A performance review relationship generation method and device, electronic equipment and storage medium

Abstract

This application provides a method, apparatus, electronic device, and storage medium for generating performance review relationships. The method includes: parsing and obtaining the participants in the current performance review from a pre-constructed enterprise organizational graph based on preset assessment conditions; for each participant, selecting target reviewers from the enterprise organizational graph based on a set of review rules bound to the review process nodes of the current performance review, and constructing review relationships with the corresponding participants; aggregating all review relationships into an initial set of review relationships; performing multi-dimensional anomaly detection on the initial set of review relationships to identify anomalies; generating corresponding target repair schemes based on the types of anomalies, and repairing the anomalies based on the target repair schemes to obtain the final set of review relationships. This invention significantly improves the efficiency and accuracy of review relationship configuration, while reducing the high cost of manual maintenance through automatic repair.

Description

Technical Field

[0001] This invention relates to the field of intelligent computing technology for enterprise organizational data, and more specifically, to a method, apparatus, electronic device, and storage medium for generating performance review relationships. Background Technology

[0002] In modern enterprise human resource management, performance appraisal, as the core mechanism for talent evaluation and motivation, relies heavily on a scientific and comprehensive design of review relationships for its fairness and effectiveness. Traditional performance systems are typically based on a simple direct reporting structure, with evaluations completed by an employee's sole direct supervisor. However, as corporate organizational structures become increasingly complex, matrix management, dotted-line reporting, project-based collaboration, and cross-departmental staffing have become the norm, leading to multi-dimensional, dynamic, non-unique, and even temporary characteristics in employee reporting relationships.

[0003] Therefore, in this context, for companies with thousands or even tens of thousands of employees and complex organizational structures, manually configuring review relationships or simple rule scripts is not only inefficient but also prone to various errors, failing to meet the requirements of efficiency and accuracy. Moreover, these errors will lead to the interruption of the performance appraisal process, data rework, and extremely high repair costs. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method, apparatus, electronic device and storage medium for generating performance review relationships, so as to improve the efficiency and accuracy of review relationship configuration at the same time.

[0005] Firstly, a method for generating performance review relationships is provided, including: Based on preset assessment criteria, the personnel participating in the current performance assessment are analyzed and obtained from the pre-constructed enterprise organizational chart. For each participant, based on the set of review rules bound to the review process node of the current performance appraisal, target reviewers are selected from the enterprise organizational graph, and review relationships are established with the corresponding participants. All review relationships are aggregated into an initial set of review relationships. Perform multi-dimensional anomaly detection on the initial review relationship set to identify outliers; Generate corresponding target remediation plans based on the types of anomalies, and remediate the anomalies based on the target remediation plans to obtain the final set of review relationships.

[0006] Optionally, based on the set of review rules bound to the review process nodes of the current performance appraisal, target reviewers are selected from the enterprise organizational graph, and review relationships are established with the corresponding participants. All review relationships are aggregated into an initial set of review relationships, including: For each review process node, iterate through the review rule set and find the multiple review rules that are bound to that review process node; For each review rule, an initial set of review candidates is obtained from the enterprise organizational graph according to the predefined query logic of that review rule; The initial set of review candidates is filtered based on the constraints configured in this review rule to obtain a subset of valid review candidates. According to the scoring strategy preset in the review rules, the effective review candidate subset is scored and ranked, and a preset number of target reviewers are selected based on the limit on the number of reviewers preset in the review rules. Generate a structured review relationship record for each target reviewer. The review relationship record shall include at least the reviewee identifier, reviewer identifier, relationship type, process node identifier, hit rule identifier, and generation reason. Aggregate the review relationship records generated by all participants at all review process nodes to form an initial review relationship set.

[0007] Optionally, before aggregating the review relationship records generated by all participants at all review process nodes to form the initial review relationship set, the following may also be included: Conflict resolution and deduplication are performed on all review relationship records corresponding to the same evaluated person identifier and the same process node identifier.

[0008] Optionally, multi-dimensional anomaly detection can be performed on the initial review relationship set to identify outliers, including: Iterate through all assessment participants and check whether there are corresponding review relationship records under each review process node to identify completeness and accessibility anomalies. Completeness and accessibility anomalies include an employee having no reviewers under a specific review process node, or an employee lacking a direct supervisor in the organizational chart. The system detects the status field of the reviewer in the enterprise organization diagram and verifies whether the reviewer and the person being reviewed meet the preset constraints of the same business unit, same department or same position sequence in order to identify validity anomalies. Validity anomalies include the reviewer being in a state of resignation, frozen or no permission, or the reviewer being outside the business scope allowed by the review process node. The initial set of review relationships is constructed into a directed graph, and directed cycle detection is performed to identify structural anomalies, including review loops, duplicate review relationships, or review structure anomalies caused by multi-supervisor conflicts. The number of tasks assigned to each reviewer in the initial review relationship set is counted to identify load anomalies, including when a reviewer's task count exceeds a threshold. For each identified anomaly, an anomaly set is constructed. The anomaly set includes at least the anomaly type, severity level, associated person being evaluated, corresponding process node identifier, and anomaly details.

[0009] Optionally, generating corresponding target remediation solutions based on the type of anomaly includes: For anomalies in completeness and accessibility, search upwards along the organizational reporting chain for the nearest valid supervisor as a candidate for alternative reviewer, or map to a pre-defined acting supervisor or HRBP, and generate remedial candidate solutions for replacement. For validity anomalies, HRBPs, department heads, or acting managers are sequentially assigned as supplementary reviewer candidates according to a tiered rollback strategy, generating remediation candidate solutions for supplementation. For structural anomalies, based on the priority strategy preset by the review relationship type, the edges to be deleted are determined in the relationships that form a loop, and when the priorities are the same, they are sorted according to the review relationship strength score or the scope of influence to generate repair candidate solutions for deletion or replacement. For load and fairness anomalies, under the constraints that each reviewer can undertake a maximum of a preset number of tasks and each evaluated person retains a fixed number of reviewers, at least one set of repair candidate solutions containing addition, deletion or replacement relationship operations is generated through a redistribution algorithm. The detected anomalies and corresponding repair candidate solutions are output to the interactive interface so that the target repair solution can be determined manually.

