A network topology awareness-based home broadband user portrait construction method and system, electronic equipment and storage medium

Abstract

The application provides a network topology-aware household broadband user portrait construction method and system, electronic equipment and storage medium. The method takes the physical topology of an optical access network as a rigid constraint, takes a secondary optical splitter as a priority and a primary optical splitter as a fallback to identify a household broadband unit, solves the account level portrait fragmentation problem, extracts feature values from network behavior data and terminal management data, calculates the discrimination weight based on the distribution of the feature values in all accounts in the target area, and automatically reduces the weight of high-frequency pollution features; based on the weight and the preset quality condition, the feature values are aggregated and the comprehensive confidence is calculated in the household unit, and the portrait is generated by fusing the business attributes and the existing business labels of the unit reaching the threshold. Through the synergistic effect of topology constraint and adaptive feature suppression, the accuracy and business availability of the household broadband portrait are improved.

Description

Technical Field

[0001] This invention relates to the fields of communication network data processing, broadband service operation support, and user profile construction technology, and in particular to a method, system, electronic device, and storage medium for constructing home broadband user profiles based on network topology awareness. Background Technology

[0002] As broadband networks become the core gateway to digital home life, operators have accumulated massive amounts of data covering service acceptance, network authentication, traffic interaction, and terminal access. Building user profiles based on this data to support precision marketing, network optimization, and customer retention has become an important technological direction for broadband service operations.

[0003] In home broadband scenarios, user profiles are typically analyzed using broadband accounts as the smallest unit, generating tags by statistically analyzing a single account's traffic behavior, terminal characteristics, and service attributes. However, in real-world networks, the same physical household may have multiple broadband accounts due to changes in plans, relocation, or the coexistence of multiple accounts. A single account profile cannot reflect the overall scale of terminals, network usage preferences, and network quality of the household, leading to fragmented profiles. This, in turn, affects the accuracy of business decisions such as FTTR promotion, plan recommendations, and churn warnings. Furthermore, some existing solutions attempt to aggregate accounts using terminal identifiers or network characteristics, but common features such as cookies and device models are easily shared or corrupted among a large number of accounts. Using these as the basis for aggregation can easily introduce misassociations and incorrect labels, reducing the credibility of the profile.

[0004] Existing technologies lack a method for profiling home users that uses the physical topology of the optical access network as a rigid constraint for home identification while dynamically adjusting the aggregation contribution based on features distributed across the entire network. When splitter topology information is complete, fine-grained topology cannot be prioritized for accurate identification of home units; when topology information is missing, it cannot revert to the previous level of topology to maintain identification continuity. Furthermore, it cannot automatically reduce the weight of high-frequency pollution features based on network-wide statistics, allowing low-frequency stable features to play a dominant role in home aggregation. These issues result in deficiencies in the accuracy and service availability of home broadband user profiling. Summary of the Invention

[0005] In view of the above problems, a method, system, electronic device, and storage medium for constructing home broadband user profiles based on network topology awareness are proposed to overcome or at least partially solve the above problems. Specifically: A method for constructing home broadband user profiles based on network topology awareness includes: Obtain broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags for the home broadband account; the optical access network resource data records the primary splitter identifier, secondary splitter identifier, and installation address, and the authentication activity data records the account's activity status; The topology constraint prioritizes the secondary optical splitter and uses the primary optical splitter as a fallback topology constraint, combining the primary optical splitter identifier, secondary optical splitter identifier, installation address, and active status to identify home broadband units; Feature values ​​are extracted from network behavior feature data and terminal management data, and feature discrimination weights are calculated based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area. Within a home broadband unit, feature values ​​are aggregated based on feature discrimination weights and preset feature quality conditions to calculate the overall confidence level; A home broadband user profile is generated based on the overall confidence level, service attributes in broadband service data, and existing service tags.

[0006] Optionally, a topology constraint prioritizing the secondary optical splitter and using the primary optical splitter as a fallback, combined with primary optical splitter identifiers, secondary optical splitter identifiers, installation address, and active status, is used to identify home broadband units, including: Using the broadband account as the initial node, broadband accounts with the same secondary optical splitter identifier are grouped into the same candidate home broadband unit; When the secondary optical splitter identifier is missing or invalid, broadband accounts with the same primary optical splitter identifier will be grouped into the same candidate home broadband unit. Units with similar installation addresses among the candidate home broadband units are merged to obtain merged units; The number of accounts exceeding a preset threshold is removed from the merged units to obtain the home broadband units.

[0007] Optionally, the feature discrimination weights are calculated based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area, including: Let N be the total number of household broadband accounts in the target area, and df be the number of household broadband accounts containing the feature value. The feature discrimination weight idf is calculated using the following formula. weight : IDF weight = log[1 + N / (1 + df)] / log(1 + N) Optional, preset feature quality conditions include: When a local area network (LAN) MAC address exists and repeatedly appears in the same home broadband unit on multiple dates, the LAN MAC address is considered a stable physical characteristic. When the feature discrimination weight corresponding to a single non-physical feature exceeds the preset strong evidence threshold, the non-physical feature is regarded as strong evidence. When multiple non-physical features appear simultaneously and the feature discrimination weights corresponding to each non-physical feature exceed the preset effective feature threshold, the multiple non-physical features are used as combined evidence. The preset effective feature threshold is lower than the preset strong evidence threshold. When the feature discrimination weight corresponding to a feature value is lower than the preset effective feature threshold, the feature value is regarded as a low-weight common feature. Low-weight common features will not pass the feature quality condition due to the accumulation of quantity.

[0008] Optionally, the calculation of the overall confidence level includes: The topology base score is determined based on the topology hierarchy, and the topology adjustment coefficient is determined based on the consistency of the installation address and the number of accounts within the home broadband unit. The topology reliability is determined by the topology base score and the topology adjustment coefficient. Feature credibility is determined by the sum of the number of strong feature hits, the number of effective features, and the discrimination weight. Activity consistency is determined based on the number of active days in the past 7 days, the number of active days in the past 30 days, the most recent occurrence time, and changes in account status; The overall confidence level is determined by topological confidence, feature confidence, and activity consistency.

[0009] Optionally, generating a home broadband user profile based on comprehensive confidence level, service attributes in broadband service data, and existing service tags includes: The overall confidence level is compared with a preset threshold and a preset candidate threshold, and the preset threshold is higher than the preset candidate threshold. When the overall confidence level reaches a preset threshold, a formal profile is generated by combining business attributes and existing business tags. When the overall confidence level is lower than the preset threshold and reaches the preset candidate threshold, a profile to be reviewed is generated by combining business attributes and existing business tags. When the overall confidence level is lower than the preset candidate threshold, the intermediate process is retained or no image is output.

[0010] Optionally, generating a home broadband user profile includes generating at least one of the following: device profile, behavior profile, quality profile, family structure profile, and business scenario profile. The device profile includes the total number of terminals, mobile phones, televisions, and cameras; Behavioral profiles include active time periods, gaming preferences, video preferences, and external network access; The quality profile includes WiFi quality indicators and application quality indicators; Family structure profiles include the size of the family's terminals; The business scenario profile includes FTTR promotion compatibility, urgency of equipment replacement, need for package upgrades, opportunities for integrated marketing, and auxiliary identification tags for account change risks.

[0011] A system for constructing home broadband user profiles based on network topology awareness includes: The data access module is used to obtain broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags for home broadband accounts; the optical access network resource data records the primary splitter identifier, secondary splitter identifier, and installation address, and the authentication activity data records the account activity status; The topology sensing module is used to identify home broadband units by prioritizing topology constraints for secondary optical splitters and using primary optical splitters as fallback topology constraints, combined with primary optical splitter identifiers, secondary optical splitter identifiers, installation addresses, and active status. The feature weight calculation module is used to extract feature values ​​from network behavior feature data and terminal management data, and calculate the feature discrimination weight based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area. The trusted aggregation module is used to aggregate feature values ​​within a home broadband unit based on feature discrimination weights and preset feature quality conditions, and to calculate the overall confidence level. The profile tag generation module is used to generate a profile of a home broadband user for home broadband units whose comprehensive confidence level reaches a preset threshold, based on the comprehensive confidence level, service attributes in broadband service data, and existing service tags.

[0012] An electronic device includes a processor and a memory, the memory storing program instructions, and the processor executing the program instructions to implement any of the methods described above.

[0013] A computer-readable storage medium storing program instructions that, when executed by a processor, implement any of the methods described above.

