Document Classification Device Using Sequential Filtering for Near-Duplicate Detection

The patent divides the document classification process into multiple stages with increasing calculation complexity. Simple classification processes are applied first to all documents, and only documents that remain unclassified after each stage proceed to the next more complex stage. This segmentation allows rapid filtering of obvious duplicates while progressively handling more subtle cases, thereby improving overall detection speed without requiring all documents to undergo the most complex analysis.

The patent applies simpler classification processes as preliminary steps before more complex ones. By performing basic classification first and eliminating obvious duplicates early in the process, the system prepares the data in advance for more sophisticated analysis only where needed. This preliminary action reduces the burden on subsequent complex processes and accelerates the overall detection workflow.

The patent applies multiple classification processes selectively rather than uniformly to all documents. Only documents that remain unclassified after simpler processes undergo more complex classification. This partial application of excessive action ensures that the most rigorous analysis is applied only where necessary to achieve accurate detection, while avoiding wasteful processing time on documents already classified by simpler methods.

The classification process is segmented into multiple passes with increasing complexity. Each pass handles a subset of documents that were not fully classified by previous passes. This segmentation allows the system to achieve high detection accuracy through multiple classification perspectives while minimizing total processing time by avoiding redundant application of complex processes to already-resolved documents.

The patent employs periodic classification processes with increasing complexity applied in sequence. Rather than using a single classification method, the system cycles through multiple classification processes, each building on the results of previous ones. This periodic application of progressively more sophisticated analysis maintains high processing speed by using simple methods first, then periodically applying more complex methods only to remaining documents, thereby achieving both speed and accuracy.

Simpler classification processes serve as preliminary actions that quickly resolve obvious cases. By performing these preliminary classifications first, the system establishes a baseline level of accuracy rapidly, then applies more complex processes only to the remaining ambiguous cases. This preliminary action approach maintains overall processing speed while improving detection accuracy for difficult cases that the preliminary processes could not resolve.