Equity Penetration Analysis System Design for Person-Related Enterprises

    This article explains how to design an equity penetration analysis system for identifying direct and indirect shareholding, related enterprises, control paths, and cross-holding or circular ownership structures starting from a natural person. It covers problem definition, temporal property-graph data modeling, multi-hop path algorithms, ratio accumulation and cycle handling, offline graph construction with online query services, version release gates, evaluation metrics, and compliance and explainability requirements. Readers will learn how to separate factual relationship data from rule-based control judgments and how to make penetration results traceable, explainable, and operationally stable.

    Equity Penetration Analysis System Design for Person-Related Enterprises

    In scenarios such as risk control, compliance review, related-party transaction identification, and corporate due diligence, “person equity penetration” is a frequent and complex technical problem. The goal is usually not simply to query which enterprises a person directly holds, but to start from a target natural person and perform multi-hop analysis along equity relationships, position relationships, investment relationships, and control relationships. This helps identify directly or indirectly held enterprises, related enterprises, suspected control paths, and cross-holding or circular ownership structures.

    Such analysis may look like a graph query problem, but real-world implementation faces multiple challenges: multi-source heterogeneous data, temporal version changes, cross-holding, circular paths, nominee holdings and contractual control, data quality, and privacy compliance. This article focuses on the design and implementation of a “person-related enterprise equity penetration analysis system,” with emphasis on data modeling, penetration algorithms, engineering architecture, version iteration, and quality evaluation.

    1. Problem Definition: What Equity Penetration Actually Traces

    Person equity penetration generally includes three directions:

    1. Downward penetration: Starting from the target person, expand downward along “natural person—enterprise” shareholding relationships to identify directly held enterprises, indirectly held enterprises, and cumulative penetration ratios.
    2. Upward penetration: Starting from the target enterprise, reverse-search its shareholders, actual controllers, ultimate beneficial owners, and multi-level parent entities.
    3. Horizontal association: Through common shareholders, common positions, common investments, concerted-action relationships, and similar connections, identify the related network of the target person or target enterprise.

    From a technical implementation perspective, it is necessary to distinguish “shareholding ratio calculation” from “control relationship determination.” Shareholding ratios can be accumulated along a path, but control relationships cannot be judged mechanically based only on a ratio threshold. They also need to incorporate voting rights, board seats, contractual control, position relationships, and other business rules. The system should separate “factual data” from “rule-based judgment”: factual data describes relationships, while the rule engine outputs control conclusions and preserves explainable paths.

    2. Data Modeling: Entities, Relationships, Time, and Confidence

    The core data model for equity penetration is a property graph: nodes represent entities such as natural persons, enterprises, and institutions; edges represent relationships such as shareholding, position, investment, and control. Each edge should contain at least attributes such as ratio, amount, time interval, data source, confidence, and status.

    A simplified relationship schema can be abstracted as follows:

    1. -- Entity table: unified abstraction of natural persons and organizations
    2. entity(
    3. entity_id,
    4. entity_type, -- person / company / organization
    5. name,
    6. status,
    7. source_id,
    8. updated_at
    9. )
    10. -- Relationship table: equity, position, investment, control, etc.
    11. relation(
    12. relation_id,
    13. from_entity_id,
    14. to_entity_id,
    15. relation_type, -- equity / position / investment / control
    16. ratio, -- shareholding ratio, nullable
    17. amount, -- subscribed or paid-in amount, nullable
    18. valid_from,
    19. valid_to,
    20. source_id,
    21. confidence,
    22. status
    23. )

    Several key points matter during modeling:

    • Temporal versioning: Equity relationships change over time. Queries must support penetration “as of a given date” rather than always using the latest snapshot.
    • Source and confidence: Data quality varies by source. The system should record source and confidence and indicate the basis in results.
    • Status filtering: Statuses such as deregistered, revoked, or moved-out can affect association judgments and should not be expanded indiscriminately.
    • Relationship deduplication: The same pair of entities may have multiple relationships with different times and sources. These need to be merged according to business rules or retained as an evidence chain.

    3. Penetration Algorithms: Path Enumeration, Ratio Accumulation, and Cycle Handling

    A typical equity penetration algorithm is multi-hop path search. In engineering practice, breadth-first or depth-first strategies can be used, with constraints such as maximum depth, maximum number of paths, and minimum ratio threshold to avoid combinatorial explosion.

