A data-driven quality difference root cause reasoning method and system

Abstract

The embodiment of the application can more accurately quantify and evaluate the real network experience of users under different applications by collecting digital environment data in real time and combining a user network experience perception model based on machine learning, and can realize active perception of network quality prior to user reporting of faults, not only depending on network layer indexes; can accurately locate the root cause of poor quality according to a root cause knowledge base and evidence chain reasoning, and can improve fault troubleshooting efficiency and reduce manual dependence; by quickly and accurately locating the root cause of poor quality, operators and service providers can more effectively perform network optimization and fault processing, thereby improving operation and maintenance efficiency, reducing user complaint rate, and realizing coordinated leap of operation and maintenance efficiency and satisfaction.

Description

Technical Field

[0001] This invention relates to the field of network technology, and in particular to a data-driven method and system for inferring the root causes of poor quality. Background Technology

[0002] With the acceleration of digital transformation and the widespread adoption of smart home devices, users' demands for the quality of their home network experience are increasing. Poor network quality has become a key factor affecting operator service satisfaction. Traditional network operation and maintenance mainly relies on monitoring device-level performance indicators, lacking the ability to quantitatively perceive the actual user experience, making it difficult to meet the current demand for network service quality assurance centered on user experience.

[0003] In home broadband scenarios, poor quality can be caused by a variety of factors, including gateway performance, intranet environment, internet quality, terminal capabilities, and application services. Existing technologies typically rely on manual on-site inspections or basic tools like Ping / Traceroute for troubleshooting. Their data collection focuses on network layer transmission metrics, which are insufficiently covered for key dimensions affecting user experience, such as terminal status, wireless environment, and application interaction.

[0004] Existing methods cannot detect changes in user network experience in real time, and can only respond passively after users report faults. They also suffer from vague root cause identification due to limited data dimensions and analytical methods, and rely heavily on human experience, resulting in low troubleshooting efficiency. Furthermore, they lack a systematic data-driven reasoning mechanism, making it impossible to achieve accurate automated root cause attribution based on multi-source evidence chains, and thus failing to support the needs of operators for proactive operation and maintenance. Summary of the Invention

[0005] In view of the above problems, a data-driven method and system for inferring the root causes of poor quality is proposed to overcome or at least partially solve these problems. Specifically:

[0006] A data-driven method for inferring the root causes of poor product quality includes:

[0007] Real-time collection of digital environment data; inputting the collected digital environment data into the user network experience perception model; calculating and outputting the user's experience score at the current time point.

[0008] If the experience score is lower than the preset threshold, associate the digital environment data contained within the preset time period before and after the current time point and form a poor quality data snapshot; based on the failure scenarios in the root cause knowledge base, search for and construct the evidence chain related to the failure scenarios from the poor quality data snapshot; generate several poor quality root cause hypotheses based on the evidence chain, and evaluate the confidence level of each poor quality root cause hypothesis;

[0009] Output the hypothesis with the highest confidence level as the primary root cause of poor quality, along with the chain of evidence for the primary root cause, information on poor quality, and solutions.

[0010] Optional digital environment data includes: network quality data, device status data, WiFi environment data, port rate data, and connected device data.

[0011] Optionally, real-time digital environment data is collected and input into the user network experience perception model to calculate and output the user's experience score at the current time, including:

[0012] The probe plugin built into the home gateway continuously collects digital environment data and uploads the collected data to the cloud data processing and analysis platform according to a predetermined cycle.

[0013] The user network experience perception model, located on the cloud-based data processing and analysis platform, receives the uploaded data and calculates and outputs the user's experience score at the current time point based on machine learning algorithms.

[0014] Optionally, a poor-quality data snapshot is a snapshot of the data that includes time series data.

[0015] Optionally, based on the failure scenarios in the root cause knowledge base, a chain of evidence related to the failure scenarios is searched and constructed from the poor-quality data snapshots, including:

[0016] Extract features related to the failure scenarios defined in the root cause knowledge base from the poor quality data snapshot; search for evidence from the relevant features and construct an evidence chain based on the preset root cause scenarios and root cause judgment conditions.