[0010] Optionally, the severity levels of anomalies include blocking and warning levels. Blocking anomalies are used to prevent the performance process from being initiated, while warning anomalies allow the process to continue but indicate that attention is needed. Based on the target remediation plan, the anomalies are remediated, resulting in the final set of review relationships, including: In response to a user's one-click batch repair operation for the target repair solution corresponding to a warning-level anomaly, or in response to a user's manual confirmation operation for the target repair solution corresponding to a blocking-level anomaly, the target repair solution corresponding to each anomaly is converted into a patch instruction. The patch instruction includes the repair operation type, target review relationship, and new and old reviewer identifiers. The initial review relationship set is updated based on the patch instructions to obtain the updated review relationship set.

[0011] Based on the updated set of review relationships, multi-dimensional anomaly detection is re-executed until no blocking-level anomalies are found, resulting in the final set of review relationships.

[0012] Optionally, after pre-building the enterprise organizational chart, an organizational chart snapshot is generated, and a reverse index of the enterprise organizational chart is constructed. The reverse index includes the mapping relationship from reviewers to those being reviewed. The method also includes: In response to organizational change events during the current performance appraisal period, calculate the set of differences between the current organizational state and the current organizational graph snapshot; the set of differences includes changes in node attributes and / or organizational relationship edges in the organizational graph. Based on the difference set and the pre-built reverse index, determine the affected review relationship influence domain; For participants within the influence domain of the review relationship, incremental review regeneration, anomaly detection, and repair operations are performed.

[0013] Secondly, a device for generating performance review relationships is provided, comprising: The acquisition unit is used to parse and acquire the participants in the current performance appraisal from a pre-constructed enterprise organizational graph based on preset assessment conditions. The building unit is used to select target reviewers from the enterprise organizational graph for each participant based on the review rule set bound to the review process node of the current performance appraisal, and to build review relationships with the corresponding participants. All review relationships are aggregated into an initial review relationship set. The detection unit is used to perform multi-dimensional anomaly detection on the initial review relationship set to identify outliers; The repair unit is used to generate a corresponding target repair plan based on the type of the anomaly, and repair the anomaly based on the target repair plan to obtain the final review relationship set.

[0014] Thirdly, an electronic device is provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements any of the methods of the first aspect.

[0015] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when the computer program is executed by a processor, it implements any of the methods of the first aspect.

[0016] This invention provides a method, apparatus, electronic device, and storage medium for generating performance review relationships. By constructing an enterprise organizational chart, it can clearly display the hierarchy and reporting relationships between various departments and employees. By traversing the chart, it can automatically identify potential review relationships, achieving automated generation of review relationships on a scale of tens of thousands of people, thus improving work efficiency. In addition, through pre-verification, it reduces the error rate and can detect and repair typical errors such as loops, duplications, and overload in advance, ensuring the integrity and reachability of the review chain, significantly improving the efficiency and accuracy of review relationship configuration. At the same time, through automatic repair, it reduces the high cost of manual maintenance.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart of a method for generating performance review relationships provided by an embodiment of the present invention is shown; Figure 2 A flowchart is shown for another method for generating performance review relationships provided by an embodiment of the present invention; Figure 3 A schematic diagram of the structure of a performance review relationship generation device provided in an embodiment of the present invention is shown; Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present invention is shown. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0021] This invention provides a method for generating performance review relationships, such as... Figure 1 As shown, it includes the following steps: Step S101: Based on the preset assessment conditions, analyze and obtain the participants in the current performance assessment from the pre-constructed enterprise organizational chart.

[0022] In this embodiment of the invention, the enterprise organizational graph (OrgGraph) is uniformly constructed and solidified before the performance process is initiated. This enterprise organizational graph is a directed heterogeneous graph that integrates multi-dimensional data, including at least basic personnel information (such as employee unique identifier user_id, employment status status, job grade, department org_unit), the main reporting line reports_to, and optionally, dotted reporting lines dotted_reports_to, belonging departments belong_to, roles held role_of, committee member relationships, etc.

[0023] These multi-source data come from HR systems, IM collaboration platforms, project management systems, etc. To ensure data consistency, the heterogeneous data is first aligned with entities, and a stable, unique employee identifier, user_id, is used as the primary key. A coding mapping table is then established for departments, positions, and roles to complete field standardization and value range stabilization.

[0024] Specifically, the organizational chart is modeled in the form of nodes and edges: nodes include employees, departments, positions and roles (such as HRBP, acting supervisor); edges include main report, dotted line report, affiliated department, role held, and optional collaborative relationships.

[0025] This organizational chart can be stored in a graph database or a relational database. It also provides a graph query interface for this organizational chart (e.g., to retrieve employee information). Direct supervisor interface Acquiring employees Inter-level supervisor interface Acquiring employees Peer interface Acquiring employees Human Resources Business Manager Interface Acquiring employees direct subordinate interface Interface for retrieving corresponding calibration / final review committee members based on grouping keys Acquiring employees Collaboration interface within a specified time window ).in, To efficiently support standardized graph query interfaces, the following index structure can be constructed, for example: Construct ancestor skip lists or doubling lists for the primary reporting relationship to accelerate queries by supervisors at every level. or ); Create a peer index based on department and direct supervisor clusters to improve... performance; Collaboration weights are aggregated based on time windows and recent interaction summaries are retained to form a collaboration adjacency index. In addition, an inverse index can be constructed, including a mapping from reviewers to those being reviewed and a mapping from candidate reviewers to potential evaluation tasks. This is used to quickly locate the affected relationship range when organizational change events occur, supporting incremental recalculation and self-healing during the execution period.

