{"paper_id":"2502.19305","title":"Corporate Fraud Detection in Rich-yet-Noisy Financial Graph","authors":["Shiqi Wang","Zhibo Zhang","Libing Fang","Cam-Tu Nguyen","Wenzhong Li"],"paper_date":"2025-02-26","paper_type":"empirical","paper_url":"https://arxiv.org/pdf/2502.19305v2","page_url":"https://thequant.space/flowcharts/corporate-fraud-detection-in-rich-yet-noisy-financial-graph/","scores":{"empirical_rigor":8,"math_complexity":6.5,"quadrant":"Holy Grail"},"proposes":{"status":"available","statement":"The paper proposes Knowledge-enhanced GCN with Robust Two-stage Learning (KeGCN_R), a graph-based model that uses knowledge graph embeddings and a two-stage learning process to handle information overload and hidden fraud.","quote":"we propose a novel graph-based method, namely, Knowledge-enhanced GCN with Robust Two-stage Learning (${\\rm KeGCN}_{R}$), which leverages Knowledge Graph Embeddings to mitigate the information overload and effectively learns rich representations.","summary":"Corporate fraud detection aims to automatically recognize companies that conduct wrongful activities such as fraudulent financial statements or illegal insider trading. Previous learning-based methods fail to effectively integrate rich interactions in the company network. To close this gap, we…","strategy":null},"assumptions":{"status":"available","items":[{"assumption":"Regulators prefer precision in declaring frauds, leading to an asymmetric noise structure where many fraud cases remain undetected (hidden fraud).","type":"model","stated":"explicit","quote":"It has an asymmetric noise structure, which is based on the fact that regulators prefer precision in declaring frauds with the cost of letting some fraud cases go undetected (hidden frauds);"},{"assumption":"Label noise depends on both the instance and its neighborhood (Instance and Neighbor Dependence).","type":"model","stated":"explicit","quote":"It is instance and neighborhood dependence (see Figure 1)."},{"assumption":"A company labeled as non‑fraud for at least 8 years can be treated as a clean non‑fraud data point.","type":"data","stated":"explicit","quote":"we can reasonably trust a case that has been labeled as non-fraud for at least 8 years and consider it as a clean non-fraud data point."}]},"datasets":{"status":"available","items":[{"name":"Financial graphs from 18‑year records of Chinese A‑share companies","provider":"CSMAR","kind":"fundamentals","access":"public","period":"2003-2020","universe":"Chinese A‑share listed companies (Shanghai and Shenzhen exchanges)","quote":"We first extract all the annual financial records of listed companies on two stock exchange markets (Shanghai and Shenzhen) during the period from 2003 to 2020.","known_id":"csmar"}],"known":[{"id":"the-graph","name":"The Graph (subgraphs)","provider":"The Graph","kind":"crypto-onchain","access":"public","quote":"Multi-path Weighted Convolution Layers This phase aims to learn representations from com-pany subgraphs Gk = ⟨Vc, Wmp k , Xit⟩, each obtained from a meta-path ρk and an initialization strategy (Xatt or Xke).","mentions":16,"used_in":1}],"sample":{"start":null,"end":null,"market":null,"asset_class":null,"frequency":null},"data_resolution":"weekly-monthly"},"transaction_costs":{"status":"not_applicable","modeled":"n/a","headline_net_of_costs":null,"detail":null,"quote":null,"source":"front_matter","rule":"R3","candidates":{"front_matter":"n/a","evidence_card":"n/a","evidence_card_quoted":false},"meaning":"the paper reports no trading or execution result"},"results":{"status":"available","evaluation_type":"forecast-accuracy","findings":[{"finding":"KeGCN_R achieves the best AUC across all three datasets, outperforming all baselines, especially on the MBM dataset with the most serious hidden fraud problem.","kind":"performance","quote":"our method outperforms all baselines and achieves the best AUC across all the datasets, suggesting that our model can better handle the challenges of corporate fraud detection. Particularly, our method outperforms other methods by a large margin in the MBM dataset, which is the one with the most serious hidden fraud pr"},{"finding":"GNN‑based methods generally perform better than DNNs, indicating that relational information between companies improves detection results.","kind":"predictive-accuracy","quote":"GNN‑based methods generally perform better than DNN, showing that relational information between companies can help improve detection results."}],"reported_performance":null,"out_of_sample":{"method":"holdout","headline_out_of_sample":null}},"independent_evidence":{"status":"none_found","supports":[],"contradicts":[],"qualifies":[],"compared_with":28,"citations":{"cited_by_in_corpus":0,"by_independent_authors":0,"how_cited":{"used_as_baseline":0,"uses_method_or_data":0,"extends":0,"discusses_findings":0,"mentions":0,"unclassified":0},"independent_papers":[],"scope":"Citations between papers in this corpus only; citations from outside it are not counted."}},"implementation":{"status":"available","public":true,"code_url":"https://github.com/wangskyGit/KeHGN-R","note":null},"replication":{"status":"none_found","card":null,"note":"No replication of this paper has been run by thequant.space yet."},"extraction":{"text_source":"fulltext","model":"openai/gpt-oss-120b","version":1,"extracted_at":"2026-11-14T17:44:06Z"},"provenance":{"claims":"Everything here is what the authors state in the paper. Nothing was re-run or reproduced by the platform.","quotes":"Assumptions, datasets, cost detail and findings were extracted by a language model. Every quote is verbatim paper text. Where an item has `support`, \"full\" means every figure in it appears in the quoted sentences and \"partial\" means a figure is stated elsewhere in the paper; items with a figure found nowhere in the paper were discarded.","datasets_known":"datasets.known lists well-known vendors and public datasets by a stable id, found by a dictionary over the full text where the sentence says the data is used; it is independent of the model and covers only names in the dictionary.","flags":"transaction_costs.modeled, results.evaluation_type and results.out_of_sample.method are platform classifications (structured-decision models); see https://thequant.space/score-guide/#penalty-flags","comparisons":"supports / contradicts / qualifies are a language model's reading of two papers' quoted findings, restricted to closely related papers that share no author. They are pointers to read, not verified replications, and coverage is partial.","implementation":"code_url is a repository the paper itself presents as its own (rule-scored, ambiguous cases judged by a model), checked for liveness when it was recorded."},"docs":"https://thequant.space/agents/"}