[0017] Optionally, the method also includes: normalizing the extracted features related to the fault scenarios defined in the root cause knowledge base.

[0018] Optionally, several root cause hypotheses of poor quality are generated based on the chain of evidence, and the confidence level of each root cause hypothesis is evaluated, including:

[0019] The chain of evidence is input into the decision tree algorithm, which generates several poor-quality root cause hypotheses. At the same time, the decision tree algorithm is used to evaluate the fit between the chain of evidence and each poor-quality root cause hypothesis, and the fit is output as the confidence level of the poor-quality root cause hypothesis.

[0020] Optionally, the method further includes: outputting the second-highest confidence root cause hypothesis of the quality problem as the secondary root cause, along with the evidence chain of the secondary root cause, quality problem information, and solutions; wherein, the quality problem information includes the time when the quality problem occurred, the device that occurred, the application that occurred, the point of failure that occurred, and the specific cause type.

[0021] Optionally, the method also includes: addressing the quality issues based on the solution and feeding the solution results back to the user network experience perception model, the root cause knowledge base, and the decision tree algorithm model.

[0022] A data-driven system for reasoning about the root causes of poor product quality, comprising:

[0023] The data receiving and preprocessing module is used to collect digital environment data in real time, input the collected digital environment data into the user network experience perception model, calculate and output the user's experience score at the current time point;

[0024] The root cause reasoning engine module is used to associate digital environment data contained within a preset time period before and after the current time point when the experience score is lower than a preset threshold, and form a poor quality data snapshot; based on the failure scenarios in the root cause knowledge base, it searches for and constructs evidence chains related to the failure scenarios from the poor quality data snapshot; generates several poor quality root cause hypotheses based on the evidence chains, and evaluates the confidence level of each poor quality root cause hypothesis based on the completeness of the evidence chains;

[0025] The output module is used to output the hypothesis of the poor quality root cause with the highest confidence as the main root cause, along with the evidence chain of the main root cause, information on the poor quality, and solutions.

[0026] This invention, through real-time collection of digital environment data and combined with a machine learning-based user network experience perception model, can more accurately quantify and evaluate users' real network experience under different applications, rather than relying solely on network layer indicators. This enables proactive perception of network quality before user reports of faults. Based on a root cause knowledge base and evidence chain reasoning, it accurately locates the root causes of poor network quality, improving troubleshooting efficiency and reducing reliance on manual intervention. By quickly and accurately locating the root causes of poor network quality, it can help operators and service providers optimize networks and handle faults more effectively, thereby improving operational efficiency, reducing user complaint rates, and achieving a synergistic leap in operational efficiency and user satisfaction. Attached Figure Description

[0027] To more clearly illustrate the technical solution of the present invention, the accompanying drawings used in the description of the present invention will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of a data-driven method for reasoning about the root causes of poor quality provided in an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of poor-quality data snapshots and evidence chain construction provided in an embodiment of the present invention. Detailed Implementation

[0030] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0031] Reference Figure 1 The diagram illustrates a data-driven method for inferring the root causes of poor product quality, as provided in an embodiment of the present invention. Specifically, it may include the following steps:

[0032] Step 101: Collect digital environment data in real time, input the collected digital environment data into the user network experience perception model, calculate and output the user's experience score at the current time point.

[0033] In an embodiment of the present invention, a probe plug-in built into a home gateway can be used to continuously collect digital environment data and upload the collected data to a cloud data processing and analysis platform according to a predetermined cycle; the user network experience perception model located on the cloud data processing and analysis platform receives the uploaded data and calculates and outputs the user's experience score at the current time point based on machine learning algorithms.

[0034] The digital environment data may include: network quality data, device status data, WiFi environment data, port speed data, and connected device data.

