content types with poor stability, it means that the user's favorite content types are uncertain or the user browses randomly without actual reading needs. Therefore, random content recommendations are made to reduce computing power investment.

[0053] For users with a normal stability in content type preferences, such users are high-quality users of the platform. Therefore, personalized content types are recommended to such users to improve their satisfaction with content viewing experience, and the quality of content pushed after the recommendation strategy is executed is evaluated. The recommendation strategy for each user is adjusted in real time based on the evaluation results.

[0054] The method for obtaining the stability index is:

[0055] By formula:

[0056] Calculate and obtain the stability index P of the current user;

[0057] In the formula, A i is the number of content types viewed in a unit time period, A0 is the predicted number of content types viewed in a unit time period, and B i is the duration of viewing each content, B0 is the predicted viewing duration based on historical data, m is the total number of contents viewed in a unit time period, and υ i1 ,υ i2 is the preset proportional coefficient, C i is the viewing frequency of each type of content in a unit time period, C0 is the predicted viewing frequency of each type of content in a unit time period, W is the activity of the current user, and K is the coefficient of variation.

[0058] The stability process of judging the content type preferred by each user is as follows:

[0059] Compare the stability index P with the preset stability interval [P0, P1];

[0060] If P∈[P0, P1], then the stability of the content type preferred by the current user is judged to be normal;

[0061] If P>P1, it is judged that the stability of the content type preferred by the current user is poor;

[0062] If P∈[0, P0], it is judged that the stability of the content type preferred by the current user is good.

[0063] In this embodiment, a method for obtaining a stability index is provided. Specifically, the formula is: Calculate the stability index P of the current user. Obviously, if |A i The larger the value of -A0| is, the greater the difference between the number of content types viewed by the current user in a unit time period and the predicted number of content types viewed. i The larger the value of -C0| is, the greater the difference between the frequency of viewing each type of content by the current user in a unit time period and the predicted frequency of viewing each type of content is, and the greater the stability index obtained, |B iThe larger the value of -B0| is, the greater the difference between the duration of viewing each content and the predicted viewing duration is, indicating that the user is not very interested in each content. Therefore, the worse the stability of judging the preferred content type is, the greater the stability index is. The above content can dynamically reflect the stability of each user's preferred content type, so that users can be accurately classified to achieve different types of content recommendations, thereby achieving the effect of saving computing power resources and improving user satisfaction with targeted recommendations.

[0064] The coefficient of variation

[0065]

[0066] In the formula, ε is a dynamic value; t0 and t1 are the starting point and the ending point of the unit time period, respectively. ij (t) is the time-varying curve of the number of content types viewed by the current user in a unit time period, C ij (t) is the time-varying curve of the frequency of viewing each type of content by the current user in a unit time period, q is the number of unit time periods, j1 , j2 is the conversion coefficient, which is selected based on the experimental data fitting;

[0067] The process of obtaining the dynamic value is:

[0068] By formula: Calculate and obtain the dynamic value ε of the current user;

[0069] in, It is the online time of the current user after each login. is the average online duration of the current user after each login, N is the number of logins in a unit time period, N0(t) is the curve of the number of logins in a unit time period changing with time, and Γ is the dimensionless coefficient.

[0070] In this embodiment, the formula Calculate the coefficient of variation K using the formula: The dynamic value ε of the current user is obtained by calculation; the cumulative change in the number of content types viewed by the current user in a unit time period and the cumulative change in the frequency of viewing each type of content can accurately reflect the change in the current user's preferred content type, and finally realize the dynamic impact on the stability index P, so as to improve the accuracy of the stability index P, better fit the actual situation of the user, and reduce the error; and the setting of the dynamic value can be based on the fluctuation of the online time after each login in the unit time and The slope represented by comprehensively reflects the dynamic value of the current user. Obviously, the larger the dynamic value, the greater the impact on the variation coefficient K.

[0071] The process of establishing the recommended strategy is as follows:

[0072] For users who prefer less stable content types, randomly push each type of content;

[0073] For users who prefer a stable content type, the formula is: Calculate the recommended coefficient g for each type i , m is the number of contents corresponding to each type in a unit time period, according to the recommendation coefficient g i After sorting in descending order, select the content corresponding to the first X types for push;

[0074] For users who prefer stable content types, we will sort the content types from high to low based on the frequency of content types viewed in a unit time period, and push the content of the type that ranks first;

[0075] The process of evaluating the quality of content pushed after the recommendation strategy is executed is as follows:

[0076] Randomly obtain the time-varying curve B of the current user's stay time for viewing each content within Y preset time periods i (t) and the predicted change curve of the reading time over time B0(t);

[0077] By formula: Calculate and obtain the push quality coefficient f;

[0078] The push quality coefficient f and the warning threshold f obtained based on historical data th Compare;

[0079] If f ≥ f th , then it is judged that the quality of the content pushed after the current recommendation strategy is executed is poor and needs to be adjusted; if f<f th , it is judged that the quality of the content pushed after the current recommendation strategy is executed is good, and the execution continues;

[0080] As a further technical solution, the process of adjusting the recommendation strategy is:

[0081] Take the current time as the starting point and the end point after a unit time interval, and perform the recommendation coefficient for each type again.

