Property service quality evaluation method and system based on big data

Abstract

The invention discloses a property service quality evaluation method and system based on big data, particularly relates to the technical field of property management, and is used for solving the problem of poor property service quality evaluation optimization. According to the method, multi-source data in the property service process is collected, unified standardization and object identification anchoring are executed, a structured fusion event stream is generated, a global baseline and an object local baseline are established based on historical data, interval mapping and time attenuation are executed in a sliding time window, a deviation track is generated, and a real estate service is obtained. The method comprises the steps of reflecting dynamic changes of service quality, generating a comprehensive quality level through checking of time continuity, space connectivity and cross-modal consistency according to a dimension scoring mechanism in combination with an interval rule and continuous duration penalty, attributing abnormity through a set coverage method, adjusting a baseline window and evaluation parameters after rectification and recheck, and obtaining the quality of the service. Therefore, the intelligent level of service quality evaluation is improved, the service problem is identified, the manual intervention cost is reduced, and the service efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of property management, more particularly, the present application relates to a property service quality evaluation method and system based on big data. BACKGROUND

[0002] With the acceleration of urbanization, the quality of property services has become an important factor affecting the quality of life of residents. Property management not only needs to ensure the operation of daily facilities, but also needs to respond to residents' needs and handle various emergencies. Therefore, how to accurately evaluate the quality of property services and optimize based on data is the key to improving service level and operational efficiency.

[0003] The existing technology has the following problems: The current property service quality evaluation method mostly relies on manual experience or simple index statistics, lacks deep mining and comprehensive analysis of multi-dimensional data, and is difficult to truly reflect the complex dynamic relationship in the property service process. For example, the evaluation of service quality often only relies on single-dimensional feedback such as the processing time of repair work order or the complaint frequency of customers, without considering the influence of multi-factors such as task priority, resource allocation efficiency, equipment operating status and energy consumption. Especially in the process of equipment failure handling, it may only evaluate according to the time of repair work order, ignoring the operating status of the equipment and the number of available maintenance personnel, which may lead to overlapping resource allocation and multiple maintenance tasks being assigned to the same personnel, resulting in delayed maintenance response time, affecting service quality, leading to inaccurate evaluation results, and affecting subsequent resource scheduling and service optimization. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, the present application provides a property service quality evaluation method and system based on big data to solve the problem of poor property service quality evaluation optimization in the background art.

[0005] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0006] A property service quality evaluation method based on big data, comprising the following steps:

[0007] S1, performing field standardization, time alignment and object identification anchoring on multi-source service records generated by property operation, and generating a fusion event stream;

[0008] S2, establishing a global baseline and an object local baseline based on historical data, performing interval mapping and time decay on the fusion event stream in a sliding window, and obtaining the deviation trajectory of each index;

[0009] S3, constructing a conflict resolution graph, merging the objection records of the same object and the same period according to the source priority and time sequence proximity relationship, and outputting the minimum consistent subset and its evidence chain;

[0010] S4, sequentially performing time continuity, spatial connectivity and cross-modal consistency checking, generating counterfactual review tasks and writing review labels back to the fusion event stream when the checking fails;

[0011] S5, generating a dimension score sequence according to the deviation trajectory and the review label, according to the interval rule and the duration penalty, forming a comprehensive quality grade using the dominance relationship and sequential aggregation, and attributing the anomaly to the responsible unit by set covering selection;

[0012] S6, after the rectification and recheck closed loop is completed, adjusting the baseline window width and the checking order according to the error judgment sample proportion and the stability index, updating the evaluation parameters and outputting the quality grade and trend of the region and the object.

[0013] In a preferred embodiment, S1 includes:

[0014] Unify time zones and perform unit conversion and enumeration value alignment;

[0015] Time calibration is performed on the collection end, and the device clock drift is corrected according to the reference time source;

[0016] The composite identification anchor object is composed of region, building, unit, house, equipment and task;

[0017] When different sources point to the same object in the same period and the fields conflict, merge according to the priority order of the source trust label and record the merging basis.

[0018] In a preferred embodiment, S2 includes:

[0019] The global reference baseline interval is formed by the truncated mean or quantile;

[0020] The object local baseline is generated in the sliding time window;

[0021] Time decay is applied to the event sequence, the decay uses an exponential or piecewise linear function, the window length is adaptively adjusted according to the stability index, and the parameter effective time is recorded.

[0022] In a preferred embodiment, the interval mapping of S2 includes:

[0023] The hysteresis band is configured for each index, including an entry threshold and an exit threshold;

[0024] Monotonicity constraints and in-window smoothing are applied to the deviation trajectory;

[0025] When the trajectory fluctuates back and forth near the boundary of the hysteresis band, the previous determination state is maintained until the exit threshold is crossed.

[0026] In a preferred embodiment, S3 includes:

[0027] The conflict resolution graph is constructed with events as nodes and time-adjacent and semantically compatible edges as edges.

[0028] The conflict records are merged step by step according to the source priority table and the time-adjacent threshold, and the minimum consistent subset is output.

[0029] An evidence chain is generated for each merging result, which includes source identification, timestamp, semantic compatibility rule number and merging step number.

[0030] In a preferred embodiment, S4 includes:

[0031] The time continuity check determines whether the sequence is interrupted according to the maximum interval threshold;

[0032] The spatial connectivity check determines whether the event point set is connected based on the building and device topology graph;

[0033] The cross-modal consistency check compares each source record within the associated time window according to the event semantic mapping table;

[0034] When any check fails, a counterfactual review task is generated, which specifies the verification field, target data source and deadline condition, and after completion, the review label is written back to the fused event stream.