[0026] Furthermore, to support process reproducibility and audit traceability, key fields in the graph that affect review generation (such as reporting relationships, status, and departments) are normalized and sorted at the field level to generate a uniquely determined normalized sequence. And calculate the tissue map snapshot version number using a hash function:

[0027] in, This is a normalized sequence formed by normalizing and stabilizing the key fields that affect the generation of review relationships. The key fields include at least the user_id, status, org_unit, and grade fields of employee nodes, as well as the edge set summary of organizational relationship edges (reports_to, dotted_reports_to, belongs_to, role_of).

[0028] H(·) is a stable hash function used to generate a unique version number within the same organizational context, supporting playback reproduction and audit traceability.

[0029] At the same time, the set of nodes and edges is persistently stored in an immutable manner, forming a frozen snapshot.

[0030] Based on this graph, by graph traversal or attribute filtering, the set of employees who meet the assessment criteria are selected from the organizational graph and used as participants in this performance process.

[0031] In one feasible implementation, the preset assessment criteria are provided in the form of a structured scope configuration (scope_config), which can be manually entered by HR or automatically extracted from performance notification documents by AI, and includes the following elements: The include condition is a positive rule used to specify the scope of inclusion, supporting multiple dimensions such as department, job sequence, job level range, and static list; The exclude condition is used to exclude specific groups of people, such as interns, outsourced employees, probationary employees, or designated departments. named_list_refs (optional list of references to attachments): used to supplement or override structured conditions.

[0032] First, the structured scope configuration `scope_config` is standardized and compiled into a Boolean predicate abstract syntax tree (PredicateAST), where each leaf node corresponds to an atomic condition (e.g., `grade>="P6"` or `deptin["AI_Lab","Cloud"]`). Then, all employee nodes in the enterprise organizational graph are traversed, predicate evaluation is performed for each employee, and the hit path is recorded. If an employee satisfies both the `include` and `exclude` conditions, the final attribution is determined according to a preset priority strategy (exclude takes precedence by default), and a conflict reason flag is generated.

[0033] Based on this, the output includes: a set of participants[]: a list of all employee identifiers ultimately included; and a set of groups[]: groups selected according to a configurable grouping strategy (e.g., ...). ,in, The business department to which the employee belongs; For employee job level segments; This is the cluster identifier for the employee's direct supervisor. (Generated by clustering participants[], used for subsequent calibration of meeting groups, progress statistics, or committee binding;) After selecting participants from the enterprise organizational chart based on preset assessment criteria, the system not only outputs the final list but also generates a scope resolution and verification result, which includes three key components: coverage_rate (coverage_rate = actual number of participants / organization's expected number of participants), unmapped_entities (such as invalid department codes or non-existent user_ids), and the reason for each participant's inclusion (included_by), allowing the front-end to preview "why they were included".

[0034] For example, when scope_config is defined as "include: grade ≥ P6 and department = 'AILab'; exclude: status = 'intern'", it will return all qualified full-time employees and mark E1001 with included_by = ["grade ≥ P6", "dept = AI_Lab"], while outputting an unmapped_entities warning for an ID that has left the company but is still on the list.

[0035] This mechanism ensures that the scope of participation is not only accurate, but also interpretable, auditable, and verifiable throughout the process, providing a high-quality input foundation for the generation of subsequent review relationships.

[0036] Step S102: For each participant, based on the set of review rules bound to the review process node of the current performance appraisal, select target reviewers from the enterprise organizational graph, and build review relationships with the corresponding participants. Aggregate all review relationships into an initial set of review relationships.

[0037] In this embodiment of the invention, a performance process node is, for example, a self-assessment node. Direct supervisor review node Inter-level supervisor review nodes peer review nodes Collaborative review nodes Subordinate review nodes and calibration / final audit nodes .

[0038] Each review rule is explicitly bound to a specific performance process node, used to dynamically generate a structured dependency chain of "who reviews whom". Review rules include, but are not limited to: Self-assessment rules: The person being assessed is also the assessor, and is linked to... node; Direct supervisor evaluation rules: Query the main reporting line through the get_manager(u) interface to generate... → Relationship, bound to node; Interval evaluation rules (optional): Through The interface retrieves the supervisor at each level and generates... → Relationship, bound to node; 360 Peer Evaluation Rules: Based on the peer index (clustered by department + supervisor) or collaborative adjacency index, select TopKpeers and generate... → Relationship, bound to node; 360 Collaboration Evaluation Rules: Based on the set of collaborators in the enterprise organizational chart Filter highly interactive colleagues and generate → Relationship, bound to node; Subordinate evaluation rules (optional): Pass Get direct reports, suitable for managers' 360 feedback, and bind to... node; Calibration / Final Review Rules: Mapped to committee members according to preset groups (e.g., groups[]). ),generate → The review relationship of members is bound to... or node.

[0039] For each participant, all process nodes activated by their respective performance process are traversed, and multiple rules are loaded and bound to each node. Each rule is queried independently, ultimately generating one or more structured review relationship records. All records are aggregated into a global initial review relationship set `relations[]`, which fully expresses the review dependency network of this assessment in the form of a flat-side list, providing an input basis for subsequent anomaly detection and repair.

[0040] Step S103: Perform multi-dimensional anomaly detection on the initial review relationship set to identify anomalies.

[0041] This step enables proactive error detection, preventing interruptions caused by broken links or loops after the process has been initiated.

[0042] Step S104: Generate the corresponding target repair plan according to the type of the anomaly, and repair the anomaly based on the target repair plan to obtain the final review relationship set.

[0043] In this embodiment of the invention, corresponding repair strategies are invoked for different exception types, and all repair operations are output in the form of patches (such as {op:replace,target:(E1,manager),old:M1,new:M2}), and multiple solution candidates are supported.

