A method for optimizing the Internet access perception experience based on user perception

Abstract

The present invention provides an optimization method for the Internet access perception experience based on user perception, belonging to the field of network communication technology, including: obtaining various perception indexes affecting the perception of mobile Internet users; obtaining the perception index values when the perception is not satisfied as the perception inflection point and penalty value; determining the scores corresponding to the perception index values according to the perception inflection point and penalty value; obtaining the index weights of various perception indexes affecting the user perception; using the scores corresponding to the perception index values and the index weights to calculate the overall perception score of the user in the Internet access service; determining the service proportion of each Internet access service of the user; wherein the service proportion is the ratio of the single service duration to the total network usage duration; multiplying and adding the service proportions of all services and the overall perception score in the Internet access service to obtain the user perception portrait model. This method can find users who are not satisfied with the mobile data service and improve the perception and satisfaction of mobile Internet users.

Description

Technical Field

[0001] The present invention belongs to the technical field of network communication, and particularly relates to a method for optimizing the Internet perception experience based on user perception. Background Art

[0002] After years of large-scale construction of China's 5G mobile communication network, the overall network scale has become very large, and 5G traffic has maintained rapid growth. However, the challenges faced by customer perception and customer awareness are increasing. This year, operators have clearly stated that they need to reasonably balance and coordinate the maintenance of competitive advantages, the guarantee of customer perception, and the assurance of investment benefits.

[0003] With the advent of 5G services, the bottlenecks of traditional KPI performance optimization analysis methods and service support capabilities have emerged. Users cannot evaluate the service level, support capabilities, and usage perception of operators from the perspective of experience. Currently, the network quality monitoring of operators is shifting from being centered on network metrics to being centered on user experience; the service support of operators is shifting from passive response to complaints to active care for users.

[0004] At the user perception level, traditional metrics cannot reflect user experience, and network monitoring cannot effectively detect perception problems, so perception problems cannot be accurately demarcated and located. Summary of the Invention

[0005] In order to overcome the deficiencies of the above-mentioned existing technologies, the present invention provides a method for optimizing the Internet perception experience based on user perception.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for optimizing the Internet perception experience based on user perception, comprising:

[0008] Obtaining various perception indicators that affect the perception of mobile Internet users;

[0009] Collecting historical data of various perception indicators, inputting the historical data into a clustering algorithm, and outputting the perception indicator values when perception dissatisfaction occurs as perception inflection points and penalty values; determining the scores corresponding to the perception indicator values according to the perception inflection points and penalty values;

[0010] Inputting the historical data into a self-organizing mapping algorithm, and outputting the indicator weights of various perception indicators that affect user perception; calculating the overall perception score of the user in the Internet service by using the scores corresponding to the perception indicator values and the indicator weights;

[0011] Determining the business proportion of each Internet service of the user; wherein the business proportion is the ratio of the duration of a single service to the total network usage duration;

[0012] Multiply the business proportion of all services by the overall perception score in the Internet service and then sum them up to obtain the user perception portrait model;

[0013] Use the user perception portrait model to score the user perception, find low-score users; and optimize the Internet perception experience of low-score users;

[0014] Collect the historical data of various perception indicators and input the historical data into the clustering algorithm, and the clustering algorithm is the Kmeans algorithm;

[0015] The perception inflection point is the perception value at which the perception starts to rise or fall, and the penalty value is the perception value when the perception is from 0 to 100 points.

[0016] Furthermore, the optimization of the Internet perception experience for low-score users is specifically to optimize the network of the community where the low-score users are located; including:

[0017] According to the calculated overall perception score of the user in the Internet service, find users with 60 points or below 50 points as low-score users;

[0018] Obtain the traffic and duration data of the community where the low-score users use the network; find the communities with traffic and duration higher than the threshold as the resident communities of the low-score users;

[0019] Score the network perception of the resident communities and optimize the network of the resident communities with low scores.

[0020] Furthermore, finding the community with the highest traffic and duration as the resident community of the low-score user includes:

[0021] Input the traffic and duration data of the community into the k++ clustering algorithm, and use the k++ clustering algorithm to output the communities with traffic and duration higher than the threshold as the resident communities of the low-score users.

[0022] Furthermore, the scoring of the network perception of the resident community has the calculation formula:

[0023]

[0024] where n is the average value of the perception scores of the users in the resident community.