[0035] In a preferred embodiment, S5 includes:

[0036] The indicator value is mapped to a dimension score according to the interval rule, and a duration penalty is applied according to the cumulative active duration, which monotonically increases with the cumulative duration;

[0037] In the directed acyclic structure, sequential aggregation is performed in a predetermined order, and when the current sequence dimension does not meet the standard, a degradation rule is triggered, and according to the dominance relation table, the comprehensive quality level is derived from the dimension score and accompanied by a threshold trigger record.

[0038] In a preferred embodiment, the abnormality attribution of S5 includes:

[0039] The set of responsibility units that minimizes the number of covered unattributed abnormalities is selected in a set covering manner;

[0040] When there are the same number of set coverages, the responsibility units are selected from large to small according to the hierarchical number; if they are still the same, the one with the smallest sum of spatial envelope areas of the set is selected;

[0041] The responsibility units are defined according to the hierarchy of regions, buildings, devices and posts;

[0042] The responsibility unit identification, covered abnormality list and residual abnormality list are output.

[0043] In a preferred embodiment, S6 includes:

[0044] Adjust the baseline window width, threshold range, and verification order based on the proportion of incorrectly judged samples, hysteresis error, and variance stability.

[0045] When the proportion of sources failing the review exceeds a preset threshold, the review priority and sampling frequency of the sources are increased.

[0046] Add an effective timestamp to the updated assessment parameters and enable them in the next assessment cycle, while retaining historical parameters for retrospective auditing.

[0047] A big data-based property service quality assessment system, used to implement the aforementioned big data-based property service quality assessment method, includes:

[0048] The rehabilitation data acquisition module is used to collect electromyography, movement trajectory, and pressure distribution signals through sensors, and to perform data preprocessing to extract time-frequency features and generate multimodal feature data.

[0049] The dynamic representation module is used to construct a modal relationship graph, update and fuse electromyography, motion trajectory and pressure distribution features based on graph convolutional network, determine the correlation between modalities, and generate a unified dynamic motion representation of the scapula.

[0050] The rehabilitation plan verification module is used to generate a personalized baseline by combining the initial state and rehabilitation goal features, optimize the training plan according to the real-time deviation, and compare and analyze the actual movements of rehabilitation patients with the target movements to obtain the verification simulation information generated during the movement deviation detection process.

[0051] The rehabilitation adjustment module is used to analyze the verification simulation information and generate different signals, and to carry out corresponding rehabilitation management based on the generated signals.

[0052] The technical effects and advantages of this invention are as follows:

[0053] This invention collects multi-source service records generated during property operation and performs unified field standardization, time alignment, and object identification anchoring on these data to generate a structured fused event stream, ensuring data consistency and comparability. Based on this, a global baseline and object-specific local baselines are established based on historical data. A sliding time window is used to map and decay the event stream over time, generating a deviation trajectory for each indicator. This comprehensively reflects the dynamic changes in the quality of various services during the property service process. Furthermore, through checks on time continuity, spatial connectivity, and cross-modal consistency, the accuracy and completeness of the data are ensured, avoiding erroneous assessments due to missing or inconsistent data.

[0054] A dimensional scoring mechanism based on interval rules and duration penalties is adopted to conduct a detailed evaluation of each service quality indicator. A comprehensive quality level is formed through dominance relationships and sequential aggregation. Anomalies are attributed to the smallest responsible unit using a set coverage method, ensuring that each anomaly can be traced back to a specific responsible unit. In addition, after the rectification and re-inspection loop is completed, the baseline window width and verification order are automatically adjusted according to the proportion of erroneous judgment samples and stability indicators. The evaluation parameters are updated and the quality level and trend of the region and object are output, thereby improving the accuracy of the evaluation. It can also optimize operational decisions through data-driven methods, reduce the cost of manual intervention, and improve service efficiency and customer satisfaction. Attached Figure Description

[0055] Figure 1 This is a flowchart of a property service quality assessment method based on big data according to the present invention.

[0056] Figure 2 This is a schematic diagram of the structure of a property service quality assessment system based on big data according to the present invention. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] Example 1: As Figure 1 As shown, a big data-based method for assessing the quality of property services includes the following steps:

[0059] S1. Perform field standardization, time alignment, and object identifier anchoring on multi-source service records generated by property operations to generate a fused event stream;

[0060] S2. Based on historical data, establish a global baseline and object local baseline, perform interval mapping and time decay on the fused event stream within a sliding window, and obtain the deviation trajectory of each indicator.

[0061] S3. Construct a conflict resolution graph, merge dispute records of the same object at the same time according to source priority and temporal proximity, and output the minimum consistent subset and its evidence chain.

[0062] S4. Perform temporal continuity, spatial connectivity and cross-modal consistency checks in sequence. If the checks fail, generate a counterfactual review task and write the review label back to the fusion event stream.

[0063] S5. Based on the deviation trajectory and review labels, generate a dimensional scoring sequence according to interval rules and duration penalties, form a comprehensive quality level by using dominance relationships and sequential aggregation, and assign the anomaly to the responsible unit by set coverage selection.

[0064] S6. After the rectification and re-inspection loop is completed, adjust the baseline window width and verification order according to the proportion of incorrect judgment samples and stability indicators, update the evaluation parameters, and output the quality level and trend of the region and object.