[0014] Compared with existing technologies, this invention uses the physical topology of the optical access network as a rigid constraint for identifying home broadband units. In the case of broadband account changes, relocation, or multiple accounts coexisting, the secondary optical splitter is used as the priority topology constraint and the primary optical splitter as the fallback topology constraint. Home units are divided by combining the installation address and account activity status, which improves the granularity of profile analysis from a single account to the actual physical home, eliminating the interference of fragmented profiles at the account level on business judgment.

[0015] During the feature aggregation process, this invention calculates feature discrimination weights within the statistical scope of the entire network or target area, which automatically reduces the weight of common cookies and weak features shared by a large number of accounts, while low-frequency and stable physical features such as local area network MAC play a leading role in the aggregation, thus solving the problem of misaggregation and incorrect labeling caused by high-frequency polluting features.

[0016] Based on this, the present invention integrates topology credibility, feature credibility, and activity consistency to generate a comprehensive confidence level for home broadband units. When a preset threshold is reached, it integrates broadband service attributes and existing service tags to output a home broadband user profile. This makes the profile results both interpretable with physical topology support and capable of periodic updates, providing accurate and reliable home-level decision-making basis for operational scenarios such as FTTR promotion, package upgrades, equipment replacement, and churn warning. Attached Figure Description

[0017] 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.

[0018] Figure 1 This is a flowchart of a method for constructing a home broadband user profile based on network topology awareness, provided by an embodiment of the present invention. Figure 2 This is a system architecture diagram for constructing home broadband user profiles based on network topology awareness, provided by an embodiment of the present invention. Figure 3 This is a flowchart of the topology-aware home unit identification process provided in an embodiment of the present invention; Figure 4 This is a flowchart of the adaptive weighting and feature quality threshold provided in the embodiments of the present invention; Figure 5 This is a mapping diagram of family portrait tags and application scenarios provided in the embodiments of the present invention. Detailed Implementation

[0019] 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.

[0020] Reference Figures 1 to 5This invention provides a method for constructing a home broadband user profile based on network topology awareness, which may specifically include: Obtain broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags for the home broadband account; the optical access network resource data records the primary splitter identifier, secondary splitter identifier, and installation address, and the authentication activity data records the account's activity status; The topology constraint prioritizes the secondary optical splitter and uses the primary optical splitter as a fallback topology constraint, combining the primary optical splitter identifier, secondary optical splitter identifier, installation address, and active status to identify home broadband units; Feature values ​​are extracted from network behavior feature data and terminal management data, and feature discrimination weights are calculated based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area. Within a home broadband unit, feature values ​​are aggregated based on feature discrimination weights and preset feature quality conditions to calculate the overall confidence level; A home broadband user profile is generated based on the overall confidence level, service attributes in broadband service data, and existing service tags.

[0021] In the specific implementation process, the method for constructing a home broadband user profile provided in this embodiment can be performed according to the following steps.

[0022] Step S101: Collect multi-source data.

[0023] Broadband service data is obtained from the broadband service operation support system. The broadband service data is indexed by the home broadband account, and each record includes fields such as account identifier, contracted bandwidth, package type, service activation date, and service status. The contracted bandwidth field records the promised bandwidth value when the account was activated, and the package type field records the package category, such as a bundled package or a single broadband package. Authentication activity data comes from the authentication logs of the remote authentication dial-up user service system or the broadband remote access server. Each record includes the account identifier, first authentication time, most recent successful authentication time, and the number of active days in days. The active days are counted as follows: if the account has at least one successful authentication record within the statistical period, that record is considered an active day; the cumulative number of active days within the statistical period is the active days. This activity information constitutes the account's activity status. Optical access network resource data comes from the optical access network resource management system, recording the affiliation of each home broadband account in the access network physical topology. Each record includes the account identifier, primary splitter identifier, secondary splitter identifier, and installation address. The primary and secondary optical splitter identifiers are unique codes for optical splitting equipment in an optical access network, typically represented by the equipment serial number or a resource code assigned in the network management system. The installation address is the user-side address text registered in the installation work order.

[0024] Network behavior feature data originates from a deep packet inspection (DPI) platform. Deployed at the core or aggregation layer of broadband networks, the DPI platform acquires user plane data packets via optical splitting or mirroring. After parsing application-layer protocols and matching them against a feature library, it extracts multiple types of network behavior features. One type of feature is Hypertext Transfer Protocol (HTTP) Cookies, extracted from the Cookie field in the HTTP request header. Cookies are text information stored by a web server in a user's browser, used to identify a user session or user identity. Another type of feature is the access domain name, extracted from the Host field in the Domain Name System (DNS) query request or the HTTP request header. A third type of feature is the application name, determined by matching traffic features with the protocol fingerprints of known applications using the DPI feature library. A fourth type of feature is video application identifiers and game application identifiers, extracted from specific fields in video streaming protocols or game protocols. Finally, a fifth type of feature is camera application access records, identified from the interactive signaling of real-time streaming protocols or session initiation protocols.

[0025] Terminal management data originates from the terminal management platform. The platform periodically interacts with the session-side optical network unit or home gateway via Simple Network Management Protocol (SMLP) or TR-069 protocol to obtain the operating status and configuration information of the terminal devices. Terminal management data includes a list of LAN-side Media Access Control (MAC) addresses, device identifier, device model, and device brand. The MAC address is the hardware address of the network interface, programmed by the manufacturer during production and is globally unique. The device identifier is a unique identifier for the terminal device, while the device model and brand identify the manufacturer and product series.

[0026] Existing business tags originate from the tag management system and are various tags generated for broadband accounts based on historical business data during operation. These include tags such as high-value customer tags, customer tags indicating a tendency to complain, or tags representing specific service usage preferences. During data collection, a source system identifier and a data statistics window date range are simultaneously recorded for each type of data. The source system identifier distinguishes different data sources, and the data statistics window date range marks the coverage time interval of this batch of data, providing a basis for subsequent profile updates and traceability.

[0027] Step S102: Perform label standardization processing.

[0028] For the multi-source data collected in step S101, the following processes are performed sequentially: format unification, null value filtering, outlier removal, and duplicate record processing. Broadband accounts have different encoding formats in different business systems. Some systems add leading zeros or additional prefix characters before the account code. The standardization process unifies the broadband accounts from all sources into a raw account format with the prefix and padding removed. The raw account format only retains the original business account's number or character sequence.

[0029] Time fields have multiple representations in different systems, such as Unix timestamps, ISO date strings, or integer date encoding. Standardization converts all time fields into a unified business date format. Business dates are represented by eight-digit integers representing year, month, and day. For example, January 15, 2024 is represented as 20240115.

[0030] The standardization process for primary and secondary optical splitter identifiers includes encoding format verification and outlier removal. Identifiers that do not conform to the established encoding rules are marked and removed. The established encoding rules are the device encoding specifications defined in the optical access network resource management system, such as encoding length range and character set limitations.

[0031] The standardization of installation addresses includes removing spaces, punctuation marks, and redundant suffixes from the address text, converting characters to lowercase, and restoring the full names of common abbreviations of administrative divisions, such as replacing the abbreviations of city / district names in the address text with the complete city / district names.

[0032] Media access control addresses exist in various formats across different systems, including those separated by colons, hyphens, or no separators. Standardization removes all separators and converts the characters to consecutive hexadecimal characters. For required fields such as broadband account and splitter identifier, if a value is empty, the record is marked as invalid and removed from the dataset used in subsequent calculations. For optional fields such as installation address, if a value is empty, the record is retained but marked as missing. Duplicate records from the same data source within the same statistics window are only retained for the record with the most recent collection timestamp.

[0033] Step S103: Construct a basic account profile.

[0034] Using the standardized broadband account as the primary key, a basic profile record is generated for each account. The fields of the basic profile record include the validity date, the most recent authentication time, the number of active days in the last seven days, the number of active days in the last thirty days, the associated primary splitter identifier, the secondary splitter identifier, the standardized installation address, the contracted bandwidth, the number of terminals, and the device type.

[0035] The valid date field is primarily taken from the service activation date in the broadband service data. If the service activation date field has a null value, the first appearance date of the account in the authentication activity data is used as the supplementary valid date value. The most recent authentication time field is taken from the most recent successful authentication time in the authentication activity data. The active days in the last seven days and the active days in the last thirty days are taken from the active days statistics of two different statistical windows in the autonomous authentication activity data, respectively. The contracted bandwidth field is taken from the broadband service data, and the terminal quantity and device type fields are initially extracted from the terminal management data. The application access record field is initially extracted from the network behavior feature data, recording the list of application names accessed by the account within the statistical period and the access frequency statistics of the corresponding applications. The quality indicator field is initially extracted from the terminal management data and network behavior feature data, recording the wireless LAN signal quality parameters and application layer experience quality parameters of the account. The basic account profile provides basic data at the account granularity for subsequent home broadband unit identification and profile tag generation. Each field is gradually called and updated in subsequent steps.