    An illustrative pseudocode is as follows:

    1. function penetrate(start_entity, max_depth, min_ratio):
    2. queue = [(start_entity, ratio=1.0, path=[])]
    3. results = []
    4. visited_on_path = set()
    5. while queue not empty:
    6. node, acc_ratio, path = queue.pop()
    7. if depth(path) >= max_depth:
    8. continue
    9. for edge in query_edges(node, relation_type="equity"):
    10. if edge.status != "valid":
    11. continue
    12. if edge.to_entity in path:
    13. record_cycle(path + [edge])
    14. continue
    15. next_ratio = acc_ratio * edge.ratio
    16. if next_ratio < min_ratio:
    17. continue
    18. new_path = path + [edge]
    19. results.append({
    20. entity: edge.to_entity,
    21. path: new_path,
    22. path_ratio: next_ratio,
    23. confidence: min_confidence(new_path)
    24. })
    25. queue.push((edge.to_entity, next_ratio, new_path))
    26. return aggregate_by_entity(results)

    Note that the penetration ratio on a single chain can be accumulated multiplicatively, but multi-path aggregation cannot simply add ratios together. Duplicate paths, dilution effects, and business definitions must be handled. Circular shareholding must be detected and recorded rather than expanded indefinitely. For control relationships, conclusions should be produced by a rule engine, for example “shareholding ratio reaches threshold,” “position relationships overlap,” or “contractual control evidence exists,” while preserving the triggered rules and evidence paths.

    4. Engineering Architecture: Combining Offline Graphs with Online Queries

    An equity penetration system typically adopts an architecture of “offline graph construction + online query analysis”:

    • Data ingestion layer: Collect multi-source data and perform cleaning, standardization, entity alignment, and relationship extraction.
    • Graph storage layer: Store entities, relationships, temporal versions, and evidence chains, supporting multi-hop queries and path retrieval.
    • Rule engine layer: Execute business rules such as shareholding ratio calculation, control relationship determination, and related-enterprise judgment.
    • Query service layer: Provide person penetration, enterprise reverse lookup, path explanation, result export, and related capabilities.
    • Permission and audit layer: Control data access by role and record query audits and result usage trails.

    At the infrastructure level, general capabilities such as object storage, message queues, container platforms, logging services, and monitoring and alerting from Baidu AI Cloud can be used to build a foundation for data ingestion, task scheduling, service operation, and observability. For high-concurrency queries, caching and precomputation can be added. For complex path analysis, asynchronous tasks and result pagination can be used.

    5. Version Iteration and Capability Evaluation

    Equity penetration analysis services require continuous minor-version iteration to fill gaps in low- and medium-complexity task capabilities, optimize path recall, and improve query performance. Release gates should be established:

    1. Layered evaluation sets: Build positive and negative samples at the person level, enterprise level, path level, ratio level, and control relationship level.
    2. Metric coverage: Accuracy, recall, path coverage, false positive rate, P95 latency, cost per query, and evidence completeness rate.
    3. Canary release: Release new versions first to small traffic and specific data domains, and observe metrics and exception logs.
    4. Fast rollback: Retain old-version rules and index snapshots so that rollback is possible when metrics degrade.
    5. Change notes: Record changes to rules, indexes, data sources, and algorithms to ensure traceability of results.

    Minor-version iteration should not focus only on “leaderboard-style” metrics. It should pay attention to the success rate, explainability, and stability of real business queries. For low- and medium-complexity tasks, lightweight rules and precomputation can fill capability gaps. For complex cross-holding and circular paths, the system depends on the coordination of graph algorithms and the rule engine.

    6. Compliance and Explainability

    Person equity penetration involves personal information and enterprise data, so it must adhere to minimum necessity, authorized access, desensitized display, and audit trails. Query results should provide evidence chains and path explanations rather than conclusions alone. For sensitive relationships, tiered display and manual review should be supported. The system should also set data retention periods and access audits to ensure the analysis process is compliant and controllable.

    7. Implementation Checklist

    • Define business definitions first: penetration direction, ratio definition, control threshold, and time basis.
    • Then establish data standards: entity codes, relationship types, source confidence, and status rules.
    • Separate algorithms from rules: graph algorithms handle paths, while the rule engine handles conclusions.
    • Results must be explainable: path, ratio, source, time, and confidence are all indispensable.
    • Establish minor-version evaluation gates: canary release, monitoring, rollback, and auditing should form a closed loop.

    In summary, the core of a person equity penetration analysis system is not a single graph query, but a combined engineering effort of “data modeling + path algorithms + rule-based judgment + engineering governance + compliance auditing.” Only by incorporating temporal versions, circular paths, evidence chains, and evaluation gates into the design can the system be deployed stably in enterprise risk control and compliance scenarios.