A method for discovering video ringback tone clustering based on different dimension indexes

Abstract

The present application relates to a kind of quality analysis methods of vRBT video ringback tone, and the present application relates to a kind of methods for discovering video ringback tone clustering problem based on different dimension indicators. Including: step one: original data acquisition, step two: data cleaning, step three: clustering threshold statistics, step four: statistics influence user number, step five: statistics comprehensive score, step six: output clustering SDR, step seven: interface presents clustering problem. The present application is based on the actual business appeal of customer, refers to the specification of China Mobile Communication Group, adopts big data technology, automatically calculates and outputs the clustering cell of vRBT video ringback tone, clustering terminal brand, clustering core network and clustering city list, simultaneously utilizes the presentation of interface graphics, helps customer to quickly discover, locate the location of the clustering problem of vRBT video ringback tone, effectively reduces cost and improves work efficiency.

Description

Technical Field

[0001] This invention relates to a quality analysis method for vRBT video ringback tones, specifically a method for discovering video ringback tone clustering problems based on different dimensional indicators. Background Technology

[0002] vRBT video ringback tones are a key 5G service developed by telecom operators and have experienced rapid growth since their rollout. With the increasing number of users, effectively identifying issues affecting vRBT video ringback tones for timely mobile network optimization has become increasingly important. However, as vRBT video ringback tones are still in their early stages, experienced personnel are currently required to manually analyze and cluster issues using the VolTE video ringback tones method. This approach is inefficient and labor-intensive. Furthermore, VolTE metrics differ from vRBT metrics, lacking stage latency indicators and failing to fully reflect the clustering issues of vRBT video ringback tones. Summary of the Invention

[0003] To address the problems existing in the prior art, the purpose of this invention is to provide a technical solution for a method to discover video ringback tone clustering problems based on different dimensional indicators.

[0004] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized by including the following steps:

[0005] Step 1: Raw Data Acquisition: Using the indicator management module in the metadata orchestration tool, create indicator formulas and indicator names based on XDR. Use the basic table management module in the metadata orchestration tool to create the BKPI table, selecting the configured indicator names and common dimensions: user, city, cell, core network element, and terminal brand. Obtain latency and stage latency indicators under the user, city, cell, core network element, and terminal brand dimensions from the underlying XDR document CDR_IMS_CALL_DELAY. Obtain packet loss rate and MOS value indicators under the user, city, cell, core network element, and terminal brand dimensions from CDR_VOLTE_VOICE_QUALITY. Obtain connection indicators under the user, city, cell, core network element, and terminal brand dimensions from CDR_IMS_MO_CALL_LEG_SIP and CDR_IMS_MT_CALL_LEG_SIP. Orchestrate the process and schedule, selecting IMC-HDFS-BKPI. -15-minute granularity, no storage, based on the selected indicator formula, selected dimension, and scheduling, generate BKPI table. After packaging and installation into the environment, the IMC module automatically associates the indicator formula and dimension fields with the corresponding XDR every 15 minutes to obtain the corresponding fields of the indicator and dimension. According to time, user, city, community, core network element, terminal brand, latency indicator, stage latency indicator, packet loss rate indicator, MOS value indicator and connection indicator, it automatically generates BKPI table1 with statistics and output every 15 minutes and pushes the data to HDFS;

[0006] Step 2: Data Cleaning: Based on BKPI Table 1, first remove null values; based on the number of data samples and the distribution of data in BKPI Table 1, use the 3σ principle to identify outliers, and then use the capping method to replace data that exceed the range of (μ -3σ, μ+3σ).

[0007] Step 3: Clustering Threshold Statistics: Outliers in BKPI Table 1 can affect threshold analysis and cause deviations in the thresholds. Therefore, based on the cleaned BKPI Table 1 data from Step 2, the threshold for each indicator is calculated daily at 2:00 AM using the Otsu method.

[0008] Step 4: Count the number of affected users: Based on BKPI Table 1, according to the threshold of each indicator in Step 3, count the number of affected users at the cell level, terminal brand level, core network element level, and city level. If a user's vRBT video RTCP packet loss rate is greater than the vRBT video RTCP packet loss rate threshold output in Step 4, then that user is counted as a user affected by vRBT video RTCP packet loss rate. All users with vRBT video RTCP packet loss rates greater than the vRBT video RTCP packet loss rate threshold output in Step 4 are filtered, deduplicated, and accumulated; the result is the number of users affected by vRBT video RTCP packet loss rate. Similarly, if a user's vRBT video RTP packet loss rate is greater than the vRBT video RTP packet loss rate threshold output in Step 4, then that user is counted as a user affected by vRBT video RTP packet loss rate. All users with vRBT video RTP packet loss rates greater than the vRBT video RTP packet loss rate threshold output in Step 4 are filtered, deduplicated, and accumulated; the result is the number of users affected by vRBT video RTP packet loss rate. If a user's vRBT video IP MOS is less than the vRBT video IP MOS output in Step 4... If the MOS threshold is used, then the vRBT video IP MOS of the affected users is calculated. All users whose vRBT video IP MOS is less than the vRBT video IP MOS threshold output in step four are filtered, deduplicated, and accumulated. The result is the number of users affected by vRBT video IP MOS. The statistical rules for the number of users affected by other indicators are the same. Then, BKPI Table 2, BKPI Table 3, BKPI Table 4, and BKPI Table 5 are generated according to the dimensions of cell, terminal brand, core network, and city, and output to HDFS for use in step six.