[0036] Step S104: Identify the home broadband unit.

[0037] This step utilizes the hierarchical physical topology of the optical access network to aggregate multiple broadband accounts belonging to the same physical household into a single household broadband unit, thereby elevating the granularity of profile analysis from the account level to the household level.

[0038] A typical physical deployment architecture for an optical access network is as follows: An optical line terminal unit (OLT) is deployed in the operator's equipment room or access point, connecting to multiple primary optical splitters via a backbone fiber. Each primary splitter connects to multiple secondary splitters via distribution fibers. Each secondary splitter connects to several user-side optical network units (ONUs) via drop fibers. Each ONU provides broadband access service to a single household and corresponds to one or more broadband accounts. In this architecture, primary splitters are typically deployed in community optical junction boxes or building low-voltage rooms, connecting to users in multiple buildings or units. Secondary splitters are typically deployed in corridor fiber distribution boxes or floor distribution boxes, connecting to the ONUs of several adjacent households. Therefore, the set of accounts connected to a secondary splitter has a closer correspondence with physical households, and the secondary splitter identifier is prioritized as a topology constraint for identifying household broadband units.

[0039] During operation, the primary and secondary optical splitter identifiers associated with each broadband account in the account basic profile generated in step S103 are read. For broadband accounts with existing secondary optical splitter identifiers and passed format verification, all broadband accounts under the same secondary optical splitter identifier are assigned to the same candidate home broadband unit. For broadband accounts with missing secondary optical splitter identifiers or failed format verification, their associated primary optical splitter identifiers are retrieved, and broadband accounts under the same primary optical splitter identifier are assigned to the same candidate home broadband unit. At this time, the candidate home broadband unit is marked as generated using the primary optical splitter fallback method.

[0040] After generating candidate home broadband units, verification and filtering are performed based on the installation address and account activity status: the text consistency of the standardized installation addresses of each account within the unit is compared to determine the degree of address matching; the activity status of each account within the unit is checked, and abnormal aggregations that do not conform to the home characteristics are excluded, considering the reasonableness of the number of accounts within the unit. After verification and filtering, the official home broadband unit identification results are obtained. The priority fallback mechanism of topology constraints ensures that when fine-grained topology data is available, accurate home division can be achieved using a secondary optical splitter; when fine-grained topology data is missing, it automatically falls back to the previous level of topology constraints to maintain the continuity of identification.

[0041] Step S105: Extract candidate features for family units.

[0042] The feature information associated with each broadband account in the standardized network behavior feature data and terminal management data from step S102 is extracted and organized into structured feature edges. Each structured feature edge includes a broadband account field, a feature type field, a feature value field, an occurrence date field, a source system field, and a count field.

[0043] The broadband account field identifies the account from which the feature value originates. The feature type field labels the attribute category of the feature value, based on its physical source and generation mechanism: Local area network media access control addresses extracted from terminal management data are classified as physical features because they are unique identifiers fixed in network interface hardware during manufacturing and have a strong physical binding relationship with terminal devices; Cookie identifiers extracted from network behavior feature data are classified as non-physical features because cookies are software-level identifiers stored by the network server in the user's terminal browser and may change or be shared due to browser settings, user operations, or cross-device synchronization; Device identifiers, device models, and device brands extracted from terminal management data, although originating from terminal hardware information, are also classified as non-physical features because multiple devices may share the same model or brand. The feature value field stores the specific feature content string. The occurrence date field records the date the feature value was observed in the data. The source system field records the identifier of the system that collected the feature data. The count field records the cumulative frequency of the feature value appearing in the statistics window.

[0044] Step S106: Calculate the adaptive weights of the features.

[0045] Based on the structured feature edge set extracted in step S105, the feature discrimination weight is calculated for each feature value that appears therein.

[0046] Feature discrimination weights are used to quantify the discriminative power of a particular feature across the entire account population. Their calculation is based on the observation that in a network environment covering a large number of household broadband accounts within a target area, certain feature values ​​repeatedly appear in a vast amount of account data. For example, public cookie identifiers issued by mainstream content delivery network (CDN) providers, serving a massive user base, may be collected from the network behavior data of thousands or even tens of thousands of accounts; similarly, certain terminal device models with extremely high shipment volumes also appear frequently in terminal management data. Because these feature values ​​are prevalent across a large number of accounts, they contribute little to distinguishing whether two or more accounts belong to the same physical household. Conversely, some feature values ​​appear only in a very small number of accounts and recur stably over multiple days. For example, the LAN media access control address of a specific brand router purchased by a household may only appear in a handful of accounts within the target area. These feature values ​​have a high discriminative power in determining account affiliation.

[0047] During the calculation, all household broadband accounts participating in this profile construction within the target area are used as the statistical population, and the total number of unique accounts included in this population is denoted as N. For each feature value whose weight is to be calculated, the number of unique accounts containing that feature value within the statistical population is counted, denoted as df. The statistical scope of df is exact matching at the feature value level, that is, only accounts with identical feature value strings are counted in the same df. The larger the df value, the more prevalent the feature value is within the statistical population, and the lower the calculated feature discrimination weight; the smaller the df value, the scarcer the feature value, and the higher the calculated feature discrimination weight. Through this adaptive weight calculation based on actual data distribution, the aggregation contribution of high-frequency feature values ​​is automatically suppressed, while the aggregation contribution of low-frequency stable feature values ​​is automatically amplified, and the entire process does not require manual pre-setting of feature blacklists or whitelists.

[0048] Step S107: Calculate the feature quality threshold.

[0049] After calculating the feature discrimination weights of each feature value in step S106, the shared features within each home broadband unit are judged to meet the aggregation trust standard based on preset feature quality conditions. The feature quality conditions are graded according to feature category and feature discrimination weight level.

[0050] If a physical feature category exists within a unit and that physical feature is observed among multiple member accounts within the unit, the credibility of the feature aggregation for that unit can be improved. If no physical feature exists within a unit, but a non-physical feature exists whose feature discrimination weight exceeds a preset strong evidence threshold, then the aggregation for that unit can be supported.

[0051] If a unit contains multiple non-physical features that are shared by multiple member accounts, and the discriminative weight of each non-physical feature exceeds the preset effective feature judgment threshold, then combined evidence can be used to support aggregation.

[0052] For low-weight common features whose discriminative weight is below the effective feature judgment threshold, the cumulative occurrence of these features among accounts within a unit is not included in the judgment criteria for aggregation credibility. Through the hierarchical judgment of feature quality thresholds, the aggregation results rely only on features with sufficient discriminative power as evidence, avoiding interference from common common features in the aggregation process.

[0053] Step S108: Calculate the overall confidence level of the household unit.

[0054] The overall confidence level of a home broadband unit is determined by combining the confidence levels of its components across three dimensions: topology confidence, feature confidence, and activity consistency. Topology confidence reflects the reliability of the operation itself, which is based on the physical topology of the optical access network to identify the home broadband unit. It is related to the actual topology level used to identify the home broadband unit in step S104, the degree of text matching of the standardized installation addresses of each account within the unit, and the number of accounts within the unit.

[0055] When the identification process uses a secondary beam splitter as a topological constraint, the topological confidence level is high because the secondary beam splitter has a closer correspondence with physical homes; when the identification process uses a primary beam splitter for backtracking, the topological confidence level is adjusted accordingly.

[0056] Feature credibility reflects the reliability of the feature aggregation results in step S107, and is related to the existence of strong features that pass the feature quality threshold within the cell, the number and type coverage of effective features, and the cumulative value of the discrimination weights corresponding to these effective features.

[0057] Activity consistency reflects the degree of consistency in network usage patterns among broadband accounts within a home broadband unit over time. It is calculated by comparing the distribution of active days for each member account within the unit across two time windows: the past seven days and the past thirty days. This comparison also considers the time interval between the most recent successful authentication for each account and changes in account service status. Account combinations with more similar activity patterns receive higher activity consistency scores. The overall confidence score for the home broadband unit is obtained by comprehensively processing the scores from the topology confidence score, feature confidence score, and activity consistency score.

[0058] Step S109: Generate family profile tags.