[0025] Furthermore, it also includes:

[0026] Obtain the network B-domain data and O-domain data of the low-score users;

[0027] Associate the user's location with the location on the map according to the network B-domain data and O-domain data, and construct a user trajectory evaluation model for observing the resident campus of the low-score users from the map.

[0028] Furthermore, the user perception portrait model is as follows:

[0029]

[0030] Furthermore, it also includes:

[0031] Construct a user perception portrait model using the tensorflow framework based on the business proportion and the overall perception score in the Internet access service.

[0032] Furthermore, the perception metrics include:

[0033] Page response success rate, page response time, page display success rate, instant messaging success rate, instant messaging server-side rtt, instant messaging terminal-side rtt, instant messaging TCP connection establishment delay, video service pause frequency per minute, video pause delay ratio, video initial playback success rate, video initial caching delay, Game service response success rate, Game service average response delay, game lag rate per game.

[0034] An Internet access perception experience optimization method based on user perception provided by the present invention has the following beneficial effects:

[0035] The present invention first obtains various perception metrics that affect the perception of mobile Internet users, and then determines the perception score of the user in a single Internet access service by analyzing the values of each metric and the degree of influence of each metric on the perception. Secondly, according to the market share of different Internet access services, the total perception score model of the user, that is, the user perception portrait model, is obtained; this model considers the scenarios of each service when the user uses the network, can fully evaluate the Internet access perception of the user and output the perception score, and evaluate the service level, support ability, and usage perception of the operator from the perspective of user experience; so that the network service provider can accurately find the users who are dissatisfied with the service, and then optimize the Internet access perception experience of the users specifically, solving the problem in the prior art that the user perception problem analysis and the improvement of the user perception experience of the operator's mobile network products cannot be obtained. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention and its design, the drawings required for this embodiment will be briefly introduced below. The drawings in the following description are only partial embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0037] Figure 1 It is a schematic diagram for calculating the user perception score of the data service of the present invention;

[0038] Figure 2Schematic diagram of data processing and analysis process for dissatisfied users' resident communities in the present invention. Detailed implementation manners

[0039] In order to enable those skilled in the art to better understand the technical solutions of the present invention and be able to implement them, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and cannot be used to limit the protection scope of the present invention.

[0040] Embodiment:

[0041] The present invention provides an optimization method for Internet access perception experience based on user perception, specifically as Figure 1 shown, including: obtaining various perception indexes that affect the perception of mobile Internet users; using machine learning to determine the index weights of various perception indexes that affect user perception, the inflection points and penalty values of the perception indexes when the perception is dissatisfied, and determining the scores corresponding to the perception index values according to the perception inflection points and penalty values; calculating the overall perception score of the user in the Internet access service by using the scores corresponding to the perception index values and the index weights; determining the service proportion of each Internet access service of the user, where the service proportion is the ratio of the single service duration to the total network usage duration; constructing a user perception portrait model according to the service proportion and the overall perception score in the Internet access service, using the user perception portrait model to score the user perception, and finding low-score users; finding the resident communities of the low-score users, scoring the resident communities of the users, and finding the low-score resident communities for network optimization.

[0042] The following are the specific implementation details of the present invention:

[0043] 1.1 Overall idea

[0044] A method and device for judging the root cause of perception of dissatisfied users based on data services, including:

[0045] Based on the network B domain data (traffic analysis data, user numbers) and O domain data (user KQI, signaling data, cell engineering parameters), establish a user trajectory evaluation model based on the user dimension and the cell dimension, conduct correlation analysis on dissatisfied users and resident communities, and find the resident communities that affect network quality.

[0046] Analyze the perception of dissatisfied users. The specific process is as follows: ① Analyze and screen the user KQI perception indicators through the correlation algorithm to complete the establishment of the index model set; ② Combine the business duration and traffic ratio of different services used by users to determine the service weights. Combine the degree of influence of different service stages on resident users to form perception factors, realize the quantitative scoring of customer perception, and form a user perception portrait model based on perception; (User perception quantitative scoring) ③ Output the overall perception score of daily users and the perception scores of 4 types of data services, indicators such as traffic and duration; ④ Combine machine learning through multiple rounds of data accumulation, output the top five resident communities of the day, the perception scores of the resident communities where users are located daily and the perception scores of 4 types of services, solve the problems of communities with low scores on the network side, improve the network quality, and improve the customer perception.

