Methods, apparatus, equipment, media, and products for predicting user-perceived degradation in cells.

Abstract

This invention relates to the field of communication technology, and provides a method, apparatus, device, medium, and product for predicting user-perceived degraded cells. The method includes: acquiring a user's core network domain document; based on the user's identity identifier, associating the core network domain document with the corresponding radio network domain data to obtain a user wide table, and extracting user-perceived features and radio network domain features from the user wide table; inputting the user-perceived features and radio network domain features into a trained dual-domain model to predict degraded users, obtaining the degraded users output by the dual-domain model, and determining the degraded features from the user-perceived features and radio network domain features; and determining the degraded cells from each cell based on the degraded features of the cells traversed by the degraded users. The method, apparatus, device, medium, and product for predicting user-perceived degraded cells provided by this invention can achieve more accurate prediction of user-perceived degraded cells.

Description

Technical Field

[0001] This invention relates to the field of communication technology, and in particular to a method, apparatus, device, medium, and product for predicting user-perceived cell degradation. Background Technology

[0002] Currently, wireless network optimization is generally guided by Key Performance Indicators (KPIs). These KPIs are typically derived from the Operations and Maintenance Center (OMC) of the wireless equipment, and are statistically analyzed using the wireless coverage cell as the smallest geographical unit and a certain time interval (e.g., 15 minutes) as the time unit.

[0003] However, the KPIs of a wireless network are statistical indicators at the cell level over a period of time, and generally cannot represent the user's service experience within that cell. Often, even if the cell's wireless KPIs are good, there may still be cases where the user experience is poor.

[0004] Therefore, there is an urgent need for a method to predict cells with perceived degradation and to identify the wireless factors that cause this degradation, so as to optimize the wireless network of the cell. Summary of the Invention

[0005] This invention provides a method, apparatus, device, medium, and product for predicting user-perceived degraded cells, in order to address the deficiency in existing wireless KPI indicators that cannot characterize user service perception.

[0006] This invention provides a method for predicting user-perceived degradation cells, comprising:

[0007] Obtain the user's core domain information;

[0008] Based on the user's identity identifier, the core network domain document is associated with the wireless network domain data corresponding to the identity identifier to obtain a user wide table, and user perception features and wireless network domain features are extracted from the user wide table respectively.

[0009] The user-perceived features and the wireless network domain features are input into the trained dual-domain model to predict derogatory users, thereby obtaining the derogatory users output by the dual-domain model. Derogatory features are determined from the user-perceived features and the wireless network domain features. The dual-domain model is trained based on the user-perceived features, wireless network domain features, and derogatory labels of positive and negative sample users.

[0010] Based on the degradation characteristics of each cell visited by the degraded user, the degraded cell is identified from the cells.

[0011] According to the method for predicting user-perceived degradation cells provided by the present invention, the step of determining the degradation cells from the cells based on the degradation characteristics of each cell traversed by the degraded user includes:

[0012] Based on the deniability features of each user in the user wide table of the deniable users, determine the fused deniability features corresponding to the user wide table;

[0013] The degradation characteristics of each cell are compared with the fused degradation characteristics, and the degradation cells are determined from the cells based on the comparison results.

[0014] According to the method for predicting user-perceived degradation cells provided by the present invention, the step of comparing the degradation features of each cell with the fused degradation features, and determining the degradation cells from the cells based on the comparison results, includes:

[0015] The degradation characteristics of each cell are compared with the fused degradation characteristics, and the statistical frequency of each cell under each degradation characteristic is obtained based on the comparison results.

[0016] Based on the statistical frequency of each cell under each degradation feature, the degradation cell is determined, and the wireless degradation feature is determined from each degradation feature, and the wireless degradation feature is optimized.

[0017] According to the method for predicting user-perceived degradation cells provided by the present invention, the step of determining the degradation cells from the cells based on the degradation characteristics of each cell traversed by the degraded user includes:

[0018] Based on the degradation characteristics of each cell traversed by the first degraded user, the first degraded cell is determined from each cell. The first degraded user is obtained by using a basic model based on the user's core domain data to predict the degraded user.

[0019] Based on the degradation characteristics of each cell traversed by the second degraded user, the second degraded cell is determined from each cell. The second degraded user is predicted using the dual-domain model based on the core network domain data and wireless network domain data of the first degraded user.

[0020] Based on the first degraded cell and the second degraded cell, the final degraded cell is determined.

[0021] According to the method for predicting user-perceived degradation cells provided by the present invention, radio network domain features are extracted from the user wide table, including:

[0022] Obtain the field values ​​of each wireless network field in the wireless network domain data of the user wide table at multiple granularities within each document period;

[0023] Based on the field type of each wireless network field, the final field value of each wireless network field is determined from the field values ​​of the multiple granularities;

[0024] Based on the final field values ​​of each wireless network field, the wireless network domain features are extracted.

[0025] According to the method for predicting user-perceived degradation cells provided by the present invention, determining the final field value of each radio network field from the field values ​​of the plurality of granularities based on the field type of each radio network field includes:

[0026] For positive field types, the minimum value among the multiple granularity field values ​​is taken as the final field value;

[0027] For reverse field types, the maximum value among the multiple granular field values ​​is taken as the final field value.