[0009] Step 5: Calculate the overall score: Based on the BKPI table 1 after data cleaning in Step 2, calculate the overall score under the cell dimension, terminal brand dimension, core network element dimension, and city dimension, and output the following tables according to the corresponding rules: BKPI table 6: fields include time, cell, and overall score; BKPI table 7: fields include time, terminal brand, and overall score; BKPI table 8: fields include time, core network element, and overall score; BKPI table 9: fields include time, city, and overall score.

[0010] Step Six: Output Cluster SDR: Combine Steps One, Four, and Five according to the following rules, and output the cluster SDR according to the following conditions:

[0011] 1) Based on BKPI Table 2, filter out users whose cumulative impact value for a single cell exceeds the cluster cell user count threshold or cluster terminal user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster cell user count threshold can be dynamically adjusted to obtain the impact user count for time, cell, and cell. 2) Based on BKPI Table 3, filter out users whose cumulative impact value for a single terminal brand exceeds the cluster terminal user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster terminal user count threshold can also be dynamically adjusted to obtain the impact user count for time and terminal brand. 3) Based on BKPI Table 4, filter out users whose cumulative impact value for a single core network exceeds the cluster core network element user count threshold. The cumulative user count represents the total number of users affected by all indicators. The core network element user count threshold can also be dynamically adjusted to obtain the impact user count for time, core network, and core network. 4) Based on BKPI Table 5, filter out users whose cumulative impact value for a single city exceeds the cluster city user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster city user count threshold can also be dynamically adjusted to obtain the impact user count for time, city, and city.

[0012] 2) Based on BKPI Table 2, filter the cumulative number of affected users of a single cell / the cumulative number of affected users of all cells * 100 > the threshold for the percentage of affected users of that cell. The threshold for the percentage of affected users of a single cell can be dynamically adjusted to obtain the percentage of affected users for time, cell, and cell. Based on BKPI Table 3, filter the cumulative number of affected users of a single terminal brand / the cumulative number of affected users of all terminal brands * 100 > the threshold for the percentage of affected users of a terminal brand. The threshold for the percentage of affected users of a terminal brand can be dynamically adjusted to obtain the percentage of affected users for time, terminal brand, and terminal brand. The percentage of users affected by the core network is calculated based on the following: Based on BKPI Table 4, the cumulative impact value of a single core network / the cumulative impact value of all core networks * 100 > the threshold for the percentage of users affected by the core network. This threshold can be dynamically adjusted to obtain the percentage of users affected by the core network over time. Similarly, based on BKPI Table 5, the cumulative impact value of a single city / the cumulative impact value of all cities * 100 > the threshold for the percentage of users affected by the city. This threshold can also be dynamically adjusted to obtain the percentage of users affected by the city over time.

[0013] 3) Based on BKPI Table 6, filter out cells whose comprehensive score is less than the comprehensive score threshold of the cell. The comprehensive score threshold of the cell can be dynamically adjusted to obtain the comprehensive score of time, cell, and cell.

[0014] Output the cells that simultaneously meet conditions 1), 2), and 3). Also, left-associate the cells with cells in BKPI table 1 and output the corresponding indicator fields. Left-associate the cells with cells in BKPI table 2 and output the number of users affected by the cells. Left-associate the cells with BKPI table 6 and output the comprehensive score of the cells.

[0015] Output terminal brands that simultaneously meet conditions 1), 2), and 3). Also, left-associate terminal brands with terminal brands in BKPI table 1 and output the corresponding indicator fields. Left-associate terminal brands with terminal brands in BKPI table 3 and output the number of users affected by the terminal brand. Left-associate terminal brands with BKPI table 7 and output the comprehensive score of the terminal brand.

[0016] Output the core network that simultaneously satisfies conditions 1), 2), and 3). Also, left-associate the core network in BKPI table 1 and output the corresponding indicator fields. Left-associate the core network in BKPI table 4 and output the number of users affected by the core network. Left-associate the core network in BKPI table 8 and output the comprehensive score of the core network.

[0017] Output the cities that simultaneously meet conditions 1), 2), and 3). Also, left-associate the cities in BKPI table 1 with the corresponding indicator fields. Left-associate the cities in BKPI table 5 with the number of users affected by the cities. Left-associate the cities in BKPI table 9 with the comprehensive score of the cities.

[0018] Step 7: Presenting the Clustering Problem in the Interface: The front-end interface is developed using Javascript and CSS, including bar charts showing the number of clustered communities, clustered terminals, clustered core network elements, and clustered cities; tables listing the clustered communities, clustered terminals, clustered core network elements, and clustered cities; and the page queries the database for SDR1 and SDR2 via middleware. SDR3 and SDR4 respectively retrieve the number of clustered cells, clustered terminals, clustered core network elements, and clustered cities, and populate them into bar charts and lists of clustered cells, terminals, core network elements, and cities. The number of clustered cells, terminals, core network elements, and cities is obtained by filtering and deduplicating the cells, terminals, core network elements, and cities in the SDR1, SDR2, SDR3, and SDR4 tables for the "Whether clustered" field, terminal brand, core network, and city, respectively. The cluster list table presents all field data from the SDR1, SDR2, SDR3, and SDR4 tables where the "Whether clustered" field is "Yes". The graphical presentation helps customers quickly and intuitively locate vRBT video ringback tone clustering problems and helps them solve vRBT video ringback tone clustering issues.