[0065] S1. Perform field standardization, time alignment, and object identifier anchoring on the multi-source service records generated by property operations to generate a fused event stream. The specific implementation is as follows:

[0066] First, the system integrates multi-source service records generated by property operations, including task processing records, equipment operation records, energy consumption records, access control records, environmental monitoring records, and user feedback records. A standard field dictionary and enumeration lookup table are established, with dictionary version numbers and lookup table version numbers clearly marked. Field standardization is performed on the original fields from each source, and the status enumeration values ​​from the source are aligned and mapped to the system's standard values. Then, the time fields of all records are uniformly converted to the system's standard time zone while retaining the original time zone information and dictionary version number. A fused event stream is generated using the unified timestamp and spatial coordinates as indexes. Each event record includes at least the object identifier, occurrence time, occurrence location, source identifier, source trust label, key field list, original time, original coordinates, dictionary version number, compliance flag, and missing field flag. Missing fields are explicitly marked with null values ​​without inference or filling.

[0067] The system performs time alignment and device clock drift correction at the acquisition end. It configures a primary reference time source and a backup reference time source and sends time synchronization requests to the acquisition end at fixed intervals. Upon receiving the acquisition end's response, it records the current time of the reference time source and the device time reported by the acquisition end, recording the time difference between them as a raw offset. After collecting multiple raw offsets over several periods, it uses a method of removing extreme values ​​and taking the median to obtain the drift estimate for the acquisition end. This drift estimate is used to correct the time field of subsequent records reported by the acquisition end. For records with upload delays, the system simultaneously records the acquisition end's device time corrected by the drift estimate and the platform's receiving time, and calculates the upload delay estimate. When the upload delay estimate exceeds a threshold, the device time is used as the occurrence time and the platform's receiving time as the arrival time, and both are saved in the event log. After time synchronization, time alignment is performed according to the alignment granularity. When the occurrence time falls between two alignment scales, the event is assigned to the nearest alignment scale.

[0068] For each event record, object identification is anchored, forming a composite identifier in the order of region, building, unit, household, equipment, and task. A lower-level identifier can only be generated if the upper-level identifier exists. If an identifier at a certain level is missing, it is explicitly marked with a null value and is not replaced by other fields. A conflict detection and merging mechanism is established for multi-source records of the same object at the same time. The same object is identified by having identical composite identifiers. The same time period is determined by the merging window duration, centered on the occurrence time and using a symmetrical fixed window. When records from different sources but with the same object and time period have inconsistencies in the list of key fields, merging is performed based on the source trusted label and trusted label priority order table. If a record with higher priority gives a valid value in a certain key field, that value is used. If the record has a null value in that field, it is deferred to the next higher priority record until a valid value is obtained. If all sources have null values ​​in that field, the null value is retained in the merging result and a "requires review" mark is added.

[0069] It should be noted that the spatial coordinates use either the building's local coordinates or the globally universal coordinates, and the options and parameter effective time are fixed in the configuration. The alignment granularity is based on the larger of the shortest sampling interval in the multi-source service records and the clock accuracy of the processing system, and the parameter effective time is recorded.

[0070] S2. Based on historical data, establish a global baseline and object-specific local baselines. Within a sliding window, perform interval mapping and time decay on the fused event stream to obtain the deviation trajectory of each indicator. The specific implementation is as follows:

[0071] Based on the fusion event flow, a global baseline and a local baseline for each evaluation indicator are established. The global baseline is generated by taking the indicator samples of the same type of object in the historical period as input. The samples are first sorted by value according to the truncated mean method. After removing the extreme parts at both ends of the sequence, the lower and upper bounds of the reference interval are determined by the center position and distribution range of the remaining samples; or the specified quantiles are directly taken as the lower and upper bounds of the reference interval according to the quantile method.

[0072] Each metric specifies the baseline method used in the configuration and records the effective time of the parameters. The local baseline of the object is generated within the sliding time window. The initial window length is set by the configuration and is adaptively adjusted according to the stability index.

[0073] The stability index is determined by the amplitude of numerical fluctuations within the window and the number of direction switching times. The window length is increased when both the amplitude of fluctuations and the number of direction switching times increase, and decreased when both decrease. Time decay is applied to the event sequence within the window, offering two methods: exponential and piecewise linear. The exponential method reduces the weight of events as they become more distant from the present, while the piecewise linear method reduces the weight in segments over multiple time periods. The weight decay rate for both methods is specified in the configuration, and the effective time of the parameters is recorded. Events with lower source credibility labels are also weighted and reduced to minimize their impact on the local baseline.

[0074] Within the sliding time window, interval mapping is performed on the fused event stream to obtain the deviation trajectory. The interval mapping configures an entry threshold and an exit threshold for each indicator to form a hysteresis band. The entry threshold is used to determine the transition from normal to abnormal, and the exit threshold is used to determine the recovery from abnormal to normal.

[0075] When the index value after time decay first crosses the entry threshold, it is marked as an abnormal state. If the value fluctuates back and forth near the boundary of the hysteresis band in subsequent moments, the previous judgment state is maintained until the exit threshold is crossed and the state is changed.

[0076] Monotonicity constraints are applied to the deviation trajectory. If the direction is reversed several times in a window but the cumulative change is insufficient to support the reversal of direction, the existing direction is maintained and the trajectory is smoothed in an in-window smoothing manner. In-window smoothing is achieved by calculating the weighted average of adjacent time moments in a fixed-length inner window. The length of the inner window is less than the length of the sliding time window. The weights are set synchronously according to time decay and source credibility labels. After interval mapping and smoothing, the deviation trajectory consisting of normal and abnormal segments is output in chronological order, and the time points of each entry and exit are recorded.