[0047] 1.2 Introduction to the analysis process and analysis algorithm of resident communities

[0048] Based on the data analysis data of Network B domain (user number attribution dictionary table data) and O domain data (cell engineering parameters data, cell performance data, user signaling data), establish a user trajectory evaluation model based on the user dimension and network dimension, and conduct an association analysis of dissatisfied users and resident communities to evaluate the situation of users' resident communities.

[0049] Data processing and analysis process of dissatisfied users' resident communities:

[0050] First, obtain various data. Based on the two dimensions of users and the network, select more than 30 variables to establish the model index set. The threshold model and time of each variable can be adaptively adjusted according to the local network situation.

[0051] Secondly, perform data cleaning and data association. Rely on multiple rounds of machine learning to mine the input data, establish a user trajectory evaluation model and an analysis algorithm for resident communities, and conduct an association analysis of users and resident communities.

[0052] Thirdly, based on the long-term historical data and analysis algorithm, accurately identify users and resident communities, and label the user's resident communities (workplace, place of residence).

[0053] Introduction to the analysis algorithm of resident communities:

[0054] ① Analyze and screen the indicators related to users and resident communities through the correlation algorithm to complete the establishment of the model index set;

[0055] ② Combine data cleaning and data association, rely on multiple rounds of machine learning to mine the input data, and establish a user trajectory evaluation model and an analysis algorithm for resident communities.

[0056] ③ According to the analysis algorithm, conduct a correlation analysis of users and resident communities.

[0057] ④ Combine long-term historical data and analysis algorithms to accurately identify users and their resident communities.

[0058] The specific algorithm model and algorithm steps are as follows:

[0059] The method for k-means to converge to local extrema is the k++ clustering algorithm. k-means++ initializes the cluster centers by making the centers between clusters far away from each other. In this case, the k++ clustering algorithm is used for model training of resident communities.

[0060] Specific algorithm steps:

[0061] The method for k-means to converge to local extrema is the k++ clustering algorithm. k-means++ initializes the cluster centers by making the centers between clusters far away from each other. In this case, the k++ clustering algorithm is used for model training of resident communities.

[0062] Specific algorithm steps:

[0063] 1) Randomly select a sample data as the first cluster center C1;

[0064] 2) Calculate the minimum distance from each sample x i to the cluster center C j ;

[0065]

[0066] 3) Select the sample point with the maximum distance as the cluster center;

[0067] 4) Repeat steps (2) and (3) until the number of clusters k is reached;

[0068] 5) Use these k cluster centers as the initialized cluster centers to run the k-means algorithm;

[0069] Through model training of 50,000 users, the k++ clustering algorithm that conforms to the resident communities of users' workplaces and residences is obtained.

[0070] Calculation method for users' resident communities:

[0071] 1. Calculate the msisdn, cell_id, traffic, and duration data from 9:00 to 12:00 and from 22:00 to 5:00 every day.

[0072] 2. Aggregate the data of the previous month on the first day of each week:

[0073] Aggregate the msisdn, cell, and traffic from 9:00 to 12:00 to obtain the top 5 resident communities with the highest traffic at the workplace.

[0074] 3. Aggregate by msisdn, cell, and duration from 9 to 12 o'clock to obtain the top 5 frequently visited cells for work duration.

[0075] 4. Aggregate by msisdn, cell, and traffic from 22:00 to 5:00 am to obtain the top 5 frequently visited cells for residential traffic.

[0076] 5. Aggregate by msisdn, cell, and duration from 22:00 to 5:00 am to obtain the top 5 frequently visited cells for residential duration.

[0077] Algorithm list:

[0078]

[0079] 1.3 Introduction to the construction of the user perception portrait model

[0080] Taking the customer's perception characteristics as the research object, through three major stages of data acquisition, modeling research, and portrait establishment, including eight specific tasks, the quantitative scoring of customer perception is realized, and finally a user perception portrait model based on the cell is formed.