[0019] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized by the 3σ principle: the probability of a value distributed in (μ-σ, μ+σ) is 0.6827, the probability of a value distributed in (μ-2σ, μ+2σ) is 0.9545, and the probability of a value distributed in (μ-3σ, μ+3σ) is 0.9973, where μ is the mean and σ is the standard deviation. It is generally believed that the values ​​of data Y are almost entirely concentrated in the (μ-3σ, μ+3σ) interval, and the probability of exceeding the (μ-3σ, μ+3σ) range is less than 0.3%. These data exceeding this range are considered outliers.

[0020] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized in that the capping method replaces the data to the left of (μ-3σ) with the value of -3σ and (μ+3σ) with the value of +3σ.

[0021] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized by the following method for calculating the threshold in step three: The indicator values ​​are plotted, with the Y-axis representing the indicator value and the X-axis representing the number of rows. An arbitrary threshold t is selected to divide the image into two parts: foreground and background. The proportion of foreground points in the image is W0, with a mean of U0, and the proportion of background points in the image is W1, with a mean of U1. The overall mean of the image is U = W0 * U0 + W1 * U1. An objective function g(t) = W0 * (U0 - U)^2 + W1 * (U1 - U)^2 is established, where g(t) is the inter-class variance when the segmentation threshold is t. The method iterates from the minimum to the maximum value of the indicators, and the t corresponding to the maximum value of g(t) is used as the threshold in step four. The thresholds of each indicator are output to memory for use in step four.

[0022] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized by the following cell comprehensive score calculation rules: Based on BKPI Table 1, each indicator of the cell is sorted by time: MOS value and call success rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator of the cell. The percentage of each indicator is obtained by dividing the total number of records of the cell and users. The percentage of each indicator is multiplied by 100 and then multiplied by the weight value of the corresponding indicator. Finally, all indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each cell.

[0023] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized by the following terminal brand score calculation rules: Based on BKPI Table 1, each indicator of the terminal brand is sorted by time: MOS value and call connection rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator of the terminal brand. The percentage of each indicator is obtained by dividing the total number of records of the cell and users. The percentage of each indicator is multiplied by 100 and then multiplied by the weight value of the corresponding indicator. Finally, all indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each terminal brand.

[0024] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized by the following core network element score calculation rules: Based on BKPI Table 1, each indicator of the core network element is sorted by time: MOS value and call success rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator of the core network element. The percentage of each indicator is obtained by dividing the total number of records of the core network element and users. The percentage of each indicator is multiplied by 100 and then multiplied by the weight value of the corresponding indicator. Finally, all indicators of the core network element are summed and multiplied by the total weight to obtain the comprehensive score of each cell.

[0025] The method for discovering video ringback tone clustering problems based on different dimensional indicators is characterized by the following city-level score calculation rules: Based on BKPI Table 1, each indicator of the city is sorted by time: MOS value and call connection rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator of the city. The percentage of each indicator is obtained by dividing the total number of records of the city and users. The percentage of each indicator is multiplied by 100 and then multiplied by the corresponding weight value of the indicator. The weight value can be dynamically adjusted. Finally, all indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each city.

[0026] Currently, vRBT video ringback tones are still in their early stages, requiring experienced personnel to manually analyze and cluster data using the VolTE video ringback tone methodology. This approach is inefficient and labor-intensive. This solution, based on the client's actual business needs and referencing China Mobile's standards, utilizes big data technology to automatically calculate and output a list of clustered cells, terminal brands, core networks, and cities for vRBT video ringback tones. Simultaneously, a graphical interface helps clients quickly identify and locate clustering issues in vRBT video ringback tones, effectively reducing costs and improving work efficiency. Attached Figure Description

[0027] Figure 1 A schematic diagram of the 3σ principle;

[0028] Figure 2 This is a diagram illustrating the capping technique.

[0029] Figure 3 This is a schematic diagram of the Otsu method. Detailed Implementation

[0030] The present invention will be further described below with reference to the accompanying drawings:

[0031] Based on actual customer business needs, this invention constructs a performance evaluation system for ringback tone services, with key indicators including the following:

[0032]

[0033] This invention includes the following steps:

[0034] Step 1:

[0035] Raw data acquisition: Using the indicator management module in the IDE metadata orchestration tool, create indicator formulas based on XDR by selecting XDR fields and functions, and name the indicator. Use the basic table management module in the metadata orchestration tool to create the BKPI table, select the configured indicator name, and select common dimensions, including users, cities, communities, core network elements, and terminal brands. From the underlying XDR document CDR_IMS_CALL_DELAY, latency and stage latency metrics are obtained at the user, city, cell, core network element, and terminal brand levels. From CDR_VOLTE_VOICE_QUALITY, packet loss rate and MOS values ​​at the user, city, cell, core network element, and terminal brand levels are obtained. From CDR_IMS_MO_CALL_LEG_SIP and CDR_IMS_MT_CALL_LEG_SIP, connection metrics at the user, city, cell, core network element, and terminal brand levels are obtained. The process is orchestrated and scheduled, selecting IMC-HDFS-BKPI-15-minute granularity, without storing the data. Based on the selected metric formulas, dimensions, and scheduling, a BKPI table is generated, packaged, and installed into the environment. Every 15 minutes, the IMC module automatically associates the indicator formulas and dimension fields with the corresponding XDR, retrieves the corresponding fields of the indicators and dimensions, and automatically generates a BKPI table with statistics and output every 15 minutes according to time, user, city, community, core network element, terminal brand, latency indicator, stage latency indicator, packet loss rate indicator, MOS value indicator and connection indicator, and pushes the data to HDFS.