[0028] According to the method for predicting user-perceived degradation cells provided by the present invention, the steps of constructing the user-perceived features include:

[0029] Obtain the core domain documents of positive and negative sample users, and calculate the statistics of each core domain field in the core domain documents within each document period;

[0030] Based on the differences in the distribution of statistics of each core network field of the positive and negative sample users, the user perception features are mined from each core network field.

[0031] According to the method for predicting user-perceived degradation cells provided by the present invention, the step of constructing the wireless domain features includes:

[0032] Obtain the wide user table for positive and negative sample users, and calculate the statistics of each wireless network field in the wireless network domain data of the wide user table within each document period;

[0033] Based on the differences in the distribution of statistics of each wireless network field of the positive and negative sample users, the wireless network domain features are mined from each wireless network field.

[0034] According to the method for predicting user-perceived degradation cells provided by the present invention, the training steps of the dual-domain model include:

[0035] The user perception characteristics and wireless network domain characteristics of the positive and negative sample users are input into the initial gradient boosting decision tree GBDT model to obtain the predicted depreciation probability of the positive and negative sample users output by the initial GBDT model.

[0036] Based on the difference between the predicted depreciation probability and the depreciation label, the initial GBDT model is trained to obtain the dual-domain model.

[0037] The present invention also provides an apparatus for predicting user-perceived degradation cells, comprising:

[0038] The document retrieval unit is used to retrieve the user's core domain documents;

[0039] The feature extraction unit is used to associate the core network domain document with the wireless network domain data corresponding to the identity identifier based on the user's identity identifier to obtain a user wide table, and extract user perception features and wireless network domain features from the user wide table respectively.

[0040] The user prediction unit is used to input the user-perceived features and the wireless network domain features into a trained dual-domain model to predict derogatory users, obtain the derogatory users output by the dual-domain model, and determine the derogatory features from the user-perceived features and the wireless network domain features. The dual-domain model is trained based on the user-perceived features, wireless network domain features, and derogatory labels of positive and negative sample users.

[0041] The cell determination unit is used to determine the degraded cell from the cells based on the degraded characteristics of each cell traversed by the degraded user.

[0042] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for predicting user-perceived degradation cells as described above.

[0043] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method for predicting user-perceived degradation cells as described above.

[0044] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the method for predicting user-perceived degraded cells as described above.

[0045] The method, apparatus, device, medium, and product for predicting user-perceived degradation cells provided by this invention incorporate user-perceived features and wireless network domain features during the training of a dual-domain model. Using this model can not only improve the accuracy of degradation user prediction, but also obtain degradation features in the explanation of user degradation causes. Furthermore, based on the degradation features of each cell traversed by the degradation user, the degradation cell can be determined from each cell, thus achieving a more accurate prediction of user-perceived degradation cells. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0047] Figure 1 This is a flowchart illustrating the method for predicting user-perceived cell degradation provided by the present invention.

[0048] Figure 2 This is one of the flowcharts illustrating the method for determining degraded cells provided by the present invention.

[0049] Figure 3 This is the second flowchart of the degraded cell determination method provided by the present invention.

[0050] Figure 4 This is a schematic diagram of the device for predicting user-perceived cell degradation provided by the present invention.

[0051] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0053] Mobile internet access is one of the main services offered by telecom operators, boasting a large user base. The quality of mobile internet access and user experience directly impact an operator's reputation. A poor experience can turn users into "dissidents," meaning those dissatisfied with the operator's brand, products, or services, resulting in a significant negative impact on the operator.

[0054] Based on operators' management and analysis of personal mobile internet service quality in recent years, the factors affecting service quality and user experience, besides issues with the user's personal terminal, are mainly related to wireless network problems. Therefore, operators' efforts to improve network quality and ensure service quality primarily focus on wireless network optimization. Currently, wireless network optimization is generally guided by wireless network KPIs. These KPIs are typically derived from the OMC (Original Data Capability) of wireless devices, and are statistically analyzed at a specific time interval (e.g., 15 minutes), with the wireless coverage cell as the smallest geographical unit. Network optimization personnel monitor whether wireless network KPIs have deteriorated and, based on experience, determine the causes of degradation. For example, "RRC connection establishment success rate" is generally related to the connection status between the terminal and the base station, while "maximum number of RRC connections" is generally related to the wireless network load. If KPIs deteriorate, various optimization methods are used to optimize the network. After optimization, if the corresponding KPIs return to normal, it indicates that the optimization is effective and the problem is resolved.

[0055] However, KPIs for wireless networks are statistical indicators at the cell level over a period of time and generally cannot represent the user's service experience within that cell. Often, a cell's wireless KPIs may be good, but there may still be instances of poor user experience. After all, the ultimate goal of operators improving network and service quality is not to maintain excellent KPIs, but to ensure a positive user experience. Therefore, network optimization departments within operators are more interested in identifying degraded cells with poor user experience and the underlying network causes, and then optimizing them to effectively improve user service experience and satisfaction.