[0081] The overall work is divided into three major stages and seven specific tasks:

[0082] Stage 1, data acquisition: ① Use various available data sources to obtain and preprocess the input data. Obtain the original data of unperceived users through research; ② Ensure the accuracy of the research input data through data cleaning; ③ Also consider other dimensions, such as user-specific data, application-related content, and even data from sources other than the operator (such as planned events, weather forecasts, etc.).

[0083] Stage 2, modeling research: ① Analyze and screen various perception indicators related to user perception through the algorithm of frequently visited cells to complete the establishment of the model index set. Taking the user's frequently visited cell as the research object, through the specific tasks of data acquisition, modeling research, and portrait establishment, the quantitative scoring of customer perception is realized, and finally a user perception portrait based on the perception degree is formed, and finally it focuses on the analysis of the problem cells (frequently visited cells) of the user.

[0084] Combined with machine learning, through multiple rounds of data accumulation, establish the business weight by the proportion of the business duration of the user in different cells, and output the accurate frequently visited cell of each user.

[0085] Stage 3: Portrait establishment: Form a complete set of quantifiable calculation models through algorithm standardization; finally complete the portrait establishment of the user's frequently visited cell through model calibration.

[0086] In the discovery stage of artificial intelligence, machine learning is relied on to mine input data and extract relevant knowledge models at different levels: the cell level (including conditional features based on each cell), the cell cluster level (constructing features of cell groups according to the similarity of cells), and the user level (including features of conditions experienced by individual users). The overall goal of machine learning is to build computer models to adapt to and learn from their experiences.

[0087] Related specific machine learning functions include:

[0088] 1) Classification algorithm: It is the process of finding a model or function that describes and differentiates data classes or concepts. Then, the obtained model (i.e., the classifier) is used to determine the class to which an object belongs. An object is the entity to be classified, and it is usually represented by a tuple including a set of attribute values. The classification process assumes that the possible classes are predefined in advance. Then, the classifier model is usually obtained from a supervised learning algorithm that analyzes a set of training tuples associated with known classes.

[0089] The algorithm construction module for cells, which is used to build a multiple linear regression model algorithm using the tensorflow framework during the algorithm construction process, to obtain the influence weights of various KPI indicators on the resident cells, and at the same time select and compare multiple machine learning algorithms, and select the CART decision tree algorithm with the best evaluation effect to train the ternary classification model, and identify each cell by calculating the information gain of each node.

[0090] 2) Clustering algorithm: Group a set of objects so that objects in the same cluster are similar to each other and different from objects in other clusters. The clustering mechanism does not rely on a set of predefined classes, but these classes are obtained through an unsupervised learning process.

[0091] 1.4 Introduction to the user perception portrait analysis model and perception score algorithm:

[0092] The user perception portrait analysis model based on user perception, and the analysis includes:

[0093] Analysis of resident cell indicators, which is used to consider the analysis of the network-side reasons for the current number portability, the analysis of the network indicators of the user's resident cells, the correlation analysis of the network side affecting the user perception, the solution of problem cells, and the improvement of network indicators;

[0094] The customer perception portrait unit, which takes the perception characteristics of non-perceived customers as the research object, and realizes the quantitative scoring of customer perception through 4 stages and 9 specific tasks of data acquisition, modeling research, and portrait establishment, and finally forms a user perception portrait based on perception.

[0095] The process of establishing the customer perception portrait:

[0096] Stage 1 Data Acquisition: A total of 15,000 perceived user data have been cleaned. Analysis shows that the main factor affecting user perception on the network side is the perception of network usage, accounting for 39%. Internet problems are the main issues.

[0097] Stage 2 Modeling Research: By studying the index set, weights, factors, and thresholds, the foundation for the perception quantization algorithm is laid.

[0098] Initial Index Set Unit: Through the correlation analysis of non-perceived user data using signaling data, 24 related service perception indicators for non-perceived users are screened, and machine learning is carried out. Finally, 15 indicators are determined as follows:

[0099] Table 1 Perception Degree Index

[0100]

[0101] Threshold Unit: The initial threshold of the portrait is obtained by taking the signaling data of surveyed non-perceived users. Through machine learning operations, the values of each sub-item index of all non-perceived users are obtained, and are initially set as the thresholds of the perception degree portrait model (full score penalty value and 0 score penalty value).