[0036]

[0037]

[0038] BKPI Table 1

[0039] Step Two:

[0040] Data cleaning: Based on BKPI Table 1, firstly, null values ​​are removed; considering that there are many data samples in BKPI Table 1 and the data distribution tends to be normal, the 3σ principle is used to identify outliers; then, the capping method is used to replace data that exceed the range of (μ-3σ,μ+3σ).

[0041] The 3σ rule states that the probability of a value falling within the range (μ-σ, μ+σ) is 0.6827, the probability of falling within the range (μ-2σ, μ+2σ) is 0.9545, and the probability of falling within the range (μ-3σ, μ+3σ) is 0.9973. Here, μ is the mean and σ is the standard deviation. Generally, it can be assumed that almost all values ​​of data Y are concentrated within the range (μ-3σ, μ+3σ); the probability of values ​​exceeding this range is less than 0.3%, and these outliers can be considered outliers.

[0042] Capping method:

[0043] Replace the data to the left of (μ-3σ) with the value of -3σ, and replace (μ+3σ) with the value of +3σ;

[0044] Step 3:

[0045] Clustering Threshold Statistics: Outliers in BKPI Table 1 can affect threshold analysis and cause bias. Therefore, based on the cleaned BKPI Table 1 data from step two, the Otsu method is used to calculate the threshold for each indicator at 2 AM every day. This is done once a day, and one threshold is output daily. The indicator values ​​are then displayed as follows: Figure 3 The image is plotted with the Y-axis representing the index value and the X-axis representing the number of rows. An arbitrary threshold t is chosen, derived from the index values ​​in BKPI. The image is then regularized and divided into two parts: foreground and background. The proportion of foreground points is W0, the mean of which is the proportion of points greater than threshold t, and U0 is the average index value of points greater than threshold t. The proportion of background points is W1, the mean of which is the proportion of points less than or equal to threshold t, and U1 is the average index value of points less than or equal to threshold t. The overall mean of the image is U = W0 * U0 + W1 * U1. An objective function is established: g(t) = W0 * (U0 - U)^2 + W1 * (U1 - U)^2, where g(t) is the inter-class variance when the segmentation threshold is t. The indexes are iterated from minimum to maximum. The t corresponding to the maximum g(t) is used as the threshold in step four. The thresholds for each index are output to memory for use in step four.

[0046] Step Four:

[0047] Counting the number of affected users: Based on BKPI Table 1, according to the threshold of each indicator in Step 3, the count is performed without distinguishing between dimensions. The count is done at the cell level, terminal brand level, core network element level, and city level. If a user's vRBT video RTCP packet loss rate is greater than the vRBT video RTCP packet loss rate threshold output in Step 4, then that user is counted as affected by vRBT video RTCP packet loss rate. All users with vRBT video RTCP packet loss rates greater than the vRBT video RTCP packet loss rate threshold output in Step 4 are filtered, deduplicated, and accumulated. The result is the number of users affected by vRBT video RTCP packet loss rate. Similarly, if a user's vRBT video RTP packet loss rate is greater than the vRBT video RTP packet loss rate threshold output in Step 4, then that user is counted as affected by vRBT video RTP packet loss rate. All users with vRBT video RTP packet loss rates greater than the vRBT video RTP packet loss rate threshold output in Step 4 are counted as affected by vRBT video RTP packet loss rate. The cumulative filtering and deduplication of users based on the video RTP packet loss rate threshold results in the number of users affected by the vRBT video RTP packet loss rate. If a user's vRBT video IPMOS is less than the vRBT video IPMOS threshold output in step four, then they are counted as users affected by the vRBT video IPMOS. The cumulative filtering and deduplication of all users whose vRBT video IPMOS is less than the vRBT video IPMOS threshold output in step four results in the number of users affected by the vRBT video IPMOS. The statistical rules for the number of users affected by other indicators are the same. Then, BKPI Table 2, BKPI Table 3, BKPI Table 4 and BKPI Table 5 are generated according to the dimensions of cell, terminal brand, core network and city, and output to HDFS for use in step six.

[0048]

[0049]

[0050] BKPI Table 2

[0051]

[0052]

[0053]

[0054] BKPI Table 3

[0055]

[0056]

[0057] BKPI Table 4

[0058]

[0059]

[0060] BKPI Table 5

[0061] Step 5:

[0062] Statistical composite score: Based on the BKPI table 1 after data cleaning in step two, the composite scores are calculated under the dimensions of community, terminal brand, core network element, and city.

[0063] The rules for calculating the overall score of a cell are as follows: Based on Table 1 of BKPI, each indicator of the cell is sorted by time, with MOS value and call success rate in ascending order, and latency and packet loss rate in descending order. The ranking of each indicator of the cell is obtained by dividing the total number of records of the cell and users to get the proportion of each indicator. The proportion of each indicator is multiplied by 100 and then multiplied by the corresponding weight value of the indicator. The weight can be dynamically adjusted. Finally, all the indicators of the cell are summed and multiplied by the total weight to get the overall score of each cell.

[0064] Terminal brand score calculation rules: Based on BKPI Table 1, each indicator of the terminal brand is sorted by time, with MOS value and call success rate in ascending order, and latency and packet loss rate in descending order, to obtain the ranking of each indicator of the terminal brand. The percentage of each indicator is obtained by dividing the total number of records of cells and users. The percentage of each indicator is multiplied by 100 and then multiplied by the corresponding weight value of the indicator. The weight value can be dynamically adjusted. Finally, all indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each terminal brand.