[0035] Specifically, the following data can be collected using a probe plugin deployed on the home gateway:

[0036] Network quality data: For each data stream, data such as packet loss count, average latency (distinguishing between LAN-side network latency, LAN-side service latency, WAN-side network latency, and WAN-side service latency), HTTP first packet latency, and DNS latency are collected.

[0037] Device status data: Gateway CPU utilization, memory utilization, runtime, number of connections, number of connected terminals, optical module temperature and power, etc.;

[0038] WiFi environment data: SSID, BSSID, channel, signal strength, frequency band, and nearby WiFi interference, etc.

[0039] Port rate data: Uplink and downlink speeds and traffic of WAN and LAN ports (including WiFi ports);

[0040] Data of connected devices: terminal MAC address, IP address, negotiation rate, signal strength, access type (wired / wireless), access port (2.4G / 5G), etc.

[0041] After data collection is completed, the user network experience perception model module on the cloud platform receives the uploaded data. This model uses machine learning algorithms to parse streaming data from protocols such as DNS, HTTP, TCP, and UDP in real time. It constructs a single-stream experience scoring matrix by weighting latency deviation, packet loss rate threshold exceedances, and connection success rate. Then, it uses a time-sliding window aggregation algorithm to generate user experience scores for both the terminal and application dimensions.

[0042] For example, three key features can be extracted from each data stream: latency deviation, number of packet loss rate threshold exceedances, and connection success rate. These features are then linearly weighted using pre-trained random forest weight coefficients to generate a single-stream experience score matrix with stream ID as the row and timestamp as the column. Then, a time-sliding window aggregation algorithm is used with a 15-minute window length and a 5-minute sliding step to perform multi-protocol normalization on the matrix, such as converting the DNS latency fluctuation coefficient into an HTTP equivalent value. Finally, a terminal-dimensional user experience score is generated by aggregating the average score of all streams within the window based on the terminal MAC address, and an application-dimensional user experience score is generated by aggregating the scores based on the application ID.

[0043] It is understandable that when the user experience score at the terminal level or the user experience score at the application level is lower than a preset threshold, it can be determined that a poor user network experience event has occurred.

[0044] Step 102: When the experience score is lower than the preset threshold, associate the digital environment data contained within the preset time period before and after the current time point and form a poor quality data snapshot; based on the fault scenarios in the root cause knowledge base, search for and construct the evidence chain related to the fault scenarios from the poor quality data snapshot; generate several poor quality root cause hypotheses based on the evidence chain and evaluate the confidence level of each poor quality root cause hypothesis.

[0045] In an embodiment of the present invention, when the experience score is lower than a preset threshold, i.e., when a poor quality event occurs, the root cause reasoning engine module will be activated. It will associate digital environment data contained within a preset time period before and after the occurrence of the poor quality event to form a poor quality data snapshot. This poor quality data snapshot can be a snapshot of data containing a time series. Then, features related to the fault scenario defined in the root cause knowledge base are extracted from the poor quality data snapshot; for example, to determine whether it is "high WiFi channel interference," features such as "LAN-side network latency" and "current channel utilization" need to be extracted. Then, based on the preset root cause scenario and root cause judgment conditions, evidence is searched from the relevant features and an evidence chain is constructed.

[0046] However, it should be noted that differences in the units of different features may lead to biases in the subsequent chain of evidence evaluation. Therefore, the extracted features related to the fault scenarios defined in the root cause knowledge base can be normalized.

[0047] After normalization, the inference engine collects evidence from data snapshots based on preset root cause scenarios (gateway segment, internet segment, Wi-Fi segment, device segment) and root cause judgment conditions, and constructs an evidence chain from multiple dimensions. For example, judging "Wi-Fi channel interference" may require evidence such as "LAN-side network latency > baseline value" and "current channel load is high." Root cause judgment conditions may include:

[0048] Gateway segment problem diagnosis: Check if the gateway's CPU / memory usage is too high, the running time is too long, the optical module parameters are abnormal, and if a large number of devices are simultaneously of poor quality.