[0102] Perception Factor Unit: The perception impact factor determines the weights of each business sub-item index according to the specific problem phenomenon descriptions provided by non-perceived users, combined with the degree of impact on user perception at different stages of the business, and the probability of problem occurrence. The self-organizing mapping algorithm is used to obtain the index weights of each perception degree index.

[0103] The self-organizing mapping (SOM) algorithm is an unsupervised neural network model. It finds the weights in a set of data through self-organizing learning of the input data. The SOM algorithm classifies and calculates the weights of the data by clustering the input data on a two-dimensional grid.

[0104] Table 2 Penalty Values of Each Perception Degree Index

[0105]

[0106] The following are the steps for the SOM algorithm to find the weights of a set of data:

[0107] Initialization: First, select a suitable grid size and initialize each node in the grid as an independent neuron. Each neuron has a weight vector whose size is the same as the size of the input data.

[0108] Input Data: Then, provide the input data to the SOM algorithm. These data can be in any form, such as numbers, text, images, etc.

[0109] Calculate the distance: For each input data, calculate its distance from each neuron. Different methods can be used to calculate the distance, such as Euclidean distance or Manhattan distance.

[0110] Select the best matching unit: After calculating the distance, find the neuron that is closest to the input data, i.e., the best matching unit (BMU).

[0111] Update the weights: Then, compare the input data with the weight vector of the BMU and update the weight vector of the BMU according to the difference between them. The update method uses a strategy called "Winner-Take-All".

[0112] Repeat the steps: Repeat providing the input data to the SOM algorithm until all the input data has been processed.

[0113] Output the results: Finally, the SOM algorithm generates a two-dimensional grid, where each node has a weight vector. These weight vectors can be used to represent the patterns in the input data and can be used to classify or cluster the input data.

[0114] Table 3 Index weights of each perception index

[0115]

[0116] Business weight unit: It refers to the ratio of the duration of a single business to the total duration, that is, the durations of user businesses are aggregated, and the ratio of the duration of a single business to the total duration is the business weight K of the user; it determines the influence degree of each business on the user.

[0117] Table 4 Weights of each Internet service

[0118]

[0119] Phase 3: Portrait establishment unit: Considering comprehensively the influence of business weight, business indicators, and business quality at different time periods on perception, a comprehensive scoring system of hourly aggregation, perception stratification, and business weight is formulated to quantify the user perception degree.

[0120] Scoring algorithm unit:

[0121]

[0122] Specifically, the scoring algorithm unit of the user perception degree further includes:

[0123] Perception factor analysis module: Based on learning from more than 30 data service indicators in the initial indicator set using big data algorithms, 15 key customer perception indicators are determined. These data service perception factors have the strongest correlation with the customer perception degree.

[0124] Image construction algorithm module: During the algorithm construction process, use the tensorflow framework to construct a multiple linear regression model algorithm to obtain the influence weights of each perceived KQI. At the same time, select and compare multiple machine learning algorithms, and choose the CART decision tree algorithm with the best evaluation effect to train the ternary classification model, and identify the perceived inflection points of each KQI index by calculating the information gain of each node.

[0125] Image calibration and optimization module: To prevent the model from overfitting, introduce the empirical perception constant Δx as a reference to effectively improve the robustness of the model. At the same time, introduce non-survey data to perform two-way iterative calibration on the model to improve the evaluation accuracy and universality of the model. Calibrate the threshold and algorithm, and finally maximize the matching between the model output result and the user's perception, so as to continuously improve the accuracy of the perception image.

[0126] Accumulate the survey data of previous times, continuously correct the model variables, perform two-way iteration and dynamically adjust the perception threshold to realize the adaptive optimization of the model and improve the accuracy of the model;

[0127] Calculation of cell perception score:

[0128]

n is the average perceived score of the permanent users in the cell

[0129] The cell perception score is the average of the perceived scores of the permanent users in the cell.

[0130] Analysis and output of dissatisfied users of data services:

[0131] According to the above-mentioned aggregation analysis of various indicators for the permanent cells of the user's workplace and residence, find the users who are dissatisfied with the data service perception and potential users who may port their numbers.

[0132] Innovation points of the present invention:

[0133] 1. A judgment method based on the user's permanent cell. Combining machine learning of various data, through multiple rounds of data accumulation, the model output and algorithm output of the permanent cell analysis, analyze the top permanent cells of the user within 3 months, including the permanent cell_residence and the permanent cell_workplace.