[0056] Related technologies typically use pre-trained models to infer the core network data of existing users and predict the perceived depreciation of target users. However, this approach has the following shortcomings:

[0057] 1) In reality, network optimization is based on wireless network KPIs. However, models are built based on user perception. Therefore, the poor-quality cells output by the model may not correspond to the poor-quality cells identified based on wireless KPI thresholds. If the poor-quality cells output by the model do not have wireless KPIs below the threshold, then network optimization cannot be implemented.

[0058] 2) Since degraded users are mobile, they may pass through multiple cells within a certain period. If each cell they pass through is recorded as a degraded cell, it will introduce a certain bias.

[0059] To address the aforementioned problems, this invention provides a method for predicting user-perceived degraded cells. The method involves: acquiring a user's core network domain data; associating the core network domain data with the corresponding radio network domain data based on the user's identity identifier to obtain a user wide table, and extracting user-perceived features and radio network domain features from the user wide table; inputting the user-perceived features and radio network domain features into a trained dual-domain model for degraded user prediction, obtaining the degraded users output by the dual-domain model, and determining the degraded features from the user-perceived features and radio network domain features. The dual-domain model is trained based on the user-perceived features, radio network domain features, and degraded labels of positive and negative sample users; and determining the degraded cells from each cell based on the degraded features of the cells traversed by the degraded user.

[0060] In this embodiment, user-perceived features and wireless network features are incorporated during the training of the dual-domain model. Using this model can not only improve the accuracy of predicting degraded users, but also obtain the degraded features in the explanation of user degraded causes. Then, based on the degraded features of each cell passed by the degraded user, the degraded cell can be determined from each cell, thus predicting the cell where the user perceives degraded.

[0061] The embodiments of this invention can be applied to network optimization scenarios. The subject executing this method can be an electronic device such as a terminal device, computer, server, server cluster, or specially designed predictive device for user-perceived degradation cells, or a predictive device installed in such an electronic device, which can be implemented through software, hardware, or a combination of both.

[0062] Figure 1 This is a flowchart illustrating the method for predicting user-perceived degradation cells provided by the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0063] Step 110: Obtain the user's core domain documents.

[0064] Specifically, a user's core network domain documents can be all user documents left in the core network domain. Core network domain documents mainly include documents from core network user plane interfaces (collected by base stations and UPF sides, such as N3 and N6 interfaces), as well as documents from related signaling plane interfaces (mainly N1 and N2 interfaces).

[0065] Step 120: Based on the user's identity identifier, associate the core network domain document with the wireless network domain data corresponding to the identity identifier to obtain a user wide table, and extract the user perception features and wireless network domain features from the user wide table respectively.

[0066] Specifically, considering that user experience degradation is often related to the condition of the wireless network and rarely caused by problems on the core network side, incorporating wireless network information into the predictive model can improve the model's identification accuracy. A user's identifier can be, for example, a phone number. By associating core network domain documents with the corresponding wireless network domain data based on the user's phone number, a wide user table is obtained. This wide user table is based on the user's phone number and includes both core network domain documents and wireless network domain documents.

[0067] Subsequently, user-perceived features and wireless network domain features are extracted from the user wide table. Specifically, this could be the feature values ​​corresponding to each user-perceived feature and the feature values ​​corresponding to each wireless network domain feature.

[0068] The extraction of wireless network domain features can be achieved through the following methods:

[0069] Retrieve the field values ​​of each wireless network field in the wireless network domain data of the user wide table at multiple granularities within each document period;

[0070] Based on the field type of each wireless network field, the final field value of each wireless network field is determined from field values ​​at multiple granularities;

[0071] Based on the final field values ​​of each wireless network field, wireless network domain features are extracted.

[0072] Specifically, wireless network domain data can include multiple wireless network KPIs. Wireless network KPIs are generally obtained based on time-granularity statistics. In this embodiment, for a user transaction with a periodic granularity, the cells traversed by a single user within that transaction period are extracted, and the wireless metrics corresponding to these cells for that time period are obtained, i.e., multiple field values ​​corresponding to that transaction period. Taking a wireless metric with a 15-minute granularity as an example, if a user traverses a cell within one hour, resulting in a transaction with a duration of 35 minutes, it may correspond to 3 or 4 field values ​​at different granularities for that cell.

[0073] Subsequently, based on the field type of each wireless network field, the final field value of each wireless network field can be determined from field values ​​at multiple granularities. Specifically, this includes:

[0074] For positive field types, the minimum value among multiple granularity field values ​​is used as the final field value; for negative field types, the maximum value among multiple granularity field values ​​is used as the final field value.

[0075] For example, for load-related fields such as "maximum number of RRC connections" and "average downlink PRB utilization," the maximum value among three or four values ​​is taken. For service-related fields such as "QoS Flow establishment success rate" and "wireless connection success rate," the minimum value among three or four values ​​is taken. In other words, for positive fields (where a larger value corresponds to better perception), the minimum value among multiple granular field values ​​is taken; conversely, for negative fields (where a larger value corresponds to worse perception), the maximum value among multiple granular field values ​​is taken. By mapping multiple granular wireless field values ​​for a user to a single value, the feature value of the wireless network domain can be extracted.