[0049] Internet segment problem identification: Check whether the WAN-side network latency and WAN-side service latency are significantly higher than the baseline, whether the DNS resolution latency is too long, and whether the response of specific application servers is generally slow, etc.

[0050] WiFi segment problem diagnosis: Check if the LAN network latency is too high, the WiFi signal strength is too low, the negotiation rate is insufficient, there is serious interference in the WiFi channel (by analyzing neighbor WiFi information, current channel utilization, etc.), and whether it is working in a low-performance mode (such as single frequency, 5G priority mode not enabled), etc.

[0051] Troubleshooting connected terminal segments: Check the performance data reported by the terminal itself (if any), whether the terminal access type and negotiated rate match (e.g., a gigabit device connected to a 100 Mbps port), the terminal's historical poor performance, and whether the device is old (based on OSID or device model database), etc.

[0052] Once the chain of evidence is constructed, several poor-quality root cause hypotheses can be generated based on the chain of evidence, and the confidence level of each poor-quality root cause hypothesis can be evaluated. Specifically:

[0053] The evidence chain can be input into the decision tree algorithm, which generates several poor quality root cause hypotheses. At the same time, the decision tree algorithm is used to evaluate the fit between the evidence chain and each poor quality root cause hypothesis, and the fit is output as the confidence level of the poor quality root cause hypothesis.

[0054] Understandably, decision tree algorithms are a type of machine learning algorithm that can be pre-modeled using training data. The evidence chain, as input data, contains multiple feature dimensions, but different poor-quality root causes emphasize different dimensions. Therefore, a machine learning model can be built using decision tree algorithms, allowing the model to learn the relationship between the dimensions of the evidence chain and the root causes. The model then automatically evaluates and matches the evidence chain, assessing the fit between the evidence chain and the poor-quality root causes. Given that a single evidence chain may correspond to multiple possible root causes, the core task of the model is to find and output the best root cause with the highest matching degree among these options. Essentially, this process involves finding and outputting the poor-quality root cause with the highest confidence.

[0055] For example, if in a poor quality event, multiple terminals located on the same WiFi channel simultaneously experience high LAN network latency, and the probe scan detects multiple neighboring WiFi signals with strong signals on that channel, the decision tree algorithm will tend to judge it as "WiFi channel interference".

[0056] Step 103: Output the hypothesis with the highest confidence level as the primary root cause of poor quality, along with the chain of evidence for the primary root cause, information on poor quality, and solutions.

[0057] In embodiments of the present invention, the root cause hypothesis with the highest confidence level can be selected from several poor quality root cause hypotheses generated by the decision tree as the primary root cause. It is understood that the primary root cause is inferred by the decision tree algorithm and only represents one possibility leading to poor quality. Therefore, embodiments of the present invention can also output the root cause hypothesis with the second highest confidence level as a secondary root cause, along with the evidence chain, poor quality information, and solutions for the secondary root cause. Root cause hypotheses inferred by the decision tree algorithm may contain misjudgments. If only a single-dimensional primary root cause is output, it may lead to errors in root cause localization. Outputting secondary root causes can improve the accuracy of root cause localization from more dimensions.

[0058] The quality defect information can include the time when the quality defect occurred, the equipment involved, the application involved, the point of failure, and the specific cause type.

[0059] In embodiments of the present invention, the poor quality problem can be solved according to the solution, and the solution result can be fed back to the user network experience perception model, the root cause knowledge base, and the decision tree algorithm model.

[0060] Specifically, the primary and secondary root causes can be displayed to network operations personnel to help them quickly locate and resolve problems. This information can also be provided to other systems via API. Feedback from operations personnel or users (such as confirmed actual fault causes) will be used as new data input to the model training and iteration module to optimize the user network experience perception model, update rules and thresholds in the root cause knowledge base, and improve the decision tree algorithm parameters of the root cause inference engine, thereby enabling continuous learning and evolution of the system.