[0134] 2. User perception portrait model and calculation method based on user perception. Based on the analysis of research data, relevant perception KQI big data, various attribute data and performance data are associated to conduct research on perception portraits and attribute portraits. The establishment of the index set realizes the establishment of the data source for the perception system scoring. At the same time, through algorithm research, factors W, weights K, and thresholds T are established, and a set of standard algorithms are solidified, and penalty value tuning and calibration are carried out to establish a user perception satisfaction scoring system. The quantitative scoring of customer perception is realized, and finally a user perception portrait model based on user perception is formed. Through continuous model correction, a calculation method for user perception score and cell perception score is finally output.

[0135] The above embodiments are only preferred specific embodiments of the present invention, and the protection scope of the present invention is not limited thereto. Any simple changes or equivalent replacements of technical solutions that can be obviously obtained by those skilled in the art within the technical scope disclosed by the present invention all belong to the protection scope of the present invention.

Claims

1. An optimization method for Internet access perception experience based on user perception, characterized in that, Including: Obtaining various perception indexes that affect the perception of mobile Internet users; Collecting historical data of various perception indexes, inputting the historical data into a clustering algorithm, and outputting the perception index values when the perception is dissatisfied as the perception inflection point and penalty value; determining the scores corresponding to the perception index values according to the perception inflection point and penalty value; Inputting the historical data into a self-organizing mapping algorithm, and outputting the index weights of various perception indexes affecting user perception; calculating the overall perception score of the user in the Internet access service by using the scores corresponding to the perception index values and the index weights; Determining the service proportion of each Internet access service of the user; wherein the service proportion is the ratio of the duration of a single service to the total network usage duration; Multiplying and adding the service proportions of all services by the overall perception score in the Internet access service to obtain a user perception portrait model; Using the user perception portrait model to score user perception and finding low-score users; And optimizing the Internet access perception experience of low-score users; The step of collecting historical data of various perception indexes and inputting the historical data into a clustering algorithm, and the clustering algorithm is the Kmeans algorithm; The perception inflection point is the perception value at which the perception starts to rise or fall, and the penalty value is the perception value when the perception is from 0 to 100 points.

2. The optimization method for Internet access perception experience based on user perception according to claim 1, characterized in that, The step of optimizing the Internet access perception experience of low-score users is specifically to optimize the network of the cell where the low-score users are located; including: According to the calculated overall perception score of the user in the Internet access service, finding users with a score of 60 or below 50 as low-score users; Obtaining the traffic and duration data of the cell where the low-score users use the network; finding the cells with traffic and duration higher than the threshold as the resident cells of the low-score users; Scoring the network perception of the resident cells and optimizing the network of the resident cells with low scores.

3. The optimization method for Internet access perception experience based on user perception according to claim 2, characterized in that, The step of finding the cell with the highest traffic and duration as the resident cell of the low-score user includes: Inputting the traffic and duration data of the cell into the k++ clustering algorithm, and using the k++ clustering algorithm to output the cells with traffic and duration higher than the threshold as the resident cells of the low-score users.

4. The optimization method for Internet access perception experience based on user perception according to claim 2, characterized in that, The formula for scoring the network perception of the resident cells is: where n is the average value of the perception scores of the users in the resident cell.

5. The optimization method for Internet access perception experience based on user perception according to claim 2, characterized in that, It also includes: Obtaining the network B-domain data and O-domain data of low-score users; Associating the area where the user is located with the position on the map according to the network B-domain data and O-domain data, and constructing a user trajectory evaluation model for observing the resident cells of low-score users from the map.

6. The optimization method for Internet access perception experience based on user perception according to claim 4, characterized in that, It also includes: Constructing a user perception portrait model by using the tensorflow framework according to the service proportion and the overall perception score in the Internet access service.

7. The optimization method for Internet access perception experience based on user perception according to claim 1, characterized in that, The perception indexes include: Page response success rate, page response duration, page display success rate, instant messaging success rate, instant messaging server-side rtt, instant messaging terminal-side rtt, instant messaging TCP connection establishment delay, video service pause frequency per minute, video pause delay ratio, video initial playback success rate, video initial cache delay, Game service response success rate, Game service response average delay, game card rate.

Images