[0059] For home broadband units whose comprehensive confidence level calculated in step S108 reaches a preset threshold, a profile label for the home broadband unit is generated by combining the service attributes and existing service tags in the broadband service data obtained in step S101.

[0060] The contracted bandwidth and package type in the business attributes provide the basic parameters for generating business scenario dimension tags. Existing business tags supplement the profile output with historical annotation information accumulated by the family in previous operational activities. The generated profile tags include five dimensions: device profile, behavior profile, quality profile, family structure profile, and business scenario profile.

[0061] The device profile describes the overall composition of terminal devices within the home broadband unit, including the total number of terminals, mobile phones, televisions, cameras, and the device capability level and age of the optical network unit or home gateway. The device capability level is determined based on the device model obtained from the terminal management data, combined with the supported protocol standards and hardware configuration parameters of the device model. The device age is determined based on the interval between the device's first online time in the terminal management data and the current time. The behavioral profile describes the internet access behavior preferences of members within the home broadband unit, including high-activity periods, gaming application preferences, video application preferences, children's application access identifiers, and inter-network access identifiers. The children's application access identifier is obtained by matching application names in the network behavior feature data with a pre-built application classification knowledge base, which maintains a list of known applications serving children. The quality profile describes the network access quality of the home broadband unit, including Wi-Fi quality identifiers, application layer experience quality identifiers, and low-quality experience identifiers. The family structure profile describes the scale characteristics of the home broadband unit, including the family terminal scale inferred based on the number of terminals and member accounts. The business scenario profile describes the suitability of the home broadband unit for marketing promotion and customer retention, including the suitability for fiber-to-the-room promotion, the urgency of replacing optical network units or home gateway equipment, the need for package upgrades, the opportunity for integrated marketing, and auxiliary identification tags for account change risks.

[0062] Step S110: Output the traceable portrait result.

[0063] The profile tags generated in step S109 and the key intermediate data generated during the construction process are organized into structured output records. Each output record uses the home broadband unit identifier as the primary key, and is associated with a list of member broadband accounts, the topology level used to identify the home broadband unit, the corresponding primary and secondary optical splitter identifiers, the source of key features hit during the feature aggregation stage and the feature discrimination weight of each feature, the comprehensive confidence score, the content of each profile tag, the business scenario score, the data statistics date, and a description of the hit rules for each tag. The feature discrimination weight is the discrimination weight calculated in step S106 for the feature value. The business scenario score is generated based on the comprehensive evaluation results of each sub-tag in the business scenario profile. The hit rule description records which feature combinations or data conditions triggered the generation of each profile tag. The output records are distributed to one or more of the following systems through the application output module: tag service system, operation system, marketing recommendation system, network quality analysis system, or risk auxiliary identification system, for use in network operation and business marketing scenarios.

[0064] In one or more embodiments of the present invention, the topology constraint prioritizing the secondary optical splitter and the primary optical splitter as a fallback topology constraint, combined with the primary optical splitter identifier, the secondary optical splitter identifier, the installation address, and the active status to identify the home broadband unit includes: Using the broadband account as the initial node, broadband accounts with the same secondary optical splitter identifier are grouped into the same candidate home broadband unit; When the secondary optical splitter identifier is missing or invalid, broadband accounts with the same primary optical splitter identifier will be grouped into the same candidate home broadband unit. Units with similar installation addresses among the candidate home broadband units are merged to obtain merged units; The number of accounts exceeding a preset threshold is removed from the merged units to obtain the home broadband units.

[0065] The installation address similarity merging step handles situations where the same household is split into different candidate units due to relocation, replacement, or address entry discrepancies. Address similarity calculation uses standardized installation address text as input, segmenting the address string into several address elements according to its hierarchical structure. Provincial, municipal, and district-level administrative region address elements are only used to confirm the approximate area to which the address belongs, while end-point address elements such as street name, community name, building number, unit number, and house number are used for similarity calculation. For cases where the community name appears in the actual installation work order with aliases, abbreviations, or the full name inconsistent with the actual name, a pre-built address alias mapping table is used to uniformly replace these aliases before similarity calculation.

[0066] Similarity is determined based on the co-occurrence of address elements; the higher the overlap of the terminal address elements of two addresses, the higher the similarity. When any pair of address elements in two candidate home broadband units overlaps to a preset condition, the two candidate units are merged into a single merged unit. The merging process is performed pairwise among all candidate units, and the newly formed unit replaces the original unit in subsequent comparisons until no further mergeable unit pairs are possible.

[0067] The account quantity filtering step is used to eliminate misidentifications in non-residential scenarios. It counts the number of unique broadband accounts within each merged unit. When the number of accounts exceeds the preset upper limit for the number of household accounts, the merged unit is removed entirely. The upper limit for the number of household accounts is set based on the statistical distribution characteristics of household broadband user accounts within the operating area. This upper limit should cover the account quantity range of the vast majority of real household units, excluding obviously abnormal aggregations. Through quantity filtering, non-residential scenarios such as multiple independent residents in shared apartments sharing the same splitter port, small and micro enterprises accessing the internet under the guise of household broadband, and incorrectly linked accounts for equipment in building common areas can be eliminated, ensuring that the final output of household broadband units physically corresponds to a real single household.

[0068] In one or more embodiments of the present invention, calculating the feature distinguishing weight based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in a target area includes: Let N be the total number of household broadband accounts in the target area, and df be the number of household broadband accounts containing the feature value. The feature discrimination weight idf is calculated using the following formula. weight : IDF weight = log[1 + N / (1 + df)] / log(1 + N) In one alternative implementation, the system may calculate the feature discrimination weight only for the feature set covered by the candidate home broadband units identified in step S104, instead of calculating it for all feature values ​​across the entire network.

[0069] The operation process of this implementation is as follows: After the identification of the home broadband unit is completed in step S104, the structured feature edges of each member account in all home broadband units that have been extracted in step S105 are extracted, and the deduplicated feature values ​​appearing in these feature edges are used to form a feature set to be calculated; weight calculation is performed on each feature value in the feature set to be calculated, and when the df value of the feature value in all home broadband accounts in the target area is counted, all home broadband accounts in the target area are still used as the statistical population, rather than only counting within the home broadband unit.

[0070] The principle behind this approach is that the df value reflects the prevalence of a certain feature across the entire region. Only based on global statistics can it be accurately determined whether a feature is a high-frequency pollution feature. If only accounts within a household broadband unit are used as the statistical scope, the distribution density of the feature value across the entire region cannot be determined, and the discrimination measure will lose its reference benchmark. By initiating weight calculation only within the feature set covered by candidate household broadband units, while maintaining the global statistical scope of df, the number of feature values ​​requiring weight calculation can be reduced while ensuring the accuracy of the discrimination measure, thus balancing computational efficiency and global discrimination.

[0071] In one or more embodiments of the present invention, the preset characteristic quality conditions include: When a local area network (LAN) MAC address exists and repeatedly appears in the same home broadband unit on multiple dates, the LAN MAC address is considered a stable physical characteristic. When the feature discrimination weight corresponding to a single non-physical feature exceeds the preset strong evidence threshold, the non-physical feature is regarded as strong evidence. When multiple non-physical features appear simultaneously and the feature discrimination weights corresponding to each non-physical feature exceed the preset effective feature threshold, the multiple non-physical features are used as combined evidence. The preset effective feature threshold is lower than the preset strong evidence threshold. When the feature discrimination weight corresponding to a feature value is lower than the preset effective feature threshold, the feature value is regarded as a low-weight common feature. Low-weight common features will not pass the feature quality condition due to the accumulation of quantity.

[0072] During the feature aggregation process, preset feature quality conditions are used to determine whether the combination of shared features within a home broadband unit constitutes credible evidence. The determination is made sequentially according to four levels: stable physical features, strong evidence, combined evidence, and low-weight common feature exclusion rules.

[0073] Stable physical characteristics have the highest priority. When a LAN Media Access Control (MAC) address appears repeatedly on multiple different dates in two or more member accounts within the same household broadband unit, that MAC address is considered a stable physical characteristic. As a hardware address for a network interface, the MAC address possesses global uniqueness and hardware-level stability. Its shared nature across multiple accounts and repeated occurrences over several days directly indicate at the physical level that the terminal devices corresponding to these accounts are connected to the same LAN, thus inferring that these accounts belong to the same physical household. The criterion of repeated occurrences over multiple days is used to exclude temporarily accessed guest devices.

[0074] When no stable physical feature is found, the process proceeds to determine strong evidence. If a non-physical feature exists whose feature discrimination weight exceeds a preset strong evidence threshold, that feature is considered strong evidence and used to support the aggregation independently. The preset strong evidence threshold is set at a high level, ensuring that only extremely scarce non-physical features within the target area meet the criteria.