[0077] The global baseline and object-specific local baseline, time decay parameters, sliding time window length, entry and exit thresholds, monotonicity constraints, and in-window smoothing configurations are all uniformly incorporated into parameter version management. All parameters include their effective date and are written into the evidence chain index of the event log. When the deviation trajectory repeatedly lingers near the hysteresis band boundary and eventually crosses the exit threshold, a boundary crossing event is generated and written into the fused event stream. This event includes the object identifier, occurrence time, baseline method used, current threshold, source credibility label weighting description, and window length value, ensuring that subsequent reviews can trace back the judgment process.

[0078] For example, in early July, the city issued a high temperature warning. The commercial podium of Community A saw an increase in customer traffic from afternoon to evening. Environmental monitoring records and energy consumption records were continuously poured into the fusion event stream. The system first used samples of the same business type in the same season of previous years to form a global baseline interval, and then generated a local baseline of the object with a sliding time window of one week.

[0079] Starting at 4 PM that day, after interval mapping and time decay within the sliding time window, the indoor comfort-related indicators had a greater weight for samples closer to the present than for earlier samples. The indicator values ​​exceeded the entry threshold and were marked as abnormal by the hysteresis band. Affected by the merchants' self-adjustment of fresh air, there were multiple fluctuations between 3 PM and 4 PM. However, since the hysteresis band maintained the previous judgment state, it was not frequently switched to normal. The system applied monotonic constraints to the deviation trajectory and smoothed it within the window, obtaining the continuous abnormal segment from 4 PM to before closing time, as well as the entry and exit time points.

[0080] S3. Construct a conflict resolution graph, merge dispute records of the same object within the same time period according to source priority and temporal proximity, and output the minimum consistent subset and its evidence chain. The specific implementation is as follows:

[0081] Using the merged event stream as input, a conflict resolution graph is constructed for objection records within the same time period. Objects are identified by composite identifiers, which consist of region, building, unit, household, equipment, and task sequence. The "same time period" is determined by the merged window duration, which is centered on the occurrence time and uses a symmetrical fixed window. Each event record is used as an event node, and edges are established according to two types of relationships.

[0082] The first type is temporal proximity relationship, which is established when the time difference between the occurrence of two event records is not greater than the temporal proximity threshold. The second type is semantic compatibility relationship, which is established when the values ​​and meanings of key fields are compared according to the semantic compatibility rule table. When two event records are not contradictory in terms of service object, event type, scope of impact and processing links and can jointly describe the same fact or the same handling process, a semantically compatible edge is established. Each edge records its relationship type, the threshold or rule number used, and the generation time.

[0083] Conflicting records are merged step by step according to the source priority table and the temporal proximity threshold, and the smallest consistent subset is output. The source priority table is mapped from the source trust label. High trust sources have high priority and low trust sources have low priority.

[0084] The merging process is carried out on a per-object and per-time-period basis. First, the event nodes are sorted from high to low according to their source priority. Within the same priority, they are sorted according to the order of occurrence time. The event node with the highest current priority and the earliest time is selected as the seed. In the conflict resolution graph, the adjacent nodes that satisfy both the temporal proximity relationship and the semantic compatibility relationship with the seed are searched and included in the set to be merged.

[0085] If a conflict occurs in the list of key fields, the record with the higher source priority is used as the merged value. If two records have the same source priority and conflicting values, the record whose occurrence time is closer to the center of the merge window duration is used. If they still cannot be distinguished, a null value is retained and marked as requiring review. After a merge is completed, the node to be merged is removed from the candidate set and the merge step number is recorded. Then, the remaining nodes are processed using the same rules until the candidate set is empty.

[0086] The minimum consistent subset is defined as the smallest set of events that can support all the required fields of the critical field list required by the object during a given time period, provided that the semantic compatibility and temporal proximity constraints are satisfied. If any event is removed from this set, at least one required field will be missing or any established compatibility relationship will be destroyed, thus no longer satisfying the consistency requirements.

[0087] For each merge result, an evidence chain is generated and bound to the merged event record. The evidence chain includes the source identifier, occurrence timestamp, semantic compatibility rule number used, temporal proximity threshold used, event record number involved in the merge, source and reason for the value of each field in the key field list, merge step number, and parameter effective time.

[0088] The source identifier is used to point to a specific data source, the occurrence timestamp is used to restore the time sequence, the semantic compatibility rule number is used to indicate the rule entry on which it is based, and the merging step number is used to indicate the processing order of this merging within the entire time period. At the same time, the temporal proximity threshold, the merging window duration and the source priority table are written into the parameter version management and the parameter effective time is recorded to ensure that the merging process can be restored based on the evidence chain and parameter version in the future.

[0089] The minimum consistent subset after conflict resolution and evidence chain generation, along with the merged event records, is written back to the fused event stream as input for subsequent temporal continuity verification, spatial connectivity verification, and cross-modal consistency verification, and as a reliable data source for the scoring stage interval rules and duration penalties.

[0090] S4. Perform temporal continuity, spatial connectivity, and cross-modal consistency checks sequentially. If a check fails, generate a counterfactual review task and write the review tag back to the fusion event stream. The specific implementation is as follows:

[0091] The event sequence is established by taking the fused event stream as input and by object identifiers, which are also composed of region, building, unit, household, equipment, and task.

[0092] Configure a maximum interval threshold for each type of metric and each type of event. The maximum interval threshold is derived from the historical statistics or operational procedures of the global baseline interval and the local baseline of the object. The effective time of the parameter is written when configuring.