[0065] Core network element score calculation rules: Based on BKPI Table 1, each indicator of the core network element is sorted by time, with MOS value and call success rate in ascending order, and latency and packet loss rate in descending order, to obtain the ranking of each indicator of the core network element. The percentage of each indicator is obtained by dividing the total number of records of the core network element and users. The percentage of each indicator is multiplied by 100 and then multiplied by the corresponding weight value of the indicator. The weight value can be dynamically adjusted. Finally, all the indicators of the core network element are summed and multiplied by the total weight to obtain the comprehensive score of each cell.

[0066] City-level score calculation rules: Based on BKPI Table 1, each indicator of the city is sorted by time, with MOS value and call success rate in ascending order, and latency and packet loss rate in descending order, to obtain the ranking of each indicator of the city. The percentage of each indicator is obtained by dividing the total number of records of the city and users. The percentage of each indicator is multiplied by 100 and then multiplied by the corresponding weight value of the indicator. The weight value can be dynamically adjusted. Finally, all the indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each city.

[0067] Output BKPI Table 6 according to the corresponding rules, with fields including time, cell, and overall score. Output BKPI Table 7, with fields including time, terminal brand, and overall score. Output BKPI Table 8, with fields including time, core network element, and overall score. Output BKPI Table 9, with fields including time, city, and overall score.

[0068] time 1664586000 residential area SJYUH8351 Wanda B2 South Area-HLSA-1 Overall Score 93.21

[0069] BKPI Table 6

[0070] time 1664586000 Terminal Brands Huawei Overall Score 92.21

[0071] BKPI Table 7

[0072] time 1664586000 Core network elements SBC108BX Overall Score 90.11

[0073] BKPI Table 8

[0074]

[0075]

[0076] BKPI Table 9

[0077] Step Six:

[0078] Output SDR clustering: Merge steps one, four, and five according to the following rules, and output the SDR clustering based on the following conditions:

[0079] 1) Based on BKPI Table 2, filter out users whose cumulative impact value for a single cell exceeds the cluster cell user count threshold or cluster terminal user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster cell user count threshold can be dynamically adjusted to obtain the impact user count for time, cell, and cell. 2) Based on BKPI Table 3, filter out users whose cumulative impact value for a single terminal brand exceeds the cluster terminal user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster terminal user count threshold can also be dynamically adjusted to obtain the impact user count for time and terminal brand. 3) Based on BKPI Table 4, filter out users whose cumulative impact value for a single core network exceeds the cluster core network element user count threshold. The cumulative user count represents the total number of users affected by all indicators. The core network element user count threshold can also be dynamically adjusted to obtain the impact user count for time, core network, and core network. 4) Based on BKPI Table 5, filter out users whose cumulative impact value for a single city exceeds the cluster city user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster city user count threshold can also be dynamically adjusted to obtain the impact user count for time, city, and city. For example, in community 1: the cumulative number of affected users is 10, and the threshold for the number of users in the cluster is 8. Then output the community and the number of affected users at that time.

[0080] 2) Based on BKPI Table 2, filter the cumulative impact of a single cell / (cumulative impact of all cells) * 100 > the threshold for the percentage of impacted users of that cell. This threshold can be dynamically adjusted to obtain the percentage of impacted users for each time and cell. Based on BKPI Table 3, filter the cumulative impact of a single terminal brand / (cumulative impact of all terminal brands) * 100 > the threshold for the percentage of impacted users of that terminal brand. This threshold can be dynamically adjusted to obtain the percentage of impacted users for each time and terminal brand. Based on BKPI Table 4, filter the cumulative impact of a single core network / (cumulative impact of all core networks). The cumulative number of users * 100 > the threshold for the percentage of users affected by the core network. This threshold can be dynamically adjusted to obtain the percentage of users affected by the core network over time. Based on BKPI Table 5, the cumulative number of users affected by a single city / the cumulative number of users affected by all cities * 100 > the threshold for the percentage of users affected by the city. This threshold can also be dynamically adjusted to obtain the percentage of users affected by the city over time. For example, in cell 1: the cumulative number of users affected by all indicators is 10, the total number of users affected by all cells is 100, and the threshold for the percentage of users affected is 5%. Therefore, the output will show the cell and the percentage of users affected at that time.

[0081] 3) Based on BKPI Table 6, filter out cells whose comprehensive score is less than the comprehensive score threshold of the cell. The comprehensive score threshold of the cell can be dynamically adjusted to obtain the comprehensive score of time, cell, and cell.

[0082] Output the cells that simultaneously meet conditions 1), 2), and 3). Also, left-associate the cells with cells in BKPI table 1 and output the corresponding indicator fields. Left-associate the cells with cells in BKPI table 2 and output the number of users affected by the cells. Left-associate the cells with BKPI table 6 and output the cell's overall score. For example, if cell 1 has an overall score of 80.22 and an overall score threshold of 90, then output the cell and its overall score for that time period.

[0083] Output terminal brands that simultaneously satisfy conditions 1), 2), and 3). Also, left-associate terminal brands with terminal brands in BKPI table 1 and output the corresponding indicator fields. Left-associate terminal brands with terminal brands in BKPI table 3 and output the number of users affected by the terminal brand. Left-associate terminal brands with BKPI table 7 and output the comprehensive score of the terminal brand.

[0084] Output the core network that simultaneously satisfies conditions 1), 2), and 3). Also, left-associate the core network in BKPI table 1 and output the corresponding indicator fields. Left-associate the core network in BKPI table 4 and output the number of users affected by the core network. Left-associate the core network in BKPI table 8 and output the comprehensive score of the core network.

[0085] Output the cities that simultaneously meet conditions 1), 2), and 3). Also, left-associate the cities in BKPI table 1 with the corresponding indicator fields. Left-associate the cities in BKPI table 5 with the number of users affected by the cities. Left-associate the cities in BKPI table 9 with the comprehensive score of the cities.