[0061] For ease of understanding, the present invention also provides the following embodiments:

[0062] Suppose a user experiences poor video quality, such as buffering and spinning, when watching a video application (e.g., "XX Video", AppID X) on a smart TV connected to a home gateway via WiFi.

[0063] Firstly, a probe plugin deployed on the home gateway can be used to continuously collect the following data and upload it to the cloud platform while the user is watching a video:

[0064] Streaming data: For the TCP / UDP stream of this video application (AppID is X), collect its average service latency on the WAN side (wan_app_lat), average service latency on the LAN side (lan_app_lat), number of packet losses (pkt_loss_cnt), and number of packets with payload (pkt_pl_cnt).

[0065] WiFi data: WiFi signal strength (power_level), negotiation rate (nego_rate), channel (channel), and neighboring WiFi interference on that channel (e.g., nei_ssid, nei_channel, nei_power_level obtained through scanning).

[0066] Gateway status: Gateway CPU utilization (cpu_use), memory utilization (mem_use), WAN port real-time speed (wan_rxrate), etc.

[0067] Terminal data: MAC address, IP address, etc. of the smart TV.

[0068] Then, the user network experience perception model on the cloud platform receives the above data. Based on the user network experience perception model, it calculates the user's perception score at the current sampling point (within 5 minutes). Assuming the calculated score is 45 points (below the poor quality threshold of 60 points), the system determines that a "video playback stuttering" poor quality event has occurred.

[0069] The system records the timestamp of the poor quality event and associates it with all relevant collected data before and after that time point (e.g., 5 minutes before and 1 minute after), forming a data snapshot for reference. Figure 2 .

[0070] Based on the fault scenarios in the root cause knowledge base, key indicators are extracted from data snapshots, such as: wan_app_lat = 250ms, lan_app_lat = 300ms, pkt_loss_cnt (WAN) = 0, pkt_loss_cnt (LAN) = 5, pkt_pl_cnt (LAN) = 1000 (i.e., LAN-side packet loss rate of 0.5%), TV WiFi signal strength = -75dBm, negotiation rate = 54Mbps, current channel is 6, and 3 other strong WiFi signals are detected on channel 6. Gateway CPU utilization = 20%, memory utilization = 30%, WAN port downlink rate = 5Mbps (assuming user bandwidth is 100Mbps, not reaching the bottleneck).

[0071] To determine whether these indicators can serve as evidence of the root cause of poor quality, the following root cause scenarios can be investigated:

[0072] Internet segment investigation: wan_app_lat = 250ms, assuming a baseline value of 200ms, slightly higher than the baseline but without any extreme anomalies, WAN-side packet loss is 0. Preliminary assessment suggests that there may be some fluctuations in the internet segment, but this is not the primary issue.

[0073] Gateway troubleshooting: Gateway CPU and memory usage are normal, and the WAN port speed is not at a bottleneck. Preliminary investigation has ruled out gateway performance as the bottleneck.

[0074] WiFi segment troubleshooting: lan_app_lat = 300ms, significantly high. LAN packet loss rate 0.5%, indicating some packet loss. TV WiFi signal strength -75dBm, weak. Negotiation rate 54Mbps, potentially insufficient for HD video. Multiple strong neighboring WiFi signals on current channel 6, indicating high likelihood of channel interference.

[0075] Downlink terminal segment troubleshooting: Assuming there are no data issues with the TV terminal itself.

[0076] The decision tree algorithm, based on the root cause knowledge base (e.g., the knowledge base contains information such as "when the LAN side has high latency, packet loss, weak signal, and large channel interference, the height is likely to be a WiFi coverage or interference problem"), and considering the above investigation items, concludes that there is a high probability that there is a problem with the WiFi segment.