[0076] Step 130: Input the user perception features and wireless network domain features into the trained dual-domain model to predict the derogatory users, obtain the derogatory users output by the dual-domain model, and determine the derogatory features from the user perception features and wireless network domain features. The dual-domain model is trained based on the user perception features, wireless network domain features, and derogatory labels of positive and negative sample users.

[0077] Specifically, after obtaining the feature values ​​of the user-perceived features and the wireless network domain features, these feature values ​​are input into a trained dual-domain model to predict degraded users, resulting in the output of the dual-domain model: degraded users. Degraded users are customers with poor internet service experience. Degraded features are identified from the user-perceived features and wireless network domain features. These degraded features are those that affect the user's internet service experience. The treeSHAP algorithm is used to derive the top features that contribute the most to the degraded experience; these are the degraded features.

[0078] It should be noted that the dual-domain model can be trained beforehand. The dual-domain model is trained based on the user perception features of positive and negative sample users, wireless network features, and derogatory labels.

[0079] In some embodiments, the training method for the dual-domain model includes:

[0080] 1) Constructing user-perceived features:

[0081] Obtain the core domain documents of positive and negative sample users, and calculate the statistics of each core domain field in the core domain documents within each document period; based on the differences in the distribution of the statistics of each core domain field of positive and negative sample users, extract user perception features from each core domain field.

[0082] Specifically, users who file complaints are labeled as derogatory users, and users who do not file complaints are labeled as non-derogatory users. Complaining users are used as positive samples, and a similar proportion of non-complaining users are used as negative samples. The positive and negative samples are randomly and evenly mixed, and then divided into training and validation sets according to a certain ratio, for example, a 4:1 ratio, with "4" as the training set and "1" as the validation set.

[0083] Obtain core network domain documents for positive and negative sample users, mainly including documents from core network user plane interfaces (collected from base stations and UPF sides, such as N3 and N6 interfaces), as well as documents from related signaling plane interfaces (mainly N1 and N2 interfaces).

[0084] The user's core domain documents are divided into time periods, such as one hour or 45 minutes, etc., and this embodiment of the invention does not specifically limit this. Within a document period granularity, the statistics of each core network field in all core domain documents are calculated. For example, for the core network field "TCP layer RTT latency", the "average TCP layer RTT latency" of all core domain documents within this document period is calculated.

[0085] Subsequently, the statistical measures of each core network field in the positive and negative samples are compared to see if there are significant differences in their distribution. If so, the core network field is constructed as a user-perceived feature; otherwise, the core network field cannot be constructed as a user-perceived feature. This embodiment is based on the concept of big data, fully mining features from a numerical level to construct a feature library of the largest possible scale.

[0086] In addition, considering the limited computing power and model generalization, various algorithms, such as the Boruta algorithm, can be used to select a suitable number of features from the existing user perception feature library and incorporate them into the model as the final user perception features.

[0087] 2) Constructing wireless domain characteristics:

[0088] Obtain the wide user table for both positive and negative sample users, and calculate the statistics of each wireless network field in the wireless network domain data of the wide user table within each document period; based on the differences in the distribution of the statistics of each wireless network field for both positive and negative sample users, extract wireless network domain features from each wireless network field.

[0089] Based on the user wide table, various statistics for all wireless network fields in the user wide table are calculated within a document period. For example, for the field "Maximum RRC Connections", the average, maximum, minimum, median, variance, and higher-order moments of this field are calculated for all documents within this period. The distribution of the statistics for positive and negative samples is compared to see if there are significant differences. If so, these are used to construct wireless network domain features.

[0090] 3) Model Training. The training steps for the dual-domain model include: inputting the user perception features and wireless network domain features of positive and negative sample users into the initial gradient boosting decision tree (GBDT) model to obtain the predicted depreciation probabilities of positive and negative sample users output by the initial GBDT model; and training the initial GBDT model based on the difference between the predicted depreciation probabilities and the depreciation labels to obtain the dual-domain model.

[0091] Specifically, after calculating the feature values ​​of each user's perceived characteristics and wireless network domain characteristics, these feature values ​​are input into the GBDT residual regression tree algorithm for training. The degradation probability of positive samples is denoted as 1, and the degradation probability of negative samples is denoted as 0. Multiple residual trees are ensembled to calculate an approximation of the degradation probability. Inference is performed using validation set data to obtain the user's degradation probability; values ​​greater than or equal to 0.5 are classified as poor quality, and values ​​less than 0.5 are classified as non-poor quality. Combined with the sample labels in the validation set, the model's recall and precision are calculated. The overall improvement in recall and precision is used as a standard to guide the optimization and shaping of the GBDT model, ultimately forming a dual-domain model of the perceived domain and wireless network.

[0092] Step 140: Based on the degradation characteristics of each community visited by the degraded user, identify the degraded communities from among the communities.

[0093] Specifically, after obtaining the degraded users and degraded characteristics through step 130, the degraded cells can be determined from among the cells based on the degraded characteristics of each cell traversed by the degraded users. Step 140 specifically includes:

[0094] Step 141: Based on the devaluation features in the user wide table of devaluing users, determine the fused devaluation features corresponding to the user wide table;

[0095] Step 142: Compare the degradation characteristics of each cell with the merged degradation characteristics, and determine the degraded cells from each cell based on the comparison results.