[0044] This invention constructs an organizational chart of an enterprise, which clearly displays the hierarchy and reporting relationships between various departments and employees. By traversing this chart, it automatically identifies potential review relationships, achieving automated generation of review relationships on a scale of tens of thousands of people, thus improving work efficiency. In addition, through pre-verification, it reduces the error rate and can detect and repair typical errors such as loops, duplications, and overload in advance, ensuring the integrity and reachability of the review chain, significantly improving the efficiency and accuracy of review relationship configuration. At the same time, through automatic repair, it reduces the high cost of manual maintenance.

[0045] Based on the above embodiments, and using the set of review rules bound to the review process nodes of the current performance appraisal, target reviewers are selected from the enterprise organizational chart, and review relationships are established with the corresponding participants. All review relationships are aggregated into an initial set of review relationships, including: Step S102A: For each review process node, iterate through the multiple review rules bound to that review process node in the review rule set.

[0046] In this embodiment of the invention, each review rule is represented in the form of a structured object, including rule identifier rule_id, process node identifier step_key, predicate (applicable conditions), candidate_query (candidate query expression), constraints (filtering constraints), priority (conflict priority), enabled (enabled status), and version (review version number).

[0047] This embodiment only expands rules where enabled=true and the predicate evaluates to true for the current participant. For example, A node can be bound to multiple rules, which are applicable to technical and product roles respectively, and are activated by predicate.

[0048] Step S102B: For each review rule, obtain the initial set of review candidates from the enterprise organizational graph according to the predefined query logic of the review rule.

[0049] Continuing from the previous example, the corresponding graph interface can be called to execute a query and return an initial candidate list. For example, the 360 ​​collaboration evaluation rule might define candidate_query=topK(collaborators(u),5,strength), where strength is calculated by weighting collaboration frequency, timeliness, and project overlap.

[0050] Step S102C: Filter the initial set of review candidates based on the constraints configured in the review rule to obtain a subset of valid review candidates.

[0051] It should be noted that constraints may include Boolean conditions such as "rater.status=active", "rater.dept=employee.dept", "rater≠employee", and "rater cannot be a subordinate".

[0052] Constraints are applied one by one, those that are not satisfied are eliminated, and the number of filters for each constraint is recorded (e.g., filtered_reason_counts={“inactive”:2,“cross_bu”:1}) for subsequent rule execution traces.

[0053] Step S102D: According to the scoring strategy preset in the review rules, the effective review candidate subset is scored and ranked, and a preset number of target reviewers are selected based on the limit on the number of reviewers preset in the review rules.

[0054] In this embodiment of the invention, the scoring strategy is based on relationship strength scoring. To achieve this, the calculation formula is as follows:

[0055] in, This indicates employees within a preset time window. and Statistics on the frequency of collaborative interactions; Indicates employees and The time interval between the most recent collaborative interaction; express and Overlap of common projects / tasks within a preset time window (can be counted or normalized values); , , , Configure parameters for rules and incorporate them into rule version management.

[0056] System by Sort in descending order and select the Top K as target reviewers according to the number of reviewers configured according to the rules (e.g., K=5).

[0057] Step S102E: Generate a structured review relationship record for each target reviewer.

[0058] The review relationship record should include at least the reviewee identifier, reviewer identifier, relationship type, process node identifier, hit rule identifier, and generation reason.

[0059] The relationship type is used to distinguish different relationship sources under the same process node (such as master / dashed master), and supports subsequent structural validation.

[0060] Step S102F: Aggregate the review relationship records generated by all participants under all review process nodes to form an initial review relationship set.

[0061] All relationships are managed uniformly using a flat list `relations[]`, with each record carrying complete metadata, providing high-fidelity input for anomaly detection, repair, and auditing.

[0062] This embodiment achieves accurate and interpretable mapping from multi-source rules to structured review relationships through a rule-driven review relationship generation mechanism. It supports complex scenarios such as master / dashed line reporting, peer collaboration, and committee calibration, ensuring the automated and high-quality generation of "who reviews whom" relationships at a scale of tens of thousands of people. Furthermore, it ensures the representativeness of reviewers and business adaptability through relationship strength scoring and TopK truncation.

[0063] Based on the above embodiments, before aggregating the review relationship records generated by all participants at all review process nodes to form the initial review relationship set, the following is also included: Step S102G: Perform conflict resolution and deduplication on all review relationship records corresponding to the same reviewee identifier and the same process node identifier.

[0064] Continuing from the previous example, if multiple review rules are in If a node also matches rater_id=R1001, the relationship with higher priority or higher relationship strength score is retained, and the generated reason is merged (e.g., "[Top 3 technical collaborations, Top 5 project co-occurrences]"). If the same process node hits both the main and dotted line supervisors, the main line supervisor will be retained according to the policy. If a node requires a unique reviewer but generates multiple reviewers, the decision is made based on priority; if multiple reviewers are allowed, the number of reviewers is truncated according to the preset number.

[0065] All decision-making processes are written into the rule execution trajectory field, including deduplication decisions, priority decisions, etc., to ensure that the process is interpretable and the result is deterministic.

[0066] In addition, to ensure traceability, the entire rule set is normalized (field standardization, stable sorting of the set, and sorting by step_key + rule_id) before generating relations[], and the review rule version number is calculated using a hash algorithm. :

[0067] in, This is the rule sequence formed after standardizing the current set of review rules. Standardization includes at least the following: standardizing the field names and parameters of predicate, constraints, and candidate_query; performing stable sorting of the set fields; and performing stable sorting of the rule set by (step_key, rule_id).

[0068] H(·) is a stable hash function used to generate a unique version number under the same rule configuration.

[0069] When generating each review relationship record, Identifier for each review relationship It is written together and eventually solidified into the relationship chain snapshot along with the enterprise organization snapshot version to support rule source tracing and snapshot playback.

[0070] This embodiment effectively solves problems such as duplicate reviews and conflicts in the main dotted line caused by multiple rules or multiple reports from organizations. While retaining high-priority or high-intensity relationships, it merges the generated causes, which not only avoids redundant relationships interfering with the process, but also provides clean input for subsequent anomaly detection, improving the consistency and determinism of the relationship set.