[0075] When there are neither stable physical features nor strong evidence, the condition for combined evidence is determined. If multiple non-physical features are shared by multiple member accounts within a home broadband unit, and the discriminative weight of each feature exceeds a preset effective feature threshold, then these features are used as combined evidence to support aggregation. The preset effective feature threshold is lower than the preset strong evidence threshold. The logic is that while the discriminative power of a single ordinary effective feature may not reach the level of being used as independent evidence, when multiple effective features from independent sources simultaneously point to the same attribution relationship, the probability of accidental matching is significantly reduced, and the strength of the combined evidence can pass the quality threshold.

[0076] When the feature's discriminative weight is lower than the preset effective feature threshold, the feature is marked as a low-weight common feature. Its cumulative occurrence within a home broadband unit is not included in the criteria for judging the reliability of aggregation; that is, it does not pass the feature quality condition due to sheer quantity. This exclusion rule specifically targets high-frequency contamination features, preventing common cookies or generic device models from causing false aggregations due to quantity advantages.

[0077] In one or more embodiments of the present invention, calculating the overall confidence level includes: The topology base score is determined based on the topology hierarchy, and the topology adjustment coefficient is determined based on the consistency of the installation address and the number of accounts within the home broadband unit. The topology reliability is determined by the topology base score and the topology adjustment coefficient. Feature credibility is determined by the sum of the number of strong feature hits, the number of effective features, and the discrimination weight. Activity consistency is determined based on the number of active days in the past 7 days, the number of active days in the past 30 days, the most recent occurrence time, and changes in account status; The overall confidence level is determined by topological confidence, feature confidence, and activity consistency.

[0078] Topology reliability is calculated based on topology level, installation address consistency, and the number of accounts within a home broadband unit. Topology level reflects the splitter level used to identify the home broadband unit: units directly identified using a secondary splitter as a topology constraint have a higher base topology score because the secondary splitter has a close correspondence with the physical home; units identified by a primary splitter have a lower base topology score because the primary splitter has a larger coverage area and is more likely to correspond to multiple households. Installation address consistency adjusts topology reliability by comparing the matching degree of standardized installation addresses of each member account within the unit: the topology adjustment coefficient is maximized when all account addresses match completely; the topology adjustment coefficient decreases accordingly when addresses partially match or are missing. The base score is maintained when the number of accounts within the unit falls within the typical range for home broadband accounts; when it deviates from this range, the topology reliability is appropriately lowered. The final topology reliability is determined by adjusting the base topology score using the topology adjustment coefficient.

[0079] Feature credibility is determined based on the hit rate of strong features, the number of effective features, and the sum of the discriminative weights. Strong features include stable physical features and strong evidence features determined in step S107; each hit of a strong feature makes a significant positive contribution to feature credibility. The number of effective features is defined as the number of effective features that pass the feature quality threshold in step S107; a higher number of effective features indicates higher feature credibility. The sum of discriminative weights is the cumulative value of the discriminative weights corresponding to each effective feature; a high sum of weights indicates that the features participating in the aggregation have strong overall discriminative ability.

[0080] Activity consistency is calculated by comparing the behavioral patterns of each member account within a home broadband unit over a time dimension. This involves calculating the number of active days for each account within the statistical unit over the past seven days and the past thirty days within the statistical period, and then determining the degree of difference in the distribution of active days among the accounts; a smaller difference indicates a more similar activity pattern. The comparison of most recent occurrence times is achieved by calculating the maximum time interval between the most recent successful authentication times for each account; a smaller time interval indicates stronger activity synchronization among accounts. Account status changes are monitored by tracking whether each account experienced status change events such as suspension, reactivation, or account cancellation within the statistical period; higher synchronization of account status changes indicates stronger activity consistency.

[0081] In one or more embodiments of the present invention, generating a home broadband user profile based on comprehensive confidence level, service attributes in broadband service data, and existing service tags includes: The overall confidence level is compared with a preset threshold and a preset candidate threshold, and the preset threshold is higher than the preset candidate threshold. When the overall confidence level reaches a preset threshold, a formal profile is generated by combining business attributes and existing business tags. When the overall confidence level is lower than the preset threshold and reaches the preset candidate threshold, a profile to be reviewed is generated by combining business attributes and existing business tags. When the overall confidence level is lower than the preset candidate threshold, the intermediate process is retained or no image is output.

[0082] When generating a home broadband user profile, a tiered processing strategy is adopted for home broadband units based on the numerical range of the comprehensive confidence score. The comprehensive confidence score is calculated in step S108, and its numerical range, after normalization, falls within a preset numerical range. The system pre-sets two judgment thresholds of different sizes: a preset threshold and a preset candidate threshold, where the preset threshold is higher than the preset candidate threshold. These two thresholds divide the range of comprehensive confidence scores into three intervals, corresponding to the three processing methods: official profile, profile awaiting review, and no profile output.

[0083] When the overall confidence level of a home broadband unit reaches a preset threshold, it indicates that the unit's topology identification results, feature aggregation evidence, and member account activity consistency are all at a high level of credibility, and the conclusion that all broadband accounts within the unit belong to the same physical household is sufficiently supported. At this point, the system combines the service attributes and existing service tags in the broadband service data obtained in step S101 to generate a formal profile for the home broadband unit. The tags of each dimension in the formal profile are directly output as reliable results to the downstream business system, which can be used for operational decision-making scenarios such as fiber-to-the-room promotion, package upgrade recommendations, equipment replacement warnings, and converged marketing.

[0084] When the overall confidence level of a home broadband unit is lower than a preset threshold but reaches a preset candidate threshold, it indicates that the identification and aggregation results of the unit have a certain degree of credibility, but do not reach the confidence level required for direct application in business. This may indicate that some data is missing or the feature evidence is insufficient. In this case, the system still generates a verification profile for the unit based on business attributes and existing business tags. The verification profile contains the same tag dimensions as the official profile, but with additional verification markers for manual confirmation. The verification profile can be manually verified by operations personnel in conjunction with other information to confirm the accuracy of the home broadband unit's identification before deciding whether to put it into business use. This mechanism provides a flexible handling method when data quality fluctuates or feature evidence is at a critical state, avoiding the complete discarding of valuable profile information simply because the confidence level is slightly lower than the official threshold.

[0085] When the overall confidence level of a home broadband unit is lower than the preset candidate threshold, it indicates that the reliability of the unit's topology identification or feature aggregation results is insufficient. This may be due to reasons such as abnormal splitter connection relationships, lack of effective shared features among member accounts, or excessive differences in activity patterns. In this case, the system does not generate a profile output, but only retains the intermediate process data generated during this identification and aggregation process, including the member account list of the candidate home broadband unit, the extracted feature edge information, and the feature discrimination weights of each feature. This intermediate process data can be re-involved in the aggregation calculation after subsequent data updates or parameter adjustments, avoiding information loss due to incomplete data collection in a single instance or temporary network fluctuations.

[0086] In one or more embodiments of the present invention, generating a home broadband user profile includes generating at least one of a device profile, a behavior profile, a quality profile, a family structure profile, and a business scenario profile; The device profile includes the total number of terminals, mobile phones, televisions, and cameras; Behavioral profiles include active time periods, gaming preferences, video preferences, and external network access; The quality profile includes WiFi quality indicators and application quality indicators; Family structure profiles include the size of the family's terminals; The business scenario profile includes FTTR promotion compatibility, urgency of equipment replacement, need for package upgrades, opportunities for integrated marketing, and auxiliary identification tags for account change risks.

[0087] When generating household broadband user profiles, the system generates tags for each household broadband unit that meets the output conditions from five dimensions: device profile, behavior profile, quality profile, family structure profile, and business scenario profile. The data sources and generation methods for each dimension of the tags are as follows.

[0088] Device profiles describe the terminal hardware configuration within the home broadband unit. The total number of terminals is obtained by counting the number of deduplicated device identifiers associated with each member account in the terminal management data. The device identifier is used as the deduplication criterion to avoid duplicate counting of the same device due to multiple data collections on different dates. The number of mobile phones and televisions is obtained by classifying and counting the device type field associated with the device identifier. The device type field is labeled by the terminal management platform by parsing the device's Hypertext Transfer Protocol User Agent string or reading the device type registration information. The system counts separately based on the mobile phone category and television category in the device type label. The number of cameras is counted by counting the camera categories in the device type label, and also by combining the parsing results of real-time streaming protocol or session initiation protocol interaction signaling in the network behavior feature data to include camera-type devices identified by the network side.