[0082] If exists Then the current type is eliminated, and the last sorted X+1, X+2, ..., X+n are replaced in sequence; Δg0 is a preset value, which is formulated based on experimental data;

[0083] Otherwise, follow Re-push the content after sorting in descending order.

[0084] In this embodiment, by formula: Calculate the push quality coefficient f; obviously |B i The larger the deviation of (t)-B0(t)|, the larger f is. Δρ is the reference value based on historical data and empirical data. is the preset weight coefficient; therefore, the push quality coefficient f and the warning threshold f obtained based on historical data are th Compare; if f≥f th , then it is judged that the quality of the content pushed after the current recommendation strategy is executed is poor and needs to be adjusted; if f<f th , it is judged that the quality of the content pushed after the current recommendation strategy is executed is good, and the execution continues; thus, it can be based on |B i (t)-B0(t)|, to accurately evaluate the quality of the content pushed after the recommendation strategy is executed, and make real-time adjustments to focus on improving the satisfaction of users who receive personalized pushed content.

[0085] A personalized content operation and maintenance assistance system based on data analysis, wherein the assistance system is applicable to the above-mentioned personalized content operation and maintenance assistance method based on data analysis.

[0086] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A personalized content operation and maintenance auxiliary method based on data analysis, characterized in that: include: S1: Acquire platform data of each user in real time, including but not limited to the number of logins, online time, content types viewed and corresponding frequency, and duration of viewing each content; S2: Input the number of logins and online time of each user in a unit time period into the trained neural network model, and output the activity of each user; S3: Calculate the stability index of the current user based on the activity level, content type viewed and corresponding frequency of each user in a unit time period, and compare the stability index with the preset stability interval to determine the stability of the content type preferred by each user; S4: Select different recommendation strategies based on the stability of each user's preferred content type, evaluate the quality of the content pushed after the recommendation strategy is executed, and adjust the recommendation strategy for each user in real time based on the evaluation results; The method for obtaining the stability index is: By formula: ; Calculate the stability index of the current user ; In the formula, is the number of content types viewed in a unit time period. The number of content types to be viewed for the forecast within a unit time period. To check the dwell time of each content, For the forecasted query duration based on historical data, is the total number of contents viewed in a unit time period. is the preset scale factor, The frequency of viewing each type of content in a unit time period. The predicted viewing frequency of each type of content in a unit time period. is the current user's activity level, is the coefficient of variation; The stability process of judging the content type preferred by each user is as follows: The stability index With the preset stability interval Make comparisons; like , then it is judged that the stability of the content type preferred by the current user is normal; like , then it is judged that the stability of the content type that the current user prefers is poor; like , then it is judged that the stability of the content type that the current user prefers is good; The coefficient of variation ; In the formula, is a dynamic value; are the starting point and ending point of the unit time period, respectively. It is the curve of the number of content types viewed by the current user in a unit time period over time. This is the time-varying curve of the frequency of each type of content viewed by the current user in a unit time period. is the number of unit time periods, is the conversion coefficient, which is selected based on the experimental data fitting; The process of obtaining the dynamic value is: By formula: Calculate the dynamic value of the current user ; in, It is the online time of the current user after each login. is the average online time of the current user after each login, is the number of logins within a unit time period, is the curve of the number of logins changing with time in a unit time period. is the dimensionless coefficient; The process of establishing the recommended strategy is as follows: For users who prefer less stable content types, randomly push each type of content; For users who prefer a stable content type, the formula is: Calculate the recommendation coefficient for each type , The number of contents corresponding to each type in a unit time period, according to the recommended coefficient After sorting in descending order, select the content corresponding to the first X types for push; For users who have a stable preference for content types, we will sort the content types from high to low based on the frequency of viewing them in a unit time period, and push the content of the type that ranks first.

2. The personalized content operation and maintenance auxiliary method based on data analysis according to claim 1 is characterized in that: The process of evaluating the quality of content pushed after the recommendation strategy is executed is as follows: Randomly obtain the time-varying curve of the current user's stay time for viewing each content within Y preset time periods (t) and the predicted reading time curve over time ; By formula: Calculate the push quality coefficient ; Push quality factor The warning threshold based on historical data Compare; like , then it is judged that the quality of the content pushed after the current recommendation strategy is executed is poor and needs to be adjusted; if , it is judged that the quality of the content pushed after the current recommendation strategy is executed is good, and the execution continues.

3. The personalized content operation and maintenance auxiliary method based on data analysis according to claim 2 is characterized in that: The process of adjusting the recommended strategy is: Take the current time as the starting point and the end point after a unit time interval, and perform the recommendation coefficient for each type again. ; If exists , then remove the current type and fill it in from the last sorted X+1, X+2, ...., X+n in sequence; It is a preset value, which is formulated based on experimental data; Otherwise, follow Re-push the content after sorting in descending order.

Images