[0093] It should be noted that the maximum interval threshold is determined by the historical statistical values ​​of the global baseline interval and the local baseline of the object, and the more stringent one is taken as the threshold in combination with the longest allowable window time of the operation procedure, and the parameter effective time is recorded; the event semantic mapping table consists of three parts: event type comparison, status code evolution relationship and numerical tolerance. The numerical tolerance is determined by the upper and lower bounds of the global baseline interval and the stability index of the local baseline of the object. The version number of the mapping table and the parameter effective time are recorded together; the building and equipment topology map consists of building space nodes, equipment installation nodes and allowable reachable edges. The map version number and generation time are written into the configuration and the parameter effective time is updated with changes.

[0094] The process of time continuity verification is as follows:

[0095] Sort events under the same object identifier from earliest to latest by occurrence time, calculate the time difference between two adjacent events, and determine that the sequence is interrupted when any time difference is greater than the corresponding maximum interval threshold, and record the start and end time of the interruption, the maximum interval threshold triggered, and the effective time of the parameters used.

[0096] If all time differences within the sequence are not greater than the maximum interval threshold, the time continuity check is deemed to have passed.

[0097] For potential upload delays, both device time and platform reception time are taken into account, and the occurrence time after the aforementioned time alignment and drift correction is used as the judgment benchmark to avoid misjudgment caused by network latency.

[0098] After completing the temporal continuity check, perform spatial connectivity and cross-modal consistency checks:

[0099] Spatial connectivity verification is based on the building and equipment topology diagram, which consists of building space nodes and equipment installation nodes. Edges represent allowed reachability or logical associations. The version of the graph and the effective time of the parameters are recorded in the configuration.

[0100] The location of the recorded event is mapped to a sequence of nodes in the topology graph. If the entire sequence of nodes is located in the same connected component and there are accessible edges between adjacent nodes, the spatial connectivity check is considered to have passed. If any two adjacent events have no path to reach them in the topology graph or the path is marked as not accessible, the check is considered to have failed and the first disconnected node pair and the version of the topology graph used are recorded.

[0101] Cross-modal consistency verification is based on comparing an event semantic mapping table with an associated time window. The associated time window is centered on the occurrence time of the reference event, and its length is given by the configuration, while also recording the parameter effective time. The system extracts records falling into the associated time window from different sources and checks the correspondence between event types, the evolution of status codes, and the numerical tolerance range item by item according to the event semantic mapping table. The tolerance range is determined by the global baseline interval and the object's local baseline.

[0102] If any source record and reference event do not meet the mapping and tolerance requirements in terms of type, status, or value, the cross-modal consistency check is deemed to have failed, and the failing field, source identifier, and rule number are recorded; if all requirements are met, the check is deemed to have passed.

[0103] When any of the time continuity check, spatial connectivity check, or cross-modal consistency check fails, the system automatically generates a counterfactual review task and specifies three items in the task. The first item is the field to be verified, which must include at least the time of occurrence, the location of occurrence, the event type, the status code, and the indicator value. The list of key fields to be verified can be supplemented according to the object identifier.

[0104] The second item is the target data source, which is listed in descending order of the source trust label priority: original device logs, edge acquisition records, business system audit logs, and manual on-site verification. The system obtains and forms verification evidence independent of the original records in sequence.

[0105] The third item is the deadline, which includes the latest completion time, minimum sample size, and minimum observation duration to ensure sufficient evidence. After the review is completed, a review label is generated and written back to the fusion event stream. The review label includes whether the review result is passed, the calibrated field value, the evidence source identifier, the collection time, the processing personnel identifier, the effective time of the parameters used, and the rule number. At the same time, the index of the review label is added to the evidence chain for traceability. When the review is passed, the system updates the corresponding event record and removes the failure mark for that item. When the review fails, the failure mark is retained and used as a trigger condition for duration penalty and anomaly attribution in the subsequent scoring stage.

[0106] S5. Based on the deviation trajectory and review labels, generate a dimensional scoring sequence according to interval rules and duration penalties. Use dominance relationships and sequential aggregation to form a comprehensive quality level, and use set coverage selection to assign anomalies to responsible units. The specific implementation is as follows:

[0107] Using the deviation trajectory and verification labels as input, the values ​​of each indicator are first mapped to dimensional scores according to interval rules. The interval rules are determined by a hysteresis band consisting of an entry threshold and an exit threshold. The entry threshold is used to determine whether the condition changes from normal to abnormal, and the exit threshold is used to determine whether the condition recovers from abnormal to normal.

[0108] For time periods in an abnormal state, the cumulative activation duration is calculated. The cumulative activation duration is the total duration during which the indicator remains in an abnormal state within the evaluation period. Time periods during which the review label passes are not included in the cumulative activation duration. The dimension score is deducted according to the duration penalty rule. The duration penalty increases monotonically with the cumulative activation duration. The deduction step size and deduction limit are given in the configuration and the effective time of the parameters is recorded.

[0109] It should be noted that the dominance relationship table is jointly formulated through offline verification and operational procedures, which clarifies the dominance and downgrade conditions from the scoring range of each dimension to the overall quality level, and is stored in the configuration in a versioned manner for retrospective purposes; the deduction step size and deduction limit of the duration penalty are verified by replaying historical deviation trajectories and user feedback records, so that the downgrade intensity can reflect the impact of the cumulative activation time and avoid excessive deduction, and the parameter effective time is recorded in the configuration and included in the evidence chain index.

[0110] After completing the above processing, the dimensional score sequences are output in time order. The dimensions include service response, facility operation, environmental quality and customer perception. Each dimensional score sequence retains the time point and judgment basis for each entry and exit to ensure traceability.