[0071] Based on the above embodiments, multi-dimensional anomaly detection is performed on the initial review relationship set to identify anomalies, including: Step S103A: Traverse all assessment participants and check whether there are corresponding review relationship records for each review process node to identify completeness and accessibility anomalies.

[0072] Completeness and accessibility anomalies include an employee not having any reviewers at a specific review process node, or an employee lacking a direct supervisor in the organizational chart.

[0073] Such exceptions are typically set as blocking exceptions to prevent the process from starting.

[0074] Specifically, for each participant in the assessment, it is checked whether at least one review relationship has been generated at each review process node they should participate in (such as "direct supervisor review" or "peer review"). If an employee has no reviewers at a specific review process node, it is considered an anomaly in completeness and accessibility. Furthermore, if the employee lacks a direct supervisor in the company's organizational chart (i.e., it is impossible to locate an effective manager through the reporting chain), it is also classified as such an anomaly. The anomaly details record the employee's organizational path (e.g., AI Lab / Algorithm Group), the matched rule identifier, and the reason why the candidate reviewer pool is empty, to locate data missing or configuration omissions.

[0075] Step S103B: Detect the status field of the reviewer in the enterprise organization diagram, and verify whether the reviewer and the person being reviewed meet the preset constraints of the same business unit, same department or same job sequence, in order to identify validity anomalies.

[0076] Validity anomalies include reviewers who are out of the company, have frozen access, or lack authorization, or reviewers who are not within the permitted business scope of the review process node.

[0077] In this embodiment of the invention, the current status of the reviewer in each review relationship is checked one by one to see if it is "on duty and valid". If the reviewer is in a state of resignation, frozen or invalidation of permissions, it is determined to be an invalidity. Simultaneously, it verifies whether the reviewer and the reviewed party meet the scope constraints required for that review node (e.g., they must belong to the same business unit, and cross-department reviews are not allowed). If they do not meet these requirements, they are also marked as validity anomalies. Anomaly details include the specific constraints violated (e.g., the reviewer and the reviewed party do not belong to the same department), rule identifier, candidate pool size, and the actual number of selected candidates, facilitating the tracing of rule execution deviations.

[0078] Step S103C: Construct the initial review relation set into a directed graph and perform directed cycle detection to identify structural anomalies.

[0079] Structural anomalies include review loops, duplicate review relationships, or review structure anomalies caused by conflicts among multiple supervisors.

[0080] Continuing from the previous example, all review relationships are constructed into a graph structure with employees as nodes and review directions as directed edges, and a loop detection algorithm is run. If a review loop with a length greater than 1 is found (e.g., A reviews B, B reviews C, and C reviews A again), it is judged as a structural anomaly; In addition, the system checks whether the same person being evaluated is repeatedly assigned the same reviewer at the same review process node, or whether a management relationship conflict arises because the main reporting supervisor and the dotted reporting supervisor are included simultaneously. These situations are considered structural anomalies. The anomaly details include the complete loop path (e.g., A→B→C→A), the relevant rule identifiers, and the source of the conflict, ensuring that the problem is locatable and explainable.

[0081] Step S103D: Count the number of tasks assigned to each reviewer in the initial review relationship set to identify load anomalies.

[0082] Load anomalies include a reviewer's task count exceeding a threshold.

[0083] It should be noted that the total number of review tasks assigned to each reviewer is aggregated and compared with the load threshold set by their job level or department. If a reviewer is assigned more tasks than the allowed limit, it is considered an abnormality in terms of load and fairness. Furthermore, the uneven distribution of overall review task allocation (such as the Gini coefficient or standard deviation) is calculated. If it exceeds the preset tolerance range, it is also considered an anomaly. Anomaly details include actual task volume, thresholds, changes in fairness indicators, etc., providing a quantitative basis for subsequent optimization and reallocation.

[0084] Step S103E: Construct an anomaly set based on each identified anomaly.

[0085] An anomaly set should include at least the anomaly type, severity level, associated person being assessed, corresponding process node identifier, and anomaly details.

[0086] The anomaly details include at least one of the following: the rule identifier of the hit, the organizational path, the constraints of the failure, the size of the candidate pool, the actual number of selected candidates, and the loop path.

[0087] The severity levels are divided into blocking and warning levels: blocking level anomalies (such as missing integrity or review loops) will prevent the performance process from being initiated; warning level anomalies (such as slightly overloaded or slightly skewed distribution) allow the process to continue, but a notification should be displayed on the interactive interface.

[0088] Based on the above embodiments, generating corresponding target remediation solutions according to the type of anomalies includes: Step S104A: For integrity and accessibility anomalies, search upwards along the organizational reporting chain for the nearest valid supervisor as a candidate for alternative reviewer, or map to a preset acting supervisor or HRBP, and generate a remediation candidate solution for replacement.

[0089] In this embodiment of the invention, when it is detected that an employee is missing a direct supervisor or there is no reviewer at a specific review node, the "most recent effective supervisor search" operator is first invoked. Starting from the employee's direct supervisor, the process is traced up the main reporting chain level by level, skipping nodes that are inactive, until the first active and effective manager is found as a substitute reviewer. If the reporting chain has reached the organizational root and the report is still not found, then the HRBP, the head of the relevant department, or the designated acting manager will be tried in sequence according to the preset hierarchical strategy.

[0090] Each candidate solution comes with a reason for its creation (such as a supervisor filling in a gap or an HRBP providing a safety net) and an impact assessment, including the number of employees affected, the degree of deviation from the rules, and whether new anomalies are introduced, for reference in subsequent decision-making.

[0091] Step S104B: For validity anomalies, try assigning HRBPs, department heads, or acting managers as supplementary reviewer candidates in sequence according to the hierarchical rollback strategy, and generate remediation candidate solutions for supplementation.