[0096] Here, querying the wide user table of a degraded user yields all the cells the user visited during the current period. The value of the feature in each cell is then calculated, resulting in the feature value of each degraded feature. The fused degraded feature is obtained by fusing the feature values ​​of these features, reflecting the average degree of degraded characteristics. For example, the fused degraded feature can be obtained by weighted averaging the degraded features from the wide user table of a degraded user.

[0097] Then, the degradation characteristics of each cell are compared with the merged degradation characteristics, and the degraded cells are identified from the cells based on the comparison results. Step 142 specifically includes:

[0098] Step 142-1: Compare the degradation characteristics of each cell with the fused degradation characteristics, and obtain the statistical frequency of each cell under each degradation characteristic based on the comparison results;

[0099] Step 142-2: Based on the statistical frequency of each cell under each degradation feature, determine the degraded cells, identify the wireless degradation features from each degradation feature, and optimize the wireless degradation features.

[0100] Figure 2 This is one of the flowcharts illustrating the degraded cell determination method provided by the present invention, such as... Figure 2 As shown, for the degraded user's visits to cells A, B, and C, for degrade feature 1, each cell's degrade feature 1 is compared with the merged degrade feature 1. It is found that cell B and cell C's degrade feature 1 is worse than the merged degrade feature 1, meaning their count can be increased by one. For degrade feature 2, each cell's degrade feature 2 is compared with the merged degrade feature 2. It is found that cell A and cell B's degrade feature 2 is worse than the merged degrade feature 2, meaning their count can be increased by one. This process continues, comparing multiple degrade features to obtain the count for each cell. Cells with a count greater than a threshold are considered degraded cells.

[0101] To determine wireless degradation characteristics, the frequency of occurrence of the top degradation perception domain features and degradation wireless network domain features corresponding to the degraded cells can be calculated based on the degradation features corresponding to the degraded users. Here, degradation perception domain features refer to user-perceived features within the degradation features, and degradation wireless network domain features refer to wireless network domain features within the degradation features. The degradation wireless network domain feature with the highest frequency is selected as the wireless degradation feature and used as a reference for network optimization to optimize the wireless degradation characteristics.

[0102] The method provided in this embodiment of the invention compares the degradation characteristics of multiple cells passed by a degraded user within a time period. Only when the degradation characteristics of a cell are lower than the average value within this time period is it recorded as a degraded cell corresponding to the user.

[0103] Based on any of the above embodiments, based on the degradation characteristics of each cell traversed by the degraded user, the degraded cell is determined from each cell, including:

[0104] Based on the degradation characteristics of each cell traversed by the first degraded user, the first degraded cell is determined from each cell. The first degraded user is predicted using the basic model based on the user's core network domain data.

[0105] Based on the degradation characteristics of each cell traversed by the second degradation user, the second degradation cell is determined from each cell. The second degradation user is predicted using a dual-domain model based on the core network domain data and wireless network domain data of the first degradation user.

[0106] The final degraded cell is determined based on the first and second degraded cells.

[0107] Specifically, in practical applications, due to the extremely large volume of core network domain data for all users, performing feature calculations after associating with radio domain data requires significant computing power. This embodiment proposes a two-layer prediction method. Figure 3 This is the second flowchart illustrating the degraded cell determination method provided by the present invention, as follows: Figure 3 As shown, a basic model is first trained using user-perceived data. Based on the user's core domain data, the basic model is used to predict devalued users and obtain the first devalued user. That is, a round of pre-inference is performed on all user data, and then users whose devaluation probability is greater than a certain threshold are selected as the first devalued users. Based on the devaluation characteristics of each cell visited by the first devalued user, the first devalued cell is determined from each cell.

[0108] Based on this, the core network domain data and wireless network domain data of the first degraded user are correlated, and the user perception characteristics and wireless network domain characteristics are calculated using the method described in the above embodiments. A dual-domain model is then used to predict the second degraded user. Based on the degrade characteristics of each cell traversed by the second degraded user, the second degraded cell is determined from among the cells.

[0109] It should be noted that the method for determining the first and second degraded cells can be referred to the description in the above embodiments, and will not be repeated here.

[0110] Then, the intersection of the first and second degraded cells is taken as the final output degraded cell.

[0111] The method provided in this invention superimposes a dual-domain model on top of a basic model to resolve the excessive computation introduced by directly associating wireless network statistical indicators with core network domain user data in practical applications. Furthermore, it obtains a more accurate user-perceived degraded cell by taking the intersection of the degraded cells in the basic model and the degraded cells in the dual-domain model.

[0112] Based on any of the above embodiments, a method for predicting user-perceived cell degradation is provided, comprising:

[0113] S1, dual-domain model training. Includes:

[0114] S11, Construct user perception features. Obtain core domain documents for positive and negative sample users, and calculate the statistics of each core domain field in the core domain documents within each document period; based on the differences in the distribution of the statistics of each core domain field for positive and negative sample users, extract user perception features from each core domain field.