[0111] In a directed acyclic structure, sequential aggregation is performed in a preset order and a comprehensive quality level is formed based on the dominance relationship table. The directed acyclic structure uses dimensions as nodes and pre-judgment and post-aggregation as edges. The preset order is determined by the configuration and the effective time of the parameters is recorded. During sequential aggregation, the first-order dimension is processed first. If the standard is not met, a degradation rule is triggered. The degradation rule limits the upper limit of the comprehensive quality level to the upper limit level set for that dimension in the dominance relationship table and records the threshold trigger information.

[0112] Subsequently, subsequent dimensions are processed sequentially. After each dimension is processed, the range of the currently achievable comprehensive quality level is updated according to the dominance relationship table. When any dimension triggers a threshold, the corresponding upper limit or deduction constraint is added. The dominance relationship table clarifies the dominance relationship between the scoring range of each dimension and the comprehensive quality level, as well as the generation conditions of the threshold trigger record. The threshold trigger record includes the triggering dimension, the threshold used, the trigger time, and the parameter effective time. After the sequential aggregation is completed, the comprehensive quality level and trigger record are output, which are used to generate trends and suggestions and serve as input conditions for anomaly attribution.

[0113] To attribute anomalies and output responsible units, the process begins by extracting a set of anomalies that have not yet been attributed from the deviation trajectory and dimensional scoring sequence. An anomaly consists of an object identifier, indicator name, anomaly start and end time, associated dimensions, and location range. Responsibility unit levels are defined by region, building, unit, household, equipment, and job position, and level numbers are set: region number 1, building number 2, unit number 3, household number 4, equipment number 5, and job position number 6. The coverage relationship between candidate responsibility units and anomalies is determined based on the responsibility mapping table. Coverage is considered established when the location of the anomaly is within the management scope of the responsibility unit and the anomaly type is the responsibility of that unit.

[0114] The set of responsibility units is selected by using a set coverage method. The goal is to minimize the number of selected responsibility units while covering all unattributed anomalies. When there are multiple sets of responsibility units with the same number of selected units, the set with the larger sum of hierarchical numbers is selected. If the sum of hierarchical numbers is still the same, the set with the smallest sum of spatial envelope area is selected. The spatial envelope area is the smallest convex polygon area that can contain the management scope of the responsibility unit, and the area unit is square meters.

[0115] After the selection is completed, the responsible unit identifier, the covered anomaly list, and the residual anomaly list are output. The residual anomaly list is used to trigger the next round of review or rule update, and the attribution results and the comprehensive quality level are written into the evidence chain index.

[0116] For example, during the morning rush hour on weekdays, the elevator in Unit 2 of Building B in the community repeatedly restarted after a short period of shutdown. The equipment operation records and user feedback records have passed the consistency check and generated a verification label. Based on the deviation trajectory and the verification label, the elevator availability rate and service response time are mapped to the dimension score according to the interval rules. The cumulative activation time is calculated for abnormal periods, and the dimension score is deducted according to the duration penalty rule. The deduction step size and deduction limit are executed according to the configuration and the effective time of the parameters is recorded.

[0117] Subsequently, sequential aggregation is performed in the directed acyclic structure according to a preset order. First, the facility operation dimension is processed. If the standard is not met, a downgrade rule is triggered and the upper limit of the comprehensive quality level is limited to the upper limit of the level set for that dimension in the dominance relationship table. Then, the service response dimension and customer perception dimension are processed and the comprehensive quality level range is updated according to the dominance relationship table. At the same time, a threshold trigger record is generated, which includes the trigger dimension, usage threshold, trigger time and parameter effective time.

[0118] The selection process uses a set-based coverage approach within the set of responsibility units. Responsibility units are defined hierarchically by region, building, unit, household, equipment, and job position, with hierarchical numbers assigned. The goal is to minimize the number of selected responsibility units while covering all unattributed anomalies. If multiple sets have the same number of selected units, the set with the larger sum of hierarchical numbers is selected. If they are still the same, the set with the smallest sum of spatial envelope areas is selected. The final output includes equipment identifiers and maintenance team identifiers as responsibility unit identifiers. A list of covered anomalies and a list of residual anomalies are also output, and these are written into the evidence chain index along with the overall quality level.

[0119] S6. After the rectification and re-inspection loop is completed, adjust the baseline window width and verification order according to the proportion of incorrect judgment samples and stability indicators, update the evaluation parameters, and output the quality level and trend of the region and object. The specific implementation is as follows:

[0120] After the rectification and re-inspection loop is completed, the fusion event flow, deviation trajectory, dimension scoring sequence and review label of the previous evaluation cycle are statistically analyzed, and the proportion of incorrect judgment samples, hysteresis error and variance stability are calculated.

[0121] The numerator of the determination sample proportion is the number of determination samples overturned by counterfactual verification, and the denominator is the number of determination samples that participated in the determination and entered the verification closed loop. The samples are uniquely identified by object identifier and indicator name and checked by evidence chain index. The hysteresis error is the average value of the state switching delay time at the boundary of the hysteresis band. Specifically, for each switching event near the entry threshold and exit threshold, two time points are recorded: one is the time point when the switch should be made according to the assumption of no hysteresis, and the other is the time point when the switch is actually made using the hysteresis band rule. The time difference between the two is one delay time. The hysteresis error is obtained by averaging the values ​​of the objects and indicators within the evaluation period.