[0092] Specifically, when a reviewer is unable to perform their duties due to resignation, account suspension, or invalidation of permissions, the original relationship is no longer maintained. Instead, a tiered rollback mechanism is activated: first, the HRBP corresponding to the employee is attempted to take over the review responsibilities; if the HRBP is unavailable, the rollback proceeds to the head of the employee's department; if this is still not feasible, a pre-configured pool of proxy managers can be used. All candidate solutions are output as supplementary reviewers, with the rollback path (e.g., HRBP → Department Head), candidate pool size, and matching score clearly indicated to ensure the remediation process is both compliant and explainable.

[0093] Step S104C: For structural anomalies, based on the priority strategy preset by the review relationship type, determine the edges to be deleted in the relationships that form a loop, and when the priorities are the same, sort them according to the review relationship strength score or the scope of influence to generate repair candidate solutions for deletion or replacement.

[0094] Following the previous example, for the detected review loops (such as A evaluating B, B evaluating C, and C evaluating A), the importance of each side is first determined according to the preset relationship type priority (e.g., direct supervisor evaluation > inter-level evaluation > peer evaluation > collaborative evaluation); high-priority sides are retained first, and low-priority sides are deleted.

[0095] If multiple edges have the same priority, their relationship strength score (calculated based on factors such as collaboration frequency, timeliness, and project overlap) or impact range (number of employees involved) is further compared, and the edge with the lowest strength or least impact is selected for deletion. The remediation plan explicitly references the original loop path and explains the basis for edge breaking, ensuring transparency and reproducibility of the operation.

[0096] Step S104D: For load and fairness anomalies, under the constraints that each reviewer can undertake a maximum of a preset number of tasks and each evaluated person retains a fixed number of reviewers, at least one set of repair candidate solutions containing add, delete or replace relationship operations is generated through a redistribution algorithm.

[0097] In this embodiment of the invention, to address the problem of task allocation overload or severe overall distribution imbalance, a global optimization and reallocation is performed on the pool of selectable reviewers (same level, collaborators).

[0098] The constraints include: the number of tasks undertaken by each reviewer shall not exceed the upper limit corresponding to their job level, and each person being reviewed shall retain a fixed number of reviewers.

[0099] The optimization objective can be configured as: maximizing the overall relationship strength, minimizing the load imbalance, or a weighted balance of the two.

[0100] The algorithm employs minimum cost flow, linear programming relaxation, or heuristic greedy algorithms to solve the problem and outputs multiple allocation schemes. Each scheme includes a set of operations, a comparison of the workload of key reviewers before and after the allocation, fairness improvement indicators, and relationship strength retention rate, allowing HR to select according to business preferences.

[0101] Step S104E: Output the detected anomalies and corresponding candidate repair solutions to the interactive interface so that the target repair solution can be determined manually.

[0102] The interface displays anomaly details, reasons for repair, impact assessment (e.g., affecting 3 employees, fairness improved by 15%), and comparison of multiple solutions (e.g., strength priority, fairness priority). It supports users to batch adopt warning-level anomalies with one click and approve and confirm blocking-level anomalies one by one, realizing efficient decision-making through human-machine collaboration.

[0103] This embodiment constructs an adaptive repair candidate generation mechanism for different anomaly types. Combining strategies such as hierarchical rollback, priority edge breaking, and load optimization redistribution, it outputs interpretable, evaluable, and multi-scheme-selectable repair suggestions, and supports human-machine collaborative decision-making, ensuring compliance while taking into account efficiency and flexibility.

[0104] Based on the above embodiments, the final set of review relationships is obtained by repairing the anomalies according to the target repair scheme, including: Step S104F: In response to the user's one-click batch repair operation for the target repair solution corresponding to the warning-level anomaly, or in response to the user's manual confirmation operation for the target repair solution corresponding to the blocking-level anomaly, convert the target repair solution corresponding to each anomaly into a patch instruction.

[0105] Patch instructions all adopt a standardized format, including the type of repair operation, the target review relationship (uniquely determined by the reviewed person's identifier and the process node identifier), and the old and new reviewer identifiers. For high-risk operations (such as deleting management review relationships), the system forcibly records the operation reason, the operator, and the timestamp, generating an audit log to ensure that the entire process is traceable and rollbackable.

[0106] Step S104G: Update the initial review relationship set based on the patch instructions to obtain the updated review relationship set.

[0107] Step S104H: Re-execute multi-dimensional anomaly detection based on the updated review relationship set until there are no blocking-level anomalies, and obtain the final review relationship set.

[0108] For example, after each patch is applied, the system automatically triggers a new round of integrity, validity, structure, and load fairness checks. If the new fix introduces secondary anomalies (such as creating a new loop after replacement or causing another reviewer to be overloaded), then fix candidates are generated and the process proceeds to the next iteration.

[0109] This closed-loop mechanism ensures that the final output set of review relationships fully meets business rules and organizational constraints, and can only enter the process initiation stage with zero blocking errors, significantly improving the reliability and automation level of large-scale performance configuration.

[0110] In another feasible implementation, after there are no blocking-level anomalies and the final review relationship set is obtained, the participants in this session will be grouped together. Final review relationship set and group information The data is saved as a snapshot of the relationship chain, and a snapshot consistency check value is calculated. :

[0111] in, For the participants to gather; A snapshot of the review relationship; For grouped sets; For the enterprise organization map snapshot version number; This refers to the version number of the review rules. This serves as the marker for the repair trajectory in this closed-loop repair process.

[0112] Execution period can be based on Perform snapshot consistency checks to detect whether the relationship chain snapshot has drifted unexpectedly or been tampered with.

[0113] Based on the above embodiments, such as Figure 2 As shown, the method also includes: Step S105: In response to an organizational change event during the current performance appraisal period, calculate the set of differences between the current organizational state and the current organizational map snapshot.

[0114] In this embodiment of the invention, once the performance process is initiated, the enterprise organizational chart version and review rule version at the time of initiation are frozen as a benchmark.