[0115] S12, Construct wireless network domain features. Obtain the user wide table of positive and negative sample users, and calculate the statistics of each wireless network field in the wireless network domain data of the user wide table within each document period; based on the differences in the distribution of the statistics of each wireless network field of positive and negative sample users, extract wireless network domain features from each wireless network field.

[0116] S13, input the user perception features and wireless network domain features of positive and negative sample users into the initial gradient boosting decision tree GBDT model to obtain the predicted depreciation probability of positive and negative sample users output by the initial GBDT model; based on the difference between the predicted depreciation probability and the depreciation label, train the initial GBDT model to obtain a dual-domain model.

[0117] S2, predicts the perceived degradation of the cell. Includes:

[0118] S21, Obtain the user's core domain documents;

[0119] S22, based on the user's identity identifier, associate the core domain documents with the wireless domain data corresponding to the identity identifier to obtain the user wide table, and extract the user perception features and wireless domain features from the user wide table respectively.

[0120] S23, Input the user perception features and wireless network domain features into the trained dual-domain model to predict the degraded users, obtain the degraded users output by the dual-domain model, and determine the degraded features from the user perception features and wireless network domain features.

[0121] S24, based on the degradation characteristics of each cell traversed by the degraded user, identifies the degraded cells from among all cells. Specifically, this includes:

[0122] Based on the degradation features in the user wide table of degraded users, the fused degradation features corresponding to the user wide table are determined; the degradation features of each cell are compared with the fused degradation features, and the statistical frequency of each cell under each degradation feature is obtained according to the comparison results; based on the statistical frequency of each cell under each degradation feature, the degraded cells are determined, and the wireless degradation features are determined from each degradation feature, and the wireless degradation features are optimized.

[0123] Based on the degradation characteristics of each cell traversed by the first degraded user, the first degraded cell is determined from all cells. The first degraded user is predicted using a basic model based on the user's core network domain data. Based on the degradation characteristics of each cell traversed by the second degraded user, the second degraded cell is determined from all cells. The second degraded user is predicted using a dual-domain model based on the first degraded user's core network domain data and radio network domain data. Based on the first and second degraded cells, the final degraded cell is determined.

[0124] The method provided in this invention incorporates wireless network domain data for model training to obtain wireless factors affecting degraded user experience. This allows for more precise aggregation of data on degraded cells, obtaining the main wireless indicators of perceived degradation in the cell. This makes the direction of cell optimization for user experience clearer, leading to a more effective improvement in user experience through cell optimization. Furthermore, this proposal suggests a two-layer model inference method. After inferring the range of users with network perceived degradation from the basic perception domain model, wireless domain features are further incorporated into the dual-domain model for secondary inference. This significantly reduces computational overhead and provides guidance for actual network optimization work through the wireless features that contribute most to degradation obtained from the model interpretation.

[0125] Based on any of the above embodiments Figure 4 This is a schematic diagram of the device for predicting user-perceived degradation cells provided by the present invention, as shown below. Figure 4 As shown, the device includes:

[0126] Document acquisition unit 410 is used to acquire the user's core domain documents;

[0127] The feature extraction unit 420 is used to associate the core network domain document with the wireless network domain data corresponding to the identity based on the user's identity to obtain a user wide table, and extract user perception features and wireless network domain features from the user wide table respectively.

[0128] User prediction unit 430 is used to input the user perception features and the wireless network domain features into a trained dual-domain model to predict derogatory users, obtain the derogatory users output by the dual-domain model, and determine the derogatory features from the user perception features and the wireless network domain features. The dual-domain model is trained based on the user perception features, wireless network domain features, and derogatory labels of positive and negative sample users.

[0129] Cell determination unit 440 is used to determine the degraded cell from the cells based on the degraded characteristics of each cell passed by the degraded user.

[0130] The device provided in this embodiment of the invention incorporates user-perceived features and wireless network domain features during the training of the dual-domain model. Using this model can not only improve the accuracy of predicting degraded users, but also obtain the degraded features in the explanation of user degraded causes. Then, based on the degraded features of each cell passed by the degraded user, the degraded cell can be determined from each cell, thereby predicting the cell where the user perceives degraded.

[0131] Based on any of the above embodiments, the cell determination unit is specifically used for:

[0132] Based on the deniability features of each user in the user wide table of the deniable users, determine the fused deniability features corresponding to the user wide table;

[0133] The degradation characteristics of each cell are compared with the fused degradation characteristics, and the degradation cells are determined from the cells based on the comparison results.

[0134] Based on any of the above embodiments, the cell determination unit is specifically used for:

[0135] The degradation characteristics of each cell are compared with the fused degradation characteristics, and the statistical frequency of each cell under each degradation characteristic is obtained based on the comparison results.

[0136] Based on the statistical frequency of each cell under each degradation feature, the degradation cell is determined, and the wireless degradation feature is determined from each degradation feature, and the wireless degradation feature is optimized.

[0137] Based on any of the above embodiments, the cell determination unit is specifically used for:

[0138] Based on the degradation characteristics of each cell traversed by the first degraded user, the first degraded cell is determined from each cell. The first degraded user is obtained by using a basic model based on the user's core domain data to predict the degraded user.