[0077] The following root cause hypothesis was then generated and its confidence level was evaluated:

[0078] Hypothesis 1: WiFi channel interference causes a decrease in LAN-side communication quality (confidence level: high, 0.85). Evidence: High latency, packet loss, and significant interference on the LAN side.

[0079] Hypothesis 2: Poor WiFi signal coverage leads to unstable connection (Confidence: Medium, 0.65). Evidence: Signal strength of -75dBm is relatively weak.

[0080] Hypothesis 3: The application server response is slightly slow (confidence: low, 0.30). Evidence: wan_app_lat is slightly higher than the baseline.

[0081] Output the root cause hypotheses with the highest and second-highest confidence levels as the primary and secondary root causes, respectively:

[0082] Main cause: WiFi channel interference.

[0083] Secondary cause: Poor WiFi signal coverage.

[0084] Output details: Time [specific time], Device [Smart TV MAC], Application [XX Video], Location of poor signal quality [WiFi segment], Specific root cause [Severe interference exists on WiFi channel 6, and the device signal strength is weak], Suggestion [Try manually switching the router's WiFi channel to channels with less interference such as 1 or 11; adjust the position of the router or TV to improve the signal].

[0085] Furthermore, if the user adopts the suggestion to switch the WiFi channel and the video playback returns to smoothness, and this is confirmed in the feedback system, this success story will be used to strengthen the judgment rules regarding "channel interference" and the weight of their corresponding evidence in the root cause knowledge base. If switching the channel is ineffective, and maintenance personnel find during on-site troubleshooting that the outdated router firmware is causing unstable WiFi performance, this case will prompt the addition or adjustment of the judgment logic related to "router firmware / performance" in the root cause knowledge base, and may trigger a review of the gateway probe's collected data items (e.g., whether more detailed internal router status data needs to be collected).

[0086] This invention, through real-time collection of digital environment data and combined with a machine learning-based user network experience perception model, can more accurately quantify and evaluate users' real network experience under different applications, rather than relying solely on network layer indicators. This enables proactive perception of network quality before user reports of faults. Based on a root cause knowledge base and evidence chain reasoning, it accurately locates the root causes of poor network quality, improving troubleshooting efficiency and reducing reliance on manual intervention. By quickly and accurately locating the root causes of poor network quality, it can help operators and service providers optimize networks and handle faults more effectively, thereby improving operational efficiency, reducing user complaint rates, and achieving a synergistic leap in operational efficiency and user satisfaction.

[0087] This invention also provides a data-driven system for inferring the root causes of poor product quality, the system comprising:

[0088] The data receiving and preprocessing module is used to collect digital environment data in real time, input the collected digital environment data into the user network experience perception model, calculate and output the user's experience score at the current time point;

[0089] The root cause reasoning engine module is used to associate digital environment data contained within a preset time period before and after the current time point when the experience score is lower than a preset threshold, and form a poor quality data snapshot; based on the failure scenarios in the root cause knowledge base, it searches for and constructs evidence chains related to the failure scenarios from the poor quality data snapshot; generates several poor quality root cause hypotheses based on the evidence chains, and evaluates the confidence level of each poor quality root cause hypothesis based on the completeness of the evidence chains;

[0090] The output module is used to output the hypothesis of the poor quality root cause with the highest confidence as the main root cause, along with the evidence chain of the main root cause, information on the poor quality, and solutions.