[0122] Variance stability is a measure of the stability of an index value's fluctuations over multiple sliding time windows. Specifically, the variance of the index value is calculated over several consecutive sliding time windows, and then the dispersion of these variances is calculated. The smaller the dispersion, the better the variance stability. The stability threshold given in the configuration is used to determine whether the variance stability passes or fails, and the effective time of the parameter is recorded.

[0123] Based on the above three metrics, the baseline window width, threshold range, and verification order are adjusted. When the proportion of incorrectly judged samples exceeds the threshold, the baseline window width is expanded first, and the threshold range is appropriately expanded. Specifically, the entry threshold is increased and the exit threshold is decreased simultaneously to widen the hysteresis band, thereby reducing misjudgments caused by frequent out-of-bounds errors. When the proportion of incorrectly judged samples is below the threshold and the variance stability fails, the baseline window width is expanded without changing the threshold range to improve the stability of the local baseline. When the variance stability passes but the hysteresis error remains positive for a long time and exceeds the allowable range in the configuration, the threshold range is narrowed while keeping the baseline window width unchanged. Specifically:

[0124] Raising the entry and exit thresholds narrows the hysteresis band, thereby reducing delayed switching. The adjustment of the verification order follows the principle of prioritizing verification items with high failure rates. The failure rates of time continuity verification, spatial connectivity verification, and cross-modal consistency verification are statistically analyzed. The verification item with the highest failure rate is placed first in the next evaluation cycle, and the reasons for the adjustment and the effective time of the parameters are recorded in the evidence chain.

[0125] Source-level adaptive governance will be implemented based on the proportion of data sources that fail verification.

[0126] The numerator of the failure rate is the number of records from that source that failed the counterfactual review within the current assessment period, and the denominator is the number of records from that source that entered the review loop. When this rate exceeds a preset threshold, the review priority of that source in the source priority table is increased, and the sampling frequency of that source is also increased. Increased review priority means that the review action for that source will be triggered first during conflict resolution and consistency verification. Increased sampling frequency means that the data sampling interval for that source will be shortened and the number of samples will be increased in the next assessment period. All adjusted baseline window widths, threshold ranges, verification orders, source priorities, and sampling frequencies are timestamped and enabled in the next assessment period.

[0127] It should be noted that the thresholds involved in the embodiments can be determined according to specific scenarios and needs.

[0128] This invention collects multi-source service records generated during property operation and performs unified field standardization, time alignment, and object identification anchoring on these data to generate a structured fused event stream, ensuring data consistency and comparability. Based on this, a global baseline and object-specific local baselines are established based on historical data. A sliding time window is used to map and decay the event stream over time, generating a deviation trajectory for each indicator. This comprehensively reflects the dynamic changes in the quality of various services during the property service process. Furthermore, through checks on time continuity, spatial connectivity, and cross-modal consistency, the accuracy and completeness of the data are ensured, avoiding erroneous assessments due to missing or inconsistent data.

[0129] A dimensional scoring mechanism based on interval rules and duration penalties is adopted to conduct a detailed evaluation of each service quality indicator. A comprehensive quality level is formed through dominance relationships and sequential aggregation. Anomalies are attributed to the smallest responsible unit using a set coverage method, ensuring that each anomaly can be traced back to a specific responsible unit. In addition, after the rectification and re-inspection loop is completed, the baseline window width and verification order are automatically adjusted according to the proportion of erroneous judgment samples and stability indicators. The evaluation parameters are updated and the quality level and trend of the region and object are output, thereby improving the accuracy of the evaluation. It can also optimize operational decisions through data-driven methods, reduce the cost of manual intervention, and improve service efficiency and customer satisfaction.

[0130] Example 2: A property service quality assessment system based on big data, such as Figure 2 As shown, it specifically includes:

[0131] The property operation data module is used to perform field standardization, time alignment, and object identification anchoring on multi-source service records generated by property operation, and to generate a fused event stream;

[0132] The event deviation analysis module is used to establish a global baseline and a local baseline of the object based on historical data, and to perform interval mapping and time decay on the fused event stream within a sliding window to obtain the deviation trajectory of each indicator.

[0133] The conflict priority partitioning module is used to construct a conflict resolution graph, merge the dispute records of the same object at the same time according to the source priority and temporal proximity, and output the minimum consistent subset and its evidence chain.

[0134] The verification and determination module is used to sequentially perform temporal continuity, spatial connectivity and cross-modal consistency checks. If the check fails, a counterfactual verification task is generated and the verification label is written back to the fusion event stream.

[0135] The anomaly attribution module is used to generate a dimensional score sequence based on the deviation trajectory and review label, according to interval rules and duration penalties, to form a comprehensive quality level by using dominance relationship and sequential aggregation, and to assign the anomaly to the responsible unit by set coverage selection.

[0136] The quality assessment module is used to adjust the baseline window width and verification order based on the proportion of erroneous judgment samples and stability indicators after the rectification and re-inspection loop is completed, update the assessment parameters, and output the quality level and trend of the region and object.

[0137] The above formulas are all dimensionless calculations. Dimensionless calculations can be performed using various methods such as standardization, which will not be elaborated here. The formulas are derived from software simulations based on a large amount of collected data, and the preset parameters in the formulas can be set by those skilled in the art according to the actual situation.

[0138] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, ATA hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. The semiconductor medium can be a solid-state ATA hard disk.