[0089] Behavioral profiling describes the internet behavior characteristics of members within a home broadband unit. Active time periods are obtained by aggregating the authentication timestamps of each member's account in the authentication activity data at an hourly granularity, and counting the cumulative number of days with authentication records or traffic generation in each time period within a day to identify the distribution of high-activity time periods for the home broadband unit. Game preferences are derived by matching application names or traffic characteristics in the network behavior feature data with a game application feature database. This database maintains the protocol fingerprints and server-side Internet Protocol address ranges of known game applications; upon successful matching, the game type and percentage of game time are recorded. Video preferences are identified by comprehensively analyzing the media server domain name, video resolution switching mode, and content delivery network node in the video streaming protocol parsing results to determine the frequency and time distribution of video application usage within the home broadband unit. Inter-network access identifiers are determined by detecting whether cross-carrier network DNS resolution requests or cross-network server-side Internet Protocol address access records appear in the network behavior feature data. A corresponding identifier is generated when the home broadband unit exhibits inter-network resource access behavior within the statistical period.

[0090] The quality profile describes the network access quality of the home broadband unit. The Wi-Fi quality identifier is obtained by comprehensively evaluating wireless signal strength parameters, channel utilization, retransmission rate, and signal attenuation values ​​of access terminals reported by the optical network unit or home gateway in the terminal management data. A corresponding quality identifier is generated when these parameters exceed the preset normal operating range. The application quality identifier is derived by statistically analyzing the Transmission Control Protocol retransmission rate, application layer request-response latency, and Domain Name System (DNS) resolution failure rate collected by the deep packet inspection platform. An application quality identifier is generated when application layer experience indicators show a deteriorating trend.

[0091] The household structure profile is used to describe the scale characteristics of the household broadband unit. The scale of the household terminals is inferred by comprehensively analyzing the total number of terminals and the number of broadband accounts of members in the device profile. The ratio of the total number of terminals to the number of accounts is used as a reference indicator for the size of the household population and the level of terminal ownership, and based on this, the household broadband unit is divided into different scale categories.

[0092] The business scenario profile describes the suitability of the home broadband unit for operator marketing and customer retention. Fiber to the Room (FTTH) promotion suitability is assessed by comprehensively evaluating the equipment capability level of the optical network unit or home gateway, the quality of the wireless LAN, and the scale of home terminals. When the equipment capability is insufficient to match the current contracted bandwidth, or when there are many terminals and the wireless LAN quality is degraded, the home broadband unit is marked as a suitable candidate for FTTH promotion. Equipment replacement urgency is determined by analyzing the equipment model, age, and capability level of the optical network unit or home gateway. When the equipment model is outdated or the capability level does not match the current contracted bandwidth, an equipment replacement recommendation is generated. Package upgrade demand is determined by comparing the contracted bandwidth and actual traffic usage patterns of the home broadband unit. When the unit's traffic peaks repeatedly approach or reach the contracted bandwidth limit within the statistical period, a package upgrade recommendation is generated. Converged marketing opportunity is determined by analyzing the service types and existing service tags of each member account within the home broadband unit. When a member account only has broadband services and lacks mobile communication or Internet Protocol (IP) TV services, a converged marketing recommendation is generated. The account change risk identification tag is determined by monitoring the trend of member account activity, complaint records, and external network access behavior. When a member account shows signs such as a continuous decline in the number of active days, the existence of complaint tickets, or an increase in the frequency of external network access, a corresponding risk label is generated.

[0093] This invention also provides a home broadband user profile construction system based on network topology awareness, comprising: The data access module is used to obtain broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags for home broadband accounts; the optical access network resource data records the primary splitter identifier, secondary splitter identifier, and installation address, and the authentication activity data records the account activity status; The topology sensing module is used to identify home broadband units by prioritizing topology constraints for secondary optical splitters and using primary optical splitters as fallback topology constraints, combined with primary optical splitter identifiers, secondary optical splitter identifiers, installation addresses, and active status. The feature weight calculation module is used to extract feature values ​​from network behavior feature data and terminal management data, and calculate the feature discrimination weight based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area. The trusted aggregation module is used to aggregate feature values ​​within a home broadband unit based on feature discrimination weights and preset feature quality conditions, and to calculate the overall confidence level. The profile tag generation module is used to generate a profile of a home broadband user for home broadband units whose comprehensive confidence level reaches a preset threshold, based on the comprehensive confidence level, service attributes in broadband service data, and existing service tags.

[0094] This embodiment provides a home broadband user profile construction system based on network topology awareness. The system includes a data access module, an identifier standardization module, a topology awareness module, a feature extraction module, a feature weight calculation module, a trusted aggregation module, a profile tag generation module, a result traceability module, and an application output module.

[0095] The data access module establishes data interfaces with the broadband service operation support system, remote authentication dial-up user service system, optical access network resource management system, deep packet inspection platform, terminal management platform, and tag management system to acquire broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags for home broadband accounts. The data access module supports both scheduled batch and real-time streaming acquisition modes. During the acquisition process, source system identifiers and data statistics window date ranges are appended to various data types.

[0096] The identification standardization module connects to the output of the data access module and is used to standardize the format of the acquired raw data, filter null values, remove outliers, and process duplicate records. This module has a built-in broadband account format rule library, splitter encoding verification rules, address standardization mapping table, and terminal identifier format conversion rules. It performs standardization processing on broadband accounts, time fields, primary splitter identifiers, secondary splitter identifiers, installation addresses, terminal identifiers, cookies, and device identifiers, outputting cleaned and standardized data for use by the topology sensing module and feature extraction module.

[0097] The topology sensing module is used to identify home broadband units by prioritizing the secondary optical splitter as the topology constraint and using the primary optical splitter as the fallback constraint, combining the primary and secondary optical splitter identifiers, installation address, and activity status. When fine-grained topology data is available, the topology sensing module uses the secondary optical splitter identifier as the grouping key to divide home units. When fine-grained topology data is missing or invalid, it automatically falls back to the primary optical splitter identifier and combines address similarity calculations and account quantity filtering to output the final home broadband unit identification result.

[0098] The feature extraction module is used to extract feature values ​​from network behavior feature data and terminal management data, and organize the extracted features into structured feature edges that include fields such as account, feature type, feature value, occurrence date, source system, and number of times.

[0099] The feature weight calculation module is used to calculate the feature distinguishing weight based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area. This results in the automatic reduction of the weight of high-frequency feature values ​​and the automatic amplification of the weight of low-frequency stable feature values.

[0100] The trusted aggregation module is used to aggregate and determine the feature values ​​based on the feature values ​​and feature weights output by the feature extraction module and the feature weights output by the feature weight calculation module within the home broadband unit. It also combines preset feature quality conditions to calculate the overall confidence level of the home broadband unit by integrating three dimensions: topology confidence, feature confidence, and activity consistency.

[0101] The profile tag generation module is used to generate a family broadband user profile that includes device profile, behavior profile, quality profile, family structure profile and business scenario profile for family broadband units whose comprehensive confidence level reaches a preset threshold, by combining the business attributes in the broadband service data and existing business tags.

[0102] The results traceability module generates traceable output records for each home broadband unit. These records include the home unit identifier, member account, topology level, splitter encoding, feature source, feature weight, overall confidence level, profile label, scene score, data date, and hit rules. The hit rules record the triggering conditions and supporting evidence features for each profile label in structured fields, enabling downstream systems to trace and verify the profile results and optimize the rules.

[0103] The application output module is used to output the profiling results and traceability records to the tag service system, operation system, marketing recommendation system, network quality analysis system, or risk-assisted identification system in the form of application programming interface or data table. This invention also provides an electronic device, including a processor and a memory, wherein the memory stores program instructions, and the processor executes the program instructions to implement any of the above methods.

[0104] This invention also provides a computer-readable storage medium storing program instructions that, when executed by a processor, implement any of the methods described above.

[0105] Compared with existing technologies, this invention uses the physical topology of the optical access network as a rigid constraint for identifying home broadband units. In the case of broadband account changes, relocation, or multiple accounts coexisting, the secondary optical splitter is used as the priority topology constraint and the primary optical splitter as the fallback topology constraint. Home units are divided by combining the installation address and account activity status, which improves the granularity of profile analysis from a single account to the actual physical home, eliminating the interference of fragmented profiles at the account level on business judgment.