[0091] The above provides a detailed description of a data-driven method and system for inferring the root causes of poor quality. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A data-driven based quality defect root cause reasoning method, characterized in that, The method comprises: Real-time collection of digital environment data, input of the collected digital environment data into a user network experience perception model, calculation and output of an experience score of a user at a current time point; In the case where the experience score is lower than a preset threshold, association of the digital environment data contained in a preset time before and after the current time point, and formation of a quality difference data snapshot; searching and constructing an evidence chain related to a fault scene in a root cause knowledge base from the quality difference data snapshot; generating a plurality of quality difference root cause hypotheses according to the evidence chain, and evaluating the confidence of each quality difference root cause hypothesis; Outputting a quality difference root cause hypothesis with the highest confidence as a main root cause, and attaching the evidence chain, quality difference information and solution of the main root cause; The searching and constructing of the evidence chain related to the fault scene in the root cause knowledge base from the quality difference data snapshot comprises: Extracting features related to the fault scene defined in the root cause knowledge base from the quality difference data snapshot; searching for evidence and constructing an evidence chain from the related features according to a preset root cause scene and root cause judgment condition; The generating of a plurality of quality difference root cause hypotheses according to the evidence chain and the evaluation of the confidence of each quality difference root cause hypothesis comprise: Inputting the evidence chain into a decision tree algorithm, generating a plurality of quality difference root cause hypotheses based on the decision tree algorithm, and evaluating the fitness of the evidence chain and each quality difference root cause hypothesis by using the decision tree algorithm, and outputting the fitness as the confidence of the quality difference root cause hypothesis.

2. The method of claim 1, wherein, The digital environment data comprises: network quality data, device state data, WiFi environment data, port rate data, and devices connected to the network.

3. The method of claim 2, wherein, The real-time collection of digital environment data, the input of the collected digital environment data into a user network experience perception model, and the calculation and output of an experience score of a user at a current time point comprise: Continuously collecting the digital environment data by using a probe plug-in built in a home gateway, and uploading the collected data to a cloud data processing and analysis platform according to a predetermined period; A user network experience perception model located in the cloud data processing and analysis platform receives the uploaded data, and calculates and outputs an experience score of a user at a current time point based on a machine learning algorithm.

4. The method of claim 3, wherein, The quality difference data snapshot is a data snapshot containing a time sequence.

5. The method of claim 4, wherein, The method further comprises: normalizing the extracted features related to the fault scene defined in the root cause knowledge base.

6. The method of claim 1, wherein, The method further comprises: outputting a quality difference root cause hypothesis with the second highest confidence as a secondary root cause, and attaching the evidence chain, quality difference information and solution of the secondary root cause; wherein the quality difference information comprises the time, device, application, fault point and specific reason type of the quality difference.

7. The method of claim 6, wherein, The method further comprises: solving the quality difference problem according to the solution, and feeding back the solution result to the user network experience perception model, the root cause knowledge base and the decision tree algorithm model.

8. A data-driven based quality defect root cause reasoning system, characterized in that, The system comprises: The data receiving and preprocessing module is configured to collect digital environment data in real time, input the collected digital environment data into a user network experience perception model, calculate and output an experience score of a user at a current time point; The root cause reasoning engine module is configured to, when the experience score is lower than a preset threshold, associate the digital environment data contained in a preset time before and after the current time point, and form a quality difference data snapshot; search and construct an evidence chain related to a fault scenario in a root cause knowledge base from the quality difference data snapshot; generate a plurality of quality difference root cause hypotheses according to the evidence chain, and evaluate the confidence of each quality difference root cause hypothesis according to the completeness of the evidence chain; the searching and constructing of the evidence chain related to the fault scenario in the root cause knowledge base from the quality difference data snapshot comprises: extracting features related to the fault scenario defined in the root cause knowledge base from the quality difference data snapshot; searching for evidence and constructing an evidence chain from the related features according to a preset root cause scenario and root cause judgment condition; the generating of the plurality of quality difference root cause hypotheses according to the evidence chain and the evaluation of the confidence of each quality difference root cause hypothesis comprise: inputting the evidence chain into a decision tree algorithm, generating a plurality of quality difference root cause hypotheses based on the decision tree algorithm, and evaluating the fitness of the evidence chain and each quality difference root cause hypothesis by using the decision tree algorithm, and outputting the fitness as the confidence of the quality difference root cause hypothesis; The output module is configured to output a quality difference root cause hypothesis with the highest confidence as a main root cause, and attach the evidence chain, quality difference information and solution of the main root cause.

Images