[0139] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0140] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0141] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0142] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0143] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0144] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A big data-based property service quality evaluation method, characterized in that, The method comprises the following steps: S1, performing field standardization, time alignment and object identification anchoring on multi-source service records generated by property operation to generate a fusion event stream; S2, establishing a global baseline and an object local baseline based on historical data, performing interval mapping and time decay on the fusion event stream in a sliding window to obtain deviation trajectories of each index; S3, constructing a conflict resolution graph, merging conflicting records of the same object and the same time period according to source priority and time sequence proximity, and outputting a minimum consistent subset and an evidence chain thereof; S4, sequentially performing time continuity, spatial connectivity and cross-modal consistency checks, generating counterfactual review tasks when the checks fail, and writing review labels back to the fusion event stream; S5, generating a dimension score sequence according to the deviation trajectory and the review label, according to the interval rule and the duration penalty, forming a comprehensive quality level by using the dominance relationship and sequential aggregation, and attributing the anomaly to the responsible unit by set coverage selection; S6, after the rectification and re-examination closed loop is completed, adjusting the baseline window width and the checking order according to the error judgment sample proportion and the stability index, updating the evaluation parameters and outputting the quality level and the trend of the region and the object. 2.The property service quality evaluation method based on big data according to claim 1, characterized in that: S1 comprises: Unifying time zones, performing unit conversion and enumeration value alignment; Calibrating the time of the collection end, correcting the device clock drift according to the reference time source; Anchoring objects with region, building, unit, household, equipment and task as composite identifiers; When different sources point to the same object and the fields conflict in the same time period, merging is performed according to the priority order of the source credibility label and the merging basis is recorded. 3.The property service quality evaluation method based on big data according to claim 1, characterized in that: S2 comprises: Forming a global reference baseline interval with a truncated mean or quantile; Generating an object local baseline in a sliding time window; Applying time decay to the event sequence, using an exponential or piecewise linear function, adaptively adjusting the window length according to the stability index, and recording the parameter effective time.

4. The method of claim 3, wherein: The interval mapping of S2 comprises: Configuring an entry threshold and an exit threshold for each index to form a hysteresis band; Applying monotonicity constraints and window smoothing to the deviation trajectory; When the trajectory fluctuates near the boundary of the hysteresis band, the previous determination state is maintained until the exit threshold is crossed.

5. The method of claim 1, wherein: S3 comprises: Building a conflict resolution graph with events as nodes and time sequence proximity and semantic compatibility as edges; Merging conflicting records according to the source priority table and the time sequence proximity threshold, and outputting a minimum consistent subset; Generating an evidence chain for each merged result, which includes source identification, timestamp, semantic compatibility rule number and merge step number.

6. The method of claim 1, wherein: S4 comprises: The time continuity check is to determine whether the sequence is interrupted according to the maximum interval threshold; The spatial connectivity check is to determine whether the event point set is connected based on the building and equipment topology graph; The cross-modal consistency check is to compare the records of each source in the associated time window according to the event semantic mapping table; When any check fails, generate a counterfactual review task, specify the fields to be verified, the target data source and the deadline, and write the review label back to the fusion event stream after completion.

7. The method of claim 1, wherein: S5 comprises: Mapping the index value to a dimension score according to the interval rule, and applying a duration penalty according to the cumulative active duration, which monotonically increases with the cumulative duration; In the directed acyclic structure, the sequential aggregation is performed in the preset order, the degradation rule is triggered when the current sequence dimension does not meet the standard, and the comprehensive quality level is derived according to the score of each dimension and the threshold trigger record is attached according to the dominance relationship table. 8.The property service quality evaluation method based on big data according to claim 7, characterized in that: The abnormality attribution of S5 includes: Selecting a set of responsibility units that minimizes the number of covered unattributed abnormalities in a set covering manner; When there are the same number of set coverings, selecting the responsibility units in descending order of hierarchical number; if they are still the same, selecting the one with the smallest sum of spatial envelope area of the set; The responsibility units are defined in the hierarchy of area, building, equipment and post; Output the responsibility unit identifier, the covered abnormality list and the residual abnormality list. 9.The property service quality evaluation method based on big data according to claim 8, wherein S6 It includes: Adjusting the baseline window width, threshold interval and verification order according to the error judgment sample proportion, hysteresis error and variance stability; When the proportion of failed source review exceeds the preset threshold, the review priority and sampling frequency of the source are improved; Adding an effective timestamp to the updated evaluation parameters and enabling them in the next evaluation period, while retaining historical parameters for retrospective audit. 10.A big data-based property service quality evaluation system for implementing the big data-based property service quality evaluation method of any one of claims 1-9, characterized in that, It includes: A property operation data module for performing field standardization, time alignment and object identification anchoring on multi-source service records generated by property operation, and generating a fusion event stream; An event deviation analysis module for establishing global and local baselines based on historical data, performing interval mapping and time decay on the fusion event stream within a sliding window, and obtaining deviation trajectories for each indicator; A conflict priority division module for constructing a conflict resolution graph, merging conflicting records of the same object and time period according to source priority and time sequence proximity, and outputting the smallest consistent subset and its evidence chain; A review determination module for sequentially performing time continuity, spatial connectivity and cross-modal consistency verification, generating counterfactual review tasks when the verification fails, and writing review labels back to the fusion event stream; An abnormality attribution allocation module for generating a dimension score sequence according to the deviation trajectory and review label, forming a comprehensive quality level using dominance relationship and sequential aggregation, and attributing the abnormality to the responsibility unit using set covering selection; A quality evaluation module for adjusting the baseline window width and verification order according to the error judgment sample proportion and stability indicators after the rectification and re-inspection closed loop is completed, updating the evaluation parameters and outputting the quality level and trend of the area and object.

Images