[0139] Based on the degradation characteristics of each cell traversed by the second degraded user, the second degraded cell is determined from each cell. The second degraded user is predicted using the dual-domain model based on the core network domain data and wireless network domain data of the first degraded user.

[0140] Based on the first degraded cell and the second degraded cell, the final degraded cell is determined.

[0141] Based on any of the above embodiments, the feature extraction unit is specifically used for:

[0142] Obtain the field values ​​of each wireless network field in the wireless network domain data of the user wide table at multiple granularities within each document period;

[0143] Based on the field type of each wireless network field, the final field value of each wireless network field is determined from the field values ​​of the multiple granularities;

[0144] Based on the final field values ​​of each wireless network field, the wireless network domain features are extracted.

[0145] Based on any of the above embodiments, the feature extraction unit is specifically used for:

[0146] For positive field types, the minimum value among the multiple granularity field values ​​is taken as the final field value;

[0147] For reverse field types, the maximum value among the multiple granular field values ​​is taken as the final field value.

[0148] Based on any of the above embodiments, a feature construction unit is further included, for:

[0149] Obtain the core domain documents of positive and negative sample users, and calculate the statistics of each core domain field in the core domain documents within each document period;

[0150] Based on the differences in the distribution of statistics of each core network field of the positive and negative sample users, the user perception features are mined from each core network field.

[0151] Based on any of the above embodiments, the feature construction unit is used for:

[0152] Obtain the wide user table for positive and negative sample users, and calculate the statistics of each wireless network field in the wireless network domain data of the wide user table within each document period;

[0153] Based on the differences in the distribution of statistics of each wireless network field of the positive and negative sample users, the wireless network domain features are mined from each wireless network field.

[0154] Based on any of the above embodiments, the device further includes a model training unit, specifically used for:

[0155] The user perception characteristics and wireless network domain characteristics of the positive and negative sample users are input into the initial gradient boosting decision tree GBDT model to obtain the predicted depreciation probability of the positive and negative sample users output by the initial GBDT model.

[0156] Based on the difference between the predicted depreciation probability and the depreciation label, the initial GBDT model is trained to obtain the dual-domain model.

[0157] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include: a processor 510, a communications interface 520, a memory 530, and a communications bus 540, wherein the processor 510, the communications interface 520, and the memory 530 communicate with each other through the communications bus 540. The processor 510 can call logical instructions in the memory 530 to execute a method for predicting user-perceived degraded cells. This method includes: acquiring a user's core network domain document; based on the user's identity identifier, associating the core network domain document with the radio network domain data corresponding to the identity identifier to obtain a user wide table, and extracting user-perceived features and radio network domain features from the user wide table; inputting the user-perceived features and the radio network domain features into a trained dual-domain model to predict degraded users, obtaining degraded users output by the dual-domain model, and determining degraded features from the user-perceived features and the radio network domain features, wherein the dual-domain model is trained based on user-perceived features, radio network domain features, and degraded labels of positive and negative sample users; and determining degraded cells from each cell based on the degraded features of the cells traversed by the degraded user.

[0158] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0159] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the method for predicting user-perceived degradation cells provided by the above methods. The method includes: obtaining a user's core domain information; based on the user's identity identifier, associating the core domain information with the radio domain data corresponding to the identity identifier to obtain a user wide table, and extracting user-perceived features and radio domain features from the user wide table respectively; inputting the user-perceived features and the radio domain features into a trained dual-domain model to predict degradation users, obtaining the degradation users output by the dual-domain model, and determining degradation features from the user-perceived features and the radio domain features, wherein the dual-domain model is trained based on the user-perceived features, radio domain features, and degradation labels of positive and negative sample users; and determining degradation cells from each cell based on the degradation features of each cell traversed by the degradation user.

[0160] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements a method for predicting user-perceived degradation cells provided by the methods described above. The method includes: obtaining a user's core domain information; based on the user's identity identifier, associating the core domain information with the radio domain data corresponding to the identity identifier to obtain a user wide table, and extracting user-perceived features and radio domain features from the user wide table respectively; inputting the user-perceived features and the radio domain features into a trained dual-domain model to predict degradation users, obtaining the degradation users output by the dual-domain model, and determining degradation features from the user-perceived features and the radio domain features, wherein the dual-domain model is trained based on the user-perceived features, radio domain features, and degradation labels of positive and negative sample users; and determining degradation cells from the cells traversed by the degradation users based on the degradation features of each cell.

[0161] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0162] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0163] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting user-perceived degradation in a cell, characterized in that, include: Obtain the user's core domain information; Based on the user's identity identifier, the core network domain document is associated with the wireless network domain data corresponding to the identity identifier to obtain a user wide table, and user perception features and wireless network domain features are extracted from the user wide table respectively. The user-perceived features and the wireless network domain features are input into the trained dual-domain model to predict derogatory users, thereby obtaining the derogatory users output by the dual-domain model. Derogatory features are determined from the user-perceived features and the wireless network domain features. The dual-domain model is trained based on the user-perceived features, wireless network domain features, and derogatory labels of positive and negative sample users. Based on the degradation characteristics of each cell visited by the degraded user, the degraded cell is identified from each cell. The step of identifying degraded cells from the cells based on the degrade characteristics of each cell traversed by the degraded user includes: Based on the degradation characteristics of each cell traversed by the first degraded user, the first degraded cell is determined from each cell. The first degraded user is obtained by using a basic model based on the user's core domain data. The basic model is formed by training using user perception data. Based on the degradation characteristics of each cell traversed by the second degraded user, the second degraded cell is determined from each cell. The second degraded user is predicted using the dual-domain model based on the core network domain data and wireless network domain data of the first degraded user. Based on the first degraded cell and the second degraded cell, the final degraded cell is determined.