[0115] When an organizational change event (such as employee departure, job transfer, supervisor adjustment, or update of collaboration relationships) is received during the execution period, the latest organizational status is first obtained and compared in a structured manner with the frozen snapshot to generate a set of differences. .

[0116] This set of differences includes at least two types of changes: first, changes in node attributes, such as an employee status changing from "employed" to "departed", or updates to the department or job level field; second, changes in graph edge relationships, such as the addition, deletion, or modification of master reporting relationships (reports_to), dotted reporting relationships (dotted_reports_to), belonging relationships (belongs_to), role bindings (role_of), or collaboration relationships (collaborates_with).

[0117] For example, when the "M1001 terminated" event is received, Includes changes to the status attributes of node M1001; when the "E1001reports_to=M2001" event is received, This includes an update to one `reports_to` edge. This fine-grained difference comparison avoids recalculating all review relationships, significantly improving the efficiency of change handling during execution.

[0118] Step S106: Determine the affected review relationship influence domain based on the difference set and the pre-built reverse index.

[0119] Specifically, pre-built inverted indexes are used to quickly locate the affected party. The scope of the review relationships affected. The inverted index includes: a mapping from a reviewer to all the reviewers they have reviewed (for quickly finding "who someone reviewed"), and a mapping from a candidate reviewer to the review tasks they may have participated in (for identifying "which tasks someone has been a candidate for").

[0120] If a resignation occurs, use the reverse index to find all relationships with that person as a reviewer, as well as their potential tasks as a candidate, and add the relevant reviewers to the influence domain. If a supervisor change occurs, the affected person, their subordinates (configurable depth), and peers will be included in the affected domain and tagged accordingly. Nodes and The node needs to be recalculated; If a change in collaboration occurs, only the employee and their highly interactive collaborators will be included in the affected domain, and a targeted recalculation will be performed. or node.

[0121] For example, when M1001 leaves the company, querying the reverse index `reverse_rater_index[M1001]` yields `[E1001,E1002]`, indicating that M1001 was a reviewer for these two employees. Therefore, the affected domain A = {E1001,E1002}. This mechanism achieves precise location, minimizes the scope of repair, and avoids disturbing the review chain of unrelated employees.

[0122] Step S107: For participants within the review relationship's influence domain, perform incremental review regeneration, anomaly detection, and repair operations.

[0123] Following the previous example, the complete local processing flow is re-executed only for employees in affected domain A: First, based on the frozen rule version, the rules for the applicable review process nodes are expanded to generate new candidate review relationships; then, multi-dimensional anomaly detection is performed to identify new anomalies caused by organizational changes (such as invalid new supervisors, empty collaboration pools, sudden load increases, etc.); then, the adaptive repair mechanism is invoked to generate repair candidates, and patches are applied according to configuration policies or manual confirmation.

[0124] After the fix is ​​complete, the updated subset of review relationships is revalidated until there are no blocking-level anomalies, and a new fix version identifier is generated. Review relationships not within affected domain A remain unchanged.

[0125] In another implementation, complete repair trajectory identifiers and audit logs are recorded to ensure that each incremental change is traceable, replayable, and verifiable.

[0126] This embodiment implements an incremental recalculation mechanism during execution, accurately locating the affected domain based on organizational change events. Only the affected employees partially rerun the generation-detection-repair process, while keeping other relationships frozen. This ensures the stability of large-scale performance processes in dynamic organizational environments and achieves agile response capabilities with "flexibility within freeze," significantly improving the system's availability and robustness in real business scenarios.

[0127] Based on the same inventive concept, embodiments of the present invention provide a device for generating performance review relationships, such as... Figure 3 As shown, it includes: The acquisition unit 301 is used to parse and obtain the participants in the current performance appraisal from the pre-constructed enterprise organizational chart based on preset assessment conditions. Building unit 302 is used to select target reviewers from the enterprise organizational graph for each participant based on the review rule set bound to the review process node of the current performance appraisal, and to build review relationships with the corresponding participants. All review relationships are aggregated into an initial review relationship set. The detection unit 303 is used to perform multi-dimensional anomaly detection on the initial review relationship set to identify anomalies; Repair unit 304 is used to generate a corresponding target repair plan based on the type of the anomaly, and repair the anomaly based on the target repair plan to obtain the final review relationship set.

[0128] Based on the same technical concept, embodiments of the present invention also provide an electronic device, such as... Figure 4 As shown, it includes a processor 401, a communication interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404.

[0129] Memory 403 is used to store computer programs; The processor 401 is used to execute the program stored in the memory 403 to implement the steps of the method for generating performance review relationships.

[0130] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0131] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0132] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0133] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0134] In another embodiment of the present invention, a computer-readable storage medium is also provided, which stores a computer program that, when executed by a processor, implements the steps of the method for generating any of the above-described performance review relationships. Specific implementation details can be found in the method embodiments and will not be repeated here.

[0135] The apparatus for generating performance review relationships provided in this embodiment of the invention can be specific hardware on a device or software or firmware installed on the device. The implementation principle and technical effects of the apparatus provided in this embodiment of the invention are the same as those in the foregoing method embodiments. For the sake of brevity, any parts not mentioned in the apparatus embodiments can be referred to the corresponding content in the foregoing method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, apparatuses, and units described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0136] In the embodiments provided by this invention, it should be understood that the disclosed apparatus and method can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0137] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] In addition, the functional units in the embodiments provided by the present invention 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.

[0139] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0140] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0141] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. All should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for generating performance review relationships, characterized in that, include: Based on preset assessment criteria, the personnel participating in the current performance assessment are analyzed and obtained from the pre-constructed enterprise organizational chart. For each participant, based on the set of review rules bound to the review process node of the current performance appraisal, target reviewers are selected from the enterprise organizational graph, and review relationships are established with the corresponding participants. All review relationships are aggregated into an initial set of review relationships. Multi-dimensional anomaly detection is performed on the initial review relationship set to identify anomalies; Generate a corresponding target repair plan based on the type of the anomaly, and repair the anomaly based on the target repair plan to obtain the final review relationship set.