[0086]

[0087]

[0088]

[0089] SDR1

[0090]

[0091]

[0092]

[0093]

[0094] SDR2

[0095]

[0096]

[0097]

[0098] SDR3

[0099]

[0100]

[0101]

[0102]

[0103] SDR4

[0104] Step Seven:

[0105] The interface presents the clustering problem: The front-end interface is developed using languages ​​such as Javascript and CSS, including bar charts showing the number of clustered communities, clustered terminals, clustered core network elements, and clustered cities; tables listing the clustered communities, clustered terminals, clustered core network elements, and clustered cities; and the page queries the database for SDR1, SDR2, and SDR3 via middleware. The system, along with SDR4, retrieves the number of clustered cells, terminals, core network elements, and cities, respectively, and populates them into bar charts and lists of clustered cells, terminals, core network elements, and cities. The final counts for clustered cells, terminals, core network elements, and cities are obtained by filtering and deduplicating cells, terminals, core network elements, and cities that are clustered in SDR1, SDR2, SDR3, and SDR4 tables, respectively. The clustering list table then presents all data from the SDR1, SDR2, SDR3, and SDR4 tables where the clustering field is set to "yes." This graphical presentation helps customers quickly and intuitively locate and resolve vRBT video ringback tone clustering issues.

[0106] Note:

[0107] CDR_IMS_CALL_DELAY: vRBT delay session document;

[0108] CDR_IMS_MO_CALL_LEG_SIP: vRBT originating call session record;

[0109] CDR_IMS_MT_CALL_LEG_SIP: vRBT final call session record;

[0110] CDR_VOLTE_VOICE_QUALITY: vRBT voice quality session document;

[0111] vRBT: 5G voice call;

[0112] MOS: A communications term, often used to measure the voice quality of a communication system.

[0113] 3σ principle: Raida criterion, the most commonly used method for identifying outliers in data;

[0114] The capping method: Replace records outside three standard deviations above and below the mean of a continuous variable with records outside three standard deviations above and below the mean.

[0115] Methods for calculating standard deviation;

[0116] Otsu's method: Otsu's method for finding the largest inter-class variance;

[0117] IMC: Computation Module;

[0118] BKPI: Basic Data Table;

[0119] SDR: A data table used by upper-layer applications to query business data;

[0120] RTP: Data Transmission Protocol;

[0121] RTCP: Control Protocol;

[0122] MOS: Mean Opinion Value for End-to-End Measurement of Speech Quality;

[0123] IP MOS: Mean opinion value of the measured voice quality in the range from the far end point to the acquisition interface;

[0124] Total weight: The sum of the weights of all indicators;

[0125] SBC, SCC SA, SCP SA, SCCAS, SCPAS: Core network elements;

[0126] ICS, GM, Mw, ICS: Signaling Interface;

[0127] VolTE: An end-to-end voice solution under all-IP conditions on 4G networks;

[0128] 200 OK (Update) message: This message indicates that the request has been successful.

[0129] PRACK message: An ACK (acknowledgment of receipt) message for a PR (provisional response) message;

[0130] Invite message: A request message initiated between core network elements;

[0131] HDFS: Stores databases;

[0132] XDR: Underlying document;

[0133] IDE: Metadata management tool;

[0134] Javascript and CSS: programming languages;

[0135] Time: All times in the table are UTC (Coordinated Universal Time), in seconds. For example, 1664586000 represents 2022-10-01 09:00:00.

[0136] SJYUH8351 Wanda B2 South District-HLSA-1: Name of the communication equipment cell.

