US 12,248,457 B1
Detecting temporal anomalous data using dependency modeling
Hao Xiang Wu, Beijing (CN); Rong Zhao, Beijing (CN); Zhe Yan, Beijing (CN); Li Li Guan, Beijing (CN); and Li Bo Zhang, Beijing (CN)
Assigned to International Business Machines Corporation, Armonk, NY (US)
Filed by INTERNATIONAL BUSINESS MACHINES CORPORATION, Armonk, NY (US)
Filed on Aug. 23, 2023, as Appl. No. 18/454,115.
Int. Cl. G06F 16/23 (2019.01)
CPC G06F 16/2365 (2019.01) 20 Claims
OG exemplary drawing
 
1. A computer-based method of detecting anomalous data using dependency modeling, the method comprising:
within a target data environment, identifying implicit and explicit references between data contained in one or more data files, explicit references referring to data contained in the data files that include the same data metric or value and associated file features, implicit references referring to data contained in the data files which contain similar data metrics but have slight differences with regard to associated file features that cause similar data metrics to be overlooked;
determining dependency relationships between data fields in the data contained in the one or more data files;
constructing computational graphs depicting the determined dependency relationships as series of related data fields in the one or more data files;
identifying series of associated computational graphs within the constructed computational graphs;
calculating abnormality degree values for each of the data fields within the constructed computation graphs; and
in response to detecting an anomalous data field having a calculated abnormality degree value above a threshold value, calculating contribution values for a series of associated component data fields in the one or more data files to identify a root cause for the detected anomalous data field.