2. The method for predicting user-perceived degradation cells according to claim 1, characterized in that, The step of identifying degraded cells from the cells based on the degrade characteristics of each cell traversed by the degraded user includes: Based on the deniability features of each user in the user wide table of the deniable users, determine the fused deniability features corresponding to the user wide table; The degradation characteristics of each cell are compared with the fused degradation characteristics, and the degradation cells are determined from the cells based on the comparison results.

3. The method for predicting user-perceived degradation cells according to claim 2, characterized in that, The step of comparing the degradation characteristics of each cell with the fused degradation characteristics, and determining the degradation cells from the cells based on the comparison results, includes: The degradation characteristics of each cell are compared with the fused degradation characteristics, and the statistical frequency of each cell under each degradation characteristic is obtained based on the comparison results. Based on the statistical frequency of each cell under each degradation feature, the degradation cell is determined, and the wireless degradation feature is determined from each degradation feature, and the wireless degradation feature is optimized.

4. The method for predicting user-perceived degradation cells according to claim 1, characterized in that, Wireless network domain features are extracted from the user wide table, including: Obtain the field values ​​of each wireless network field in the wireless network domain data of the user wide table at multiple granularities within each document period; Based on the field type of each wireless network field, the final field value of each wireless network field is determined from the field values ​​of the multiple granularities; Based on the final field values ​​of each wireless network field, the wireless network domain features are extracted.

5. The method for predicting user-perceived degradation cells according to claim 4, characterized in that, The step of determining the final field value of each wireless network field from the field values ​​of the multiple granularities based on the field type of each wireless network field includes: For positive field types, the minimum value among the multiple granularity field values ​​is taken as the final field value; For reverse field types, the maximum value among the multiple granular field values ​​is taken as the final field value.

6. The method for predicting user-perceived degradation cells according to any one of claims 1 to 5, characterized in that, The steps for constructing the user-perceived features include: Obtain the core domain documents of positive and negative sample users, and calculate the statistics of each core domain field in the core domain documents within each document period; Based on the differences in the distribution of statistics of each core network field of the positive and negative sample users, the user perception features are mined from each core network field.

7. The method for predicting user-perceived degradation cells according to any one of claims 1 to 5, characterized in that, The steps for constructing the wireless domain features include: Obtain the wide user table for positive and negative sample users, and calculate the statistics of each wireless network field in the wireless network domain data of the wide user table within each document period; Based on the differences in the distribution of statistics of each wireless network field of the positive and negative sample users, the wireless network domain features are mined from each wireless network field.

8. The method for predicting user-perceived degradation cells according to any one of claims 1 to 5, characterized in that, The training steps for the dual-domain model include: The user perception characteristics and wireless network domain characteristics of the positive and negative sample users are input into the initial gradient boosting decision tree GBDT model to obtain the predicted depreciation probability of the positive and negative sample users output by the initial GBDT model. Based on the difference between the predicted depreciation probability and the depreciation label, the initial GBDT model is trained to obtain the dual-domain model.

9. An apparatus for predicting user-perceived degradation cells, characterized in that, include: The document retrieval unit is used to retrieve the user's core domain documents; The feature extraction unit is used to associate the core network domain document with the wireless network domain data corresponding to the identity identifier based on the user's identity identifier to obtain a user wide table, and extract user perception features and wireless network domain features from the user wide table respectively. The user prediction unit is used to input the user-perceived features and the wireless network domain features into a trained dual-domain model to predict derogatory users, obtain the derogatory users output by the dual-domain model, and determine the derogatory features from the user-perceived features and the wireless network domain features. The dual-domain model is trained based on the user-perceived features, wireless network domain features, and derogatory labels of positive and negative sample users. The cell determination unit is used to determine the degraded cell from the cells based on the degraded characteristics of each cell traversed by the degraded user; The step of identifying degraded cells from the cells based on the degrade characteristics of each cell traversed by the degraded user includes: Based on the degradation characteristics of each cell traversed by the first degraded user, the first degraded cell is determined from each cell. The first degraded user is obtained by using a basic model based on the user's core domain data. The basic model is formed by training using user perception data. Based on the degradation characteristics of each cell traversed by the second degraded user, the second degraded cell is determined from each cell. The second degraded user is predicted using the dual-domain model based on the core network domain data and wireless network domain data of the first degraded user. Based on the first degraded cell and the second degraded cell, the final degraded cell is determined.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method for predicting user-perceived degraded cells as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method for predicting user-perceived degraded cells as described in any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method for predicting user-perceived degraded cells as described in any one of claims 1 to 8.

Images