Claims

1. A method for discovering video ringback tone clustering problems based on different dimensional indicators, characterized in that... Includes the following steps: Step 1: Raw Data Acquisition: Using the indicator management module in the metadata orchestration tool, create indicator formulas and indicator names based on the underlying XDR document. Use the basic table management module in the metadata orchestration tool to create the basic data table BKPI, selecting the configured indicator names and common dimensions: user, city, cell, core network element, and terminal brand. Obtain latency indicators and stage latency indicators under the user, city, cell, core network element, and terminal brand dimensions from the vRBT latency session document CDR_IMS_CALL_DELAY of the underlying XDR document. Obtain packet loss rate and MOS value indicators under the user, city, cell, core network element, and terminal brand dimensions from the vRBT voice quality session document CDR_VOLTE_VOICE_QUALITY. Obtain the vRBT originating call session document CDR_IMS_MO_CALL_L... The EG_SIP and vRBT final call session document CDR_IMS_MT_CALL_LEG_SIP obtain connection indicators under the dimensions of user, city, cell, core network element, and terminal brand. The process is orchestrated and scheduled. The calculation module IMC-storage database HDFS-basic data table BKPI-15-minute granularity is selected but not stored. Based on the indicator formula, selected dimension, and scheduling of the selected indicators, the BKPI table is generated. After being packaged and installed into the environment, the calculation module IMC automatically associates the indicator formula and dimension fields with the corresponding XDR every 15 minutes to obtain the corresponding fields of the indicator and dimension. According to time, user, city, cell, core network element, terminal brand, latency indicator, stage latency indicator, packet loss rate indicator, MOS value indicator, and connection indicator, the BKPI table1 is automatically generated and output every 15 minutes, and the data is pushed to HDFS. Step 2: Data Cleaning: Based on BKPI Table 1, first remove null values; based on the number of data samples and the distribution of data in BKPI Table 1, use the 3σ principle to identify outliers, and then use the capping method to replace data that exceed the range of (μ-3σ,μ+3σ), where μ is the mean and σ is the standard deviation. Step 3: Clustering Threshold Statistics: Outliers in BKPI Table 1 can affect threshold analysis and cause deviations in the thresholds. Therefore, based on the cleaned BKPI Table 1 data from Step 2, the threshold for each indicator is calculated daily at 2:00 AM using the Otsu method. Step 4: Count the number of affected users: Based on BKPI Table 1, according to the threshold of each indicator in Step 3, count the number of affected users at the cell level, terminal brand level, core network element level, and city level. If a user's 5G voice call vRBT video RTCP packet loss rate is greater than the vRBT video RTCP packet loss rate threshold output in Step 3, then the user is counted as a user affected by vRBT video RTCP packet loss rate. All users with vRBT video RTCP packet loss rates greater than the vRBT video RTCP packet loss rate threshold output in Step 3 are filtered, deduplicated, and accumulated. The result is the number of users affected by vRBT video RTCP packet loss rate. If a user's vRBT video RTP packet loss rate is greater than the vRBT video RTP packet loss rate threshold output in Step 3, then the user is counted as a user affected by vRBT video RTP packet loss rate. All users with vRBT video RTP packet loss rates greater than the vRBT video RTP packet loss rate threshold output in Step 3 are filtered, deduplicated, and accumulated. The result is the number of users affected by vRBT video RTP packet loss rate. If a user's vRBT video IP MOS is less than the vRBT video IP MOS output in Step 3... If the MOS threshold is used, then the vRBT video IP MOS is considered as the number of users affected. All users whose vRBT video IP MOS is less than the vRBT video IP MOS threshold output in step three are filtered, deduplicated, and accumulated. The result is the number of users affected by the vRBT video IP MOS. The other indicators affect the user count statistics in the same way. Then, BKPI Table 2, BKPI Table 3, BKPI Table 4 and BKPI Table 5 are generated according to the dimensions of community, terminal brand, core network and city, respectively, and output to HDFS for use in step six. Step 5: Calculate the overall score: Based on the BKPI table 1 after data cleaning in Step 2, calculate the overall score under the cell dimension, terminal brand dimension, core network element dimension, and city dimension, and output the following tables according to the corresponding rules: BKPI table 6: fields include time, cell, and overall score; BKPI table 7: fields include time, terminal brand, and overall score; BKPI table 8: fields include time, core network element, and overall score; BKPI table 9: fields include time, city, and overall score. Step Six: Output Cluster SDR: Combine Steps One, Four, and Five according to the following rules, and output the cluster SDR according to the following conditions: 1) Based on BKPI Table 2, filter out users whose cumulative impact value for a single cell is greater than the cluster cell user count threshold or the cluster terminal user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster cell user count threshold can be dynamically adjusted to obtain the number of users affected by time, cell, and cell. 2) Based on BKPI Table 3, filter out users whose cumulative impact value for a single terminal brand is greater than the cluster terminal user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster terminal user count threshold can be dynamically adjusted to obtain the number of users affected by time, terminal brand, and terminal brand. 3) Based on BKPI Table 4, filter out users whose cumulative impact value for a single core network is greater than the cluster core network element user count threshold. The cumulative user count represents the total number of users affected by all indicators. The core network element user count threshold can be dynamically adjusted to obtain the number of users affected by time, core network, and core network. 4) Based on BKPI Table 5, filter out users whose cumulative impact value for a single city is greater than the cluster city user count threshold. The cumulative user count represents the total number of users affected by all indicators. The cluster city user count threshold can be dynamically adjusted to obtain the number of users affected by time, city, and city. 2) Based on BKPI Table 2, filter the cumulative impact of a single cell / (cumulative impact of all cells) * 100 > the cell's impact user percentage threshold. This cell's impact user percentage threshold can be dynamically adjusted to obtain the impact user percentage for time, cell, and cell. Based on BKPI Table 3, filter the cumulative impact of a single terminal brand / (cumulative impact of all terminal brands) * 100 > the terminal brand's impact user percentage threshold. This terminal brand's impact user percentage threshold can be dynamically adjusted to obtain the impact user percentage for time, terminal brand, and terminal brand. The percentage of users affected by the core network is calculated based on the following: Based on BKPI Table 4, the cumulative impact value of a single core network / the cumulative impact value of all core networks * 100 > the threshold for the percentage of users affected by the core network. This threshold can be dynamically adjusted to obtain the percentage of users affected by the core network over time. Similarly, based on BKPI Table 5, the cumulative impact value of a single city / the cumulative impact value of all cities * 100 > the threshold for the percentage of users affected by the city. This threshold can also be dynamically adjusted to obtain the percentage of users affected by the city over time. 3) Based on BKPI Table 6, filter out cells whose comprehensive score is less than the comprehensive score threshold of the cell. The comprehensive score threshold of the cell can be dynamically adjusted to obtain the comprehensive score of time, cell, and cell. Output the cells that simultaneously meet conditions 1), 2), and 3). Also, left-associate the cells with cells in BKPI table 1 and output the corresponding indicator fields. Left-associate the cells with cells in BKPI table 2 and output the number of users affected by the cells. Left-associate the cells with BKPI table 6 and output the comprehensive score of the cells. Output terminal brands that simultaneously meet conditions 1), 2), and 3). Also, left-associate terminal brands with terminal brands in BKPI table 1 and output the corresponding indicator fields. Left-associate terminal brands with terminal brands in BKPI table 3 and output the number of users affected by the terminal brand. Left-associate terminal brands with BKPI table 7 and output the comprehensive score of the terminal brand. Output the core network that simultaneously satisfies conditions 1), 2), and 3). Also, left-associate the core network in BKPI table 1 with the core network and output the corresponding indicator fields. Left-associate the core network in BKPI table 4 with the core network and output the number of users affected by the core network. Left-associate the core network in BKPI table 8 with the core network and output the comprehensive score of the core network. Output the cities that simultaneously meet conditions 1), 2), and 3). Also, left-associate the cities in BKPI table 1 with the corresponding indicator fields. Left-associate the cities in BKPI table 5 with the number of users affected by the cities. Left-associate the cities in BKPI table 9 with the comprehensive score of the cities. Step 7: Presenting the Clustering Problem in the Interface: The front-end interface is developed using Javascript and CSS, including bar charts showing the number of clustered cells, terminals, core network elements, and cities; and lists of clustered cells (SDR1, SDR2, SDR3, and SDR4); and lists of clustered terminals (SDR2, SDR3, and SDR4); the page uses middleware to query SDR1, SDR2, SDR3, and SDR4 from the database to obtain the number of clustered cells, terminals, core network elements, and cities, respectively, and populates these numbers into the corresponding lists. The system includes a bar chart showing the number of clustered cities, and lists of clustered communities, terminals, core network elements, and cities. The final counts for each cluster are obtained by filtering and deduplicating the communities, terminals, core network elements, and cities in tables SDR1, SDR2, SDR3, and SDR4 based on whether the clustering field is "yes". The clustering list table presents all data from tables SDR1, SDR2, SDR3, and SDR4 where the "yes" field indicates clustering. This graphical presentation helps customers quickly and intuitively locate and resolve vRBT video ringback tone clustering issues.