[0106] During the feature aggregation process, this invention calculates feature discrimination weights within the statistical scope of the entire network or target area, which automatically reduces the weight of common cookies and weak features shared by a large number of accounts, while low-frequency and stable physical features such as local area network MAC play a leading role in the aggregation, thus solving the problem of misaggregation and incorrect labeling caused by high-frequency polluting features.

[0107] Based on this, the present invention integrates topology credibility, feature credibility, and activity consistency to generate a comprehensive confidence level for home broadband units. When a preset threshold is reached, it integrates broadband service attributes and existing service tags to output a home broadband user profile. This makes the profile results both interpretable with physical topology support and capable of periodic updates, providing accurate and reliable home-level decision-making basis for operational scenarios such as FTTR promotion, package upgrades, equipment replacement, and churn warning.

[0108] The above is the overall concept of the present invention. For ease of understanding, the present invention also provides the following embodiments: Example 1: Overall Process of Building a Home Broadband Profile Within a business statistics period, the system obtains broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags for home broadband accounts through the data access module. Broadband service information provides service attributes such as broadband account, activation date, package, and contracted bandwidth; authentication activity data provides the first appearance time, most recent authentication time, and number of active days, constituting the account's activity status; optical access network resource data provides the primary splitter identifier, secondary splitter identifier, and installation address; network behavior characteristic data and terminal management data provide the terminal identifier, application access records, traffic characteristics, and quality-related parameters; and existing service tags provide account-level tag information accumulated during operation.

[0109] The system uses a standardization module to unify the format of accounts, dates, splitter codes, installation addresses, and terminal identifiers, filtering for null values, removing outliers, and handling duplicate records. After standardization, a basic account profile is constructed using the broadband account as the primary key. Fields include the effective date, most recent authentication time, number of active days in the last seven days, number of active days in the last thirty days, primary splitter identifier, secondary splitter identifier, installation address, contracted bandwidth, number of terminals, device type, application access records, and quality indicators. If the service activation date is missing from the broadband service data, the first occurrence date in the authentication activity data is used as the effective date to supplement it.

[0110] Subsequently, the topology-aware module identifies home broadband units based on network topology relationships: it prioritizes using secondary optical splitter identifiers as fine-grained topology constraints, grouping broadband accounts with the same secondary optical splitter identifier into the same candidate home broadband unit; when a secondary optical splitter identifier is missing or invalid, it uses the primary optical splitter identifier as the higher-level topology constraint for fallback identification. After generating candidate home broadband units, the candidate units are merged based on installation address similarity, and filtered by account activity status and group size to remove abnormal aggregations with more than a preset threshold of accounts, thus obtaining the official home broadband unit identification results.

[0111] Example 2: Adaptive Feature Weight Calculation The system extracts structured feature edges associated with accounts from network behavior feature data and terminal management data through a feature extraction module. Each feature edge includes the account, feature type, feature value, date of occurrence, source system, and number of occurrences. Feature types cover LAN media access control addresses, cookies, device identifiers, device models, device brands, domain names, application names, video application identifiers, game application identifiers, camera access records, and active time periods, etc.

[0112] For each feature value, the feature weight calculation module uses the total number N of all household broadband accounts participating in the statistics within the target area as a benchmark to count the number of unique accounts containing that feature value, df. According to the formula idf... weight The formula = log[1 + N / (1 + df)] / log(1 + N) calculates the feature discrimination weight of the feature value. The numerator measures the scarcity of the feature value logarithmically, while the denominator is a normalization factor, constraining the weight value to a range of zero to one. If a cookie identifier appears in the network behavior data of a large number of accounts, the df value is larger, and the calculated discrimination weight decreases; if a local area network access control address or stable terminal device identifier appears only in a very small number of accounts and is repeatedly observed on multiple dates, the df value is smaller, and the calculated discrimination weight increases. Through this adaptive weighting mechanism, the contribution of high-frequency contamination features in subsequent aggregation is automatically reduced, while the contribution of low-frequency stable features is automatically amplified, eliminating the need for manually preset feature filtering rules.

[0113] Example 3: Trusted Aggregation of Family Units The trusted aggregation module performs aggregation and determination of network characteristics of member accounts within the home broadband unit. The core operation of aggregation determination is as follows: the structured feature edges extracted from each member account in step S105 are summarized, and feature values ​​shared between at least two member accounts are extracted to form a shared feature set. For each feature value in the shared feature set, its feature discrimination weight is obtained from the calculation results of step S106, and whether the feature value constitutes valid aggregation evidence is determined item by item according to preset feature quality conditions.

[0114] The specific judgment process is as follows: First, the shared feature set is traversed to check if there is a physical feature whose feature type is a LAN media access control address and whose same address value has been observed on multiple dates. If such a feature exists, the media access control address is marked as a stable physical feature, and the feature credibility judgment of this unit passes directly without proceeding to the subsequent judgment process. If no stable physical feature exists, the non-physical features in the shared feature set are traversed, and the feature discrimination weight of each non-physical feature is compared with the preset strong evidence threshold. If the discrimination weight of a non-physical feature exceeds the preset strong evidence threshold, the feature is marked as strong evidence, and the feature credibility judgment of this unit passes. If neither of the first two conditions is met, non-physical features in the shared feature set whose discrimination weight exceeds the preset effective feature threshold are selected, and their number is counted. If the number reaches two or more, these features are collectively marked as combined evidence, and the feature credibility judgment of this unit passes. For feature values ​​in the shared feature set whose discrimination weight is lower than the preset effective feature threshold, they are marked as low-weight common features. Low-weight common features are not included in the count of effective features and are only recorded as auxiliary reference information.

[0115] After feature credibility determination, the credibility aggregation module further calculates the overall confidence level of the home broadband unit. The overall confidence level is determined by three components: topology credibility, feature credibility, and activity consistency. The quantification method for topology credibility is as follows: units identified using a secondary optical splitter as a topology constraint have a higher topology credibility value; units identified using a primary optical splitter with backtracking have a lower topology credibility value. Above these base values, adjustments are made based on the standardized text matching degree of the installation addresses of each account within the unit and whether the number of accounts within the unit falls within the typical home size range. Higher address matching degrees result in an adjustment coefficient closer to the upper limit, while a decrease in the adjustment coefficient occurs when the number of accounts deviates from the typical range. The quantification method for feature credibility is as follows: the pass rate in the feature credibility determination stage is used as the basic input. The feature credibility value is the highest when a stable physical feature is hit, the second highest when strong evidence is hit, and when combined evidence is hit, the feature credibility is categorized based on the sum of the number of effective features and the discrimination weight. The quantification method for activity consistency is as follows: Calculate the active days vector for each member account within the unit across two windows: the past seven days and the past thirty days. Determine the difference between the active days vectors of each account; the smaller the difference, the higher the activity consistency. Simultaneously, calculate the maximum time interval between the most recent successful authentication times for each account; the smaller the time interval, the closer the activity consistency adjustment coefficient is to the upper limit. Weighted summation of topology credibility, feature credibility, and activity consistency is then performed, with weighting coefficients set based on empirical reliability values ​​for each dimension in actual operation, to obtain the comprehensive confidence score for the home broadband unit.

[0116] After outputting the overall confidence score, the system performs tiered processing based on preset thresholds and preset candidate thresholds: when the overall confidence score reaches the preset threshold, the system outputs the official profile; when the overall confidence score is lower than the preset threshold but reaches the preset candidate threshold, the system outputs a profile to be reviewed and adds a manual confirmation mark; when the overall confidence score is lower than the preset candidate threshold, the system retains the shared feature set, the judgment results of each feature, and the intermediate variable of the overall confidence score generated during this aggregation process, and does not output a profile.

[0117] Example 4: Image Tagging and Scene Output The profile tag generation module generates multi-dimensional profiles for eligible home broadband units by combining service attributes from broadband service data and existing service tags. The output records include the home unit identifier, member account list, splitter code, installation address summary, topology level, topology credibility, feature credibility, activity consistency, overall confidence level, profile tags for each dimension, business scenario score, data date, and a description of the hit rules.

[0118] The device tag includes the total number of terminals, the number of mobile phones, the number of TVs, the number of cameras, the device's age / newness indicator, and the device capability level of the optical network unit or home gateway. The total number of terminals is calculated based on the device identifier for deduplication; the number of mobile phones and TVs is determined by classifying and counting them through the device type field in the terminal management data; the number of cameras is identified by combining the device type field and the real-time streaming protocol or session initiation protocol interaction signaling in the network behavior characteristic data; the device's age / newness indicator is divided into several intervals based on the interval between the device's first online time and the current time; the device capability level is classified according to the wireless protocol standards, port rates, and hardware configuration parameters supported by the device model.