2. The method according to claim 1, characterized in that, The set of review rules, bound to the review process node of the current performance appraisal, selects target reviewers from the enterprise organizational graph and establishes review relationships with corresponding participants. An initial set of review relationships is aggregated based on all these relationships, including: For each review process node, iterate through the review rule set and find the multiple review rules that are bound to that review process node; For each of the aforementioned review rules, an initial set of review candidates is obtained from the enterprise organizational graph according to the predefined query logic of that review rule; The initial set of review candidates is filtered based on the constraints configured in the review rules to obtain a subset of valid review candidates. According to the scoring strategy preset in the review rules, the subset of valid review candidates is scored and ranked, and a preset number of target reviewers are selected based on the limit on the number of reviewers preset in the review rules. For each target reviewer, generate a structured review relationship record, which includes at least the reviewee identifier, reviewer identifier, relationship type, process node identifier, hit rule identifier, and generation reason; Aggregate the review relationship records generated by all participants at all review process nodes to form an initial review relationship set.

3. The method according to claim 2, characterized in that, Before aggregating the review relationship records generated by all participants at all review process nodes to form the initial review relationship set, the following is also included: Conflict resolution and deduplication are performed on all review relationship records corresponding to the same evaluated person identifier and the same process node identifier.

4. The method according to claim 1, characterized in that, The multi-dimensional anomaly detection of the initial review relationship set to identify anomalies includes: Iterate through all assessment participants and check whether there are corresponding review relationship records under each review process node to identify completeness and accessibility anomalies. The completeness and accessibility anomalies include an employee having no reviewers under a review process node, or having a missing direct supervisor in the organizational chart. The system detects the status field of the reviewer in the enterprise organization diagram and verifies whether the reviewer and the person being reviewed meet the preset constraints of the same business unit, same department or same position sequence in order to identify validity anomalies. The validity anomalies include the reviewer being in a state of resignation, frozen or no permission, or the reviewer being outside the business scope allowed by the review process node. The initial set of review relationships is constructed into a directed graph, and directed cycle detection is performed to identify structural anomalies, including review loops, duplicate review relationships, or review structure anomalies caused by multi-supervisor conflicts. The number of tasks assigned to each reviewer in the initial review relationship set is counted to identify load anomalies, including when a reviewer's task count exceeds a threshold. An anomaly set is constructed based on each identified anomaly. The anomaly set includes at least the anomaly type, severity level, associated person identifier, corresponding process node identifier, and anomaly details.

5. The method according to claim 4, characterized in that, The step of generating a corresponding target repair plan based on the type of the anomaly includes: For the integrity and accessibility anomalies, the nearest valid supervisor is searched up the organizational reporting chain as a candidate for alternative reviewer, or mapped to a preset acting supervisor or HRBP, to generate a remediation candidate solution for replacement. For the aforementioned validity anomalies, an HRBP, department head, or acting manager are sequentially assigned as supplementary reviewer candidates according to a tiered rollback strategy, generating a repair candidate plan for supplementation. For the aforementioned structural anomalies, based on the priority strategy preset by the review relationship type, the edges to be deleted are determined in the relationships that form a loop, and when the priorities are the same, they are sorted according to the review relationship strength score or the scope of influence to generate repair candidate solutions for deletion or replacement. For the aforementioned load and fairness anomalies, under the constraints that each reviewer can undertake a maximum of a preset number of tasks and each evaluated person retains a fixed number of reviewers, at least one set of repair candidate schemes containing addition, deletion or replacement relationship operations is generated through a redistribution algorithm. The detected anomalies and corresponding repair candidate solutions are output to the interactive interface so that the target repair solution can be determined manually.

6. The method according to claim 5, characterized in that, The severity levels of the anomalies include blocking and warning levels. Blocking anomalies are used to prevent the performance process from being initiated, while warning anomalies allow the process to continue but indicate that attention is needed. Based on the target repair scheme, the anomalies are repaired, resulting in the final set of review relationships, including: In response to a user's one-click batch repair operation for the target repair solution corresponding to a warning-level anomaly, or in response to a user's manual confirmation operation for the target repair solution corresponding to a blocking-level anomaly, the target repair solution corresponding to each anomaly is converted into a patch instruction, wherein the patch instruction includes the repair operation type, target review relationship, and new and old reviewer identifiers; The initial review relationship set is updated based on the patch instructions to obtain the updated review relationship set. Based on the updated review relationship set, multi-dimensional anomaly detection is re-executed until no blocking-level anomalies are found, resulting in the final review relationship set.

7. The method according to claim 1, characterized in that, After pre-building the enterprise organizational chart, an organizational chart snapshot is generated, and a reverse index of the enterprise organizational chart is constructed, which includes the mapping relationship from reviewers to those being reviewed; The method further includes: In response to an organizational change event during the current performance evaluation period, calculate the set of differences between the current organizational state and the current organizational graph snapshot; the set of differences includes changes in node attributes and / or organizational relationship edges in the organizational graph. Based on the set of differences and the pre-built reverse index, determine the affected review relationship influence domain; For participants within the influence domain of the aforementioned review relationship, incremental review regeneration, anomaly detection, and repair operations are performed.

8. A device for generating performance review relationships, characterized in that, include: The acquisition unit is used to parse and acquire the participants in the current performance appraisal from a pre-constructed enterprise organizational graph based on preset assessment conditions. The construction unit is used to select target reviewers from the enterprise organizational graph for each participant based on the review rule set bound to the review process node of the current performance appraisal, and to build review relationships with the corresponding participants, and to aggregate all review relationships into an initial review relationship set. The detection unit is used to perform multi-dimensional anomaly detection on the initial review relationship set to identify anomalies; The repair unit is used to generate a corresponding target repair plan according to the type of the anomaly, and repair the anomaly based on the target repair plan to obtain the final review relationship set.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images