2. The method for discovering video ringback tone clustering problems based on different dimensional indicators according to claim 1, characterized in that... The 3σ principle states that the probability of a value falling within the range (μ-σ, μ+σ) is 0.6827, the probability of a value falling within the range (μ-2σ, μ+2σ) is 0.9545, and the probability of a value falling within the range (μ-3σ, μ+3σ) is 0.9973. Here, μ is the mean, σ is the standard deviation, and the values ​​of data Y are concentrated within the range of (μ-3σ, μ+3σ). The probability of values ​​falling outside the range is less than 0.3%, and these values ​​are considered outliers.

3. The method for discovering video ringback tone clustering problems based on different dimensional indicators according to claim 1, characterized in that... The capping method involves replacing the data to the left of (μ-3σ) with the value of -3σ, and replacing (μ+3σ) with the value of +3σ.

4. The method for discovering video ringback tone clustering problems based on different dimensional indicators according to claim 1, characterized in that... The threshold calculation method in step three is as follows: Plot the index values, with the index value on the Y-axis and the number of rows on the X-axis. Arbitrarily select a threshold t to divide the image into two parts: foreground and background. The proportion of foreground points in the image is W0, and the mean is U0. The proportion of background points in the image is W1, and the mean is U1. The mean of the overall image is U = W0 * U0 + W1 * U1. Establish the objective function g(t) = W0 * (U0 - U)^2 + W1 * (U1 - U)^2. g(t) is the inter-class variance when the segmentation threshold is t. Traverse from the minimum to the maximum index value. The t corresponding to the maximum g(t) is used as the threshold in step four. Output the threshold of each index to memory for use in step four.

5. The method for discovering video ringback tone clustering problems based on different dimensional indicators according to claim 1, characterized in that... The calculation rules for the comprehensive score of the cell are as follows: Based on BKPI Table 1, each indicator of the cell is sorted by time: MOS value and call success rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator of the cell. The percentage of each indicator is obtained by dividing the total number of records of the cell and users. The percentage of each indicator is multiplied by 100 and then multiplied by the weight value of the corresponding indicator. Finally, all the indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each cell.

6. The method for discovering video ringback tone clustering problems based on different dimensional indicators according to claim 1, characterized in that... The terminal brand score calculation rules are as follows: Based on BKPI Table 1, each indicator of the terminal brand is sorted by time: MOS value and call success rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator of the terminal brand. The percentage of each indicator is obtained by dividing the total number of records of the cell and users. The percentage of each indicator is multiplied by 100 and then multiplied by the weight value of the corresponding indicator. Finally, all indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each terminal brand.

7. The method for discovering video ringback tone clustering problems based on different dimensional indicators according to claim 1, characterized in that... The core network element score calculation rules are as follows: Based on BKPI Table 1, each indicator of the core network element is sorted by time: MOS value and call success rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator of the core network element. The percentage of each indicator is obtained by dividing the total number of records of the core network element and users. The percentage of each indicator is multiplied by 100 and then multiplied by the weight value of the corresponding indicator. Finally, all the indicators of the core network element are summed and multiplied by the total weight to obtain the comprehensive score of each cell.

8. The method for discovering video ringback tone clustering problems based on different dimensional indicators according to claim 1, characterized in that... The scoring rules for each city are as follows: Based on Table 1 of BKPI, each indicator for each city is sorted by time: MOS value and call success rate are sorted in ascending order, and latency and packet loss rate are sorted in descending order to obtain the ranking of each indicator for each city. The percentage of each indicator is obtained by dividing the total number of records of the city and users. The percentage of each indicator is multiplied by 100 and then multiplied by the corresponding weight value of the indicator. The weight value can be dynamically adjusted. Finally, all indicators of the cell are summed and multiplied by the total weight to obtain the comprehensive score of each city.

Images