[0119] Behavioral tags include high-activity periods, game user preferences, video user preferences, children's app access identifiers, and cross-network access identifiers. High-activity periods are derived by aggregating the active time of each member account in the authentication active data at an hourly granularity; game user preferences and video user preferences are derived by matching the application names in the network behavior feature data with the game application feature database and the video application feature database, respectively; children's app access identifiers are derived by matching the application names with the children's app classification knowledge base; and cross-network access identifiers are derived by detecting whether there are access records of cross-carrier Internet Protocol addresses in the network behavior feature data.

[0120] The quality label includes a Wi-Fi quality identifier, an application layer experience quality identifier, and a low-quality experience identifier. The Wi-Fi quality identifier is determined based on a combination of parameters reported in the terminal management data, such as signal strength, channel utilization, and retransmission rate. The application layer experience quality identifier is determined based on a combination of parameters, such as transmission control protocol retransmission rate, application request response latency, and domain name resolution failure rate. A low-quality experience identifier is generated when the above indicators exceed a preset degradation threshold.

[0121] The operational tags include tags for adaptability to fiber-to-the-room promotion, urgency of equipment replacement, need for package upgrades, opportunities for integrated marketing, and risk identification for account changes. Each operational tag is generated based on a comprehensive analysis of the unit's contracted bandwidth, equipment capability level, equipment age, peak traffic usage, types of subscribed services, and trends in member account activity. The matching rules for each tag are described using structured fields to record the combination of features or data conditions that trigger the tag, for manual review and rule iteration. The profiling results are distributed to the tag service system, operations system, marketing recommendation system, network quality analysis system, or risk identification system via the application output module.

[0122] Example 5: Device and Medium Implementation This embodiment provides an electronic device, including a processor and a memory. The memory stores program instructions, which, when executed by the processor, implement the aforementioned method for constructing a home broadband user profile. After reading the program instructions from the memory, the processor sequentially executes data access operations, identifier standardization operations, topology sensing operations, feature extraction operations, weight calculation operations, trusted aggregation operations, profile tag generation operations, and result output operations. The specific execution process of each operation is consistent with the corresponding steps in the aforementioned method embodiment.

[0123] This embodiment also provides a computer-readable storage medium storing program instructions. When these program instructions are executed by a processor of an electronic device, the processor completes the construction of a home broadband user profile according to the steps in the above method embodiment. The storage medium can be any one of a read-only memory, random access memory, disk storage medium, or optical disk storage medium.

[0124] The above provides a detailed description of a method, system, electronic device, and storage medium for constructing a home broadband user profile based on network topology awareness. 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, for those skilled in the art, 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 method for constructing home broadband user profiles based on network topology awareness, characterized in that, The method includes: Obtain broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags for the home broadband account; the optical access network resource data records the primary splitter identifier, secondary splitter identifier, and installation address, and the authentication activity data records the account activity status; The topology constraint prioritizes the secondary optical splitter and sets the primary optical splitter as the fallback topology constraint. The home broadband unit is identified by combining the primary optical splitter identifier, the secondary optical splitter identifier, the installation address, and the active status. Feature values ​​are extracted from the network behavior feature data and terminal management data, and feature discrimination weights are calculated based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area. Within the home broadband unit, feature values ​​are aggregated based on the feature discrimination weights and preset feature quality conditions to calculate the overall confidence level; A home broadband user profile is generated based on the comprehensive confidence level, the service attributes in the broadband service data, and the existing service tags.

2. The method according to claim 1, characterized in that, The topology constraint prioritizing the secondary optical splitter and the primary optical splitter as a fallback, combined with the primary optical splitter identifier, secondary optical splitter identifier, installation address, and active status identification of the home broadband unit, includes: Using the broadband account as the initial node, broadband accounts with the same secondary optical splitter identifier are grouped into the same candidate home broadband unit; When the secondary optical splitter identifier is missing or invalid, broadband accounts with the same primary optical splitter identifier will be grouped into the same candidate home broadband unit. The candidate home broadband units with similar installation addresses are merged to obtain the merged unit; The home broadband unit is obtained by removing units with more than a preset threshold of accounts from the merged units.

3. The method according to claim 2, characterized in that, The feature discrimination weight is calculated based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area, including: Let N be the total number of all household broadband accounts in the target area, and df be the number of household broadband accounts containing the feature value. The feature discrimination weight idf is calculated using the following formula. weight : idf weight = log[1 + N / (1 + df)] / log(1 + N)。 4. The method according to claim 3, characterized in that, The preset characteristic quality conditions include: When a local area network (LAN) MAC address exists and the LAN MAC address appears repeatedly in the same home broadband unit on multiple dates, the LAN MAC address is considered a stable physical characteristic. When the feature discrimination weight corresponding to a single non-physical feature exceeds the preset strong evidence threshold, the non-physical feature is regarded as strong evidence. When multiple non-physical features appear simultaneously and the feature discrimination weights corresponding to each non-physical feature exceed the preset effective feature threshold, the multiple non-physical features are used as combined evidence, and the preset effective feature threshold is lower than the preset strong evidence threshold. When the feature discrimination weight corresponding to the feature value is lower than the preset effective feature threshold, the feature value is regarded as a low-weight common feature. The low-weight common features do not pass the feature quality condition due to the accumulation of quantity.

5. The method according to claim 4, characterized in that, The calculation of the overall confidence level includes: The topology base score is determined based on the topology hierarchy, and the topology adjustment coefficient is determined based on the consistency of the installation address and the number of accounts within the home broadband unit. The topology reliability is determined by the topology base score and the topology adjustment coefficient. Feature credibility is determined by the sum of the number of strong feature hits, the number of effective features, and the discrimination weight. Activity consistency is determined based on the number of active days in the past 7 days, the number of active days in the past 30 days, the most recent occurrence time, and changes in account status; The overall confidence level is determined by the topological confidence level, the feature confidence level, and the activity consistency level.

6. The method according to claim 5, characterized in that, The step of generating a home broadband user profile based on the comprehensive confidence level, the service attributes in the broadband service data, and the existing service tags includes: The overall confidence level is compared with the preset threshold and the preset candidate threshold, wherein the preset threshold is higher than the preset candidate threshold; When the overall confidence level reaches the preset threshold, a formal profile is generated by combining the business attributes and the existing business tags. When the overall confidence level is lower than the preset threshold and reaches the preset candidate threshold, a profile to be reviewed is generated by combining the business attributes and the existing business tags. When the overall confidence level is lower than the preset candidate threshold, the intermediate process is retained or no image is output.

7. The method according to claim 6, characterized in that, The generated home broadband user profile includes at least one of the following: device profile, behavior profile, quality profile, family structure profile, and business scenario profile. The device profile includes the total number of terminals, mobile phones, televisions, and cameras; The behavioral profile includes active time periods, game preferences, video preferences, and access to other networks; The quality profile includes WiFi quality identifiers and application quality identifiers; The family structure profile includes the scale of family terminals; The business scenario profile includes FTTR promotion compatibility, urgency of equipment replacement, need for package upgrades, opportunities for integrated marketing, and auxiliary identification tags for account change risks.

8. A system for constructing home broadband user profiles based on network topology awareness, characterized in that, include: The data access module is used to obtain broadband service information, authentication activity data, optical access network resource data, network behavior characteristic data, terminal management data, and existing service tags of the home broadband account; the optical access network resource data records the primary splitter identifier, secondary splitter identifier, and installation address, and the authentication activity data records the account activity status; The topology sensing module is used to identify home broadband units by prioritizing topology constraints for secondary optical splitters and using primary optical splitters as fallback topology constraints, combined with the primary optical splitter identifier, secondary optical splitter identifier, installation address, and active status. The feature weight calculation module is used to extract feature values ​​from the network behavior feature data and terminal management data, and calculate the feature distinguishing weight based on the distribution of the number of household broadband accounts containing the same feature value among all household broadband accounts in the target area. A trusted aggregation module is used to aggregate feature values ​​within the home broadband unit based on the feature discrimination weights and preset feature quality conditions, and calculate the overall confidence level. The profile tag generation module is used to generate a home broadband user profile for home broadband units whose comprehensive confidence level reaches a preset threshold, based on the comprehensive confidence level, the service attributes in the broadband service data, and the existing service tags.

9. An electronic device comprising a processor and a memory, the memory storing program instructions, characterized in that, When the processor executes the program instructions, it implements the method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing program instructions, characterized in that, When the program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images