nimsearch VS ndup

You are welcome. Thanks are too rarely offered. :-)

You may also be interested in word stemming ( such as used by snowball stemmer in https://github.com/c-blake/nimsearch ) or other NLP techniques, but I don't know how internationalized/multi-lingual that stuff is, but conceptually you might want "series of stemmed words" to be the content fragments of interest.

Similarity scores have many applications. Weights on graph of cancelled downloads ranked by size might be one. :)

Of course, for your specific "truncation" problem, you might also be able to just do an edit distance against the much smaller filenames and compare data prefixes in files or use a SHA256 of a content-based first slice. ( There are edit distance algos in Nim in https://github.com/c-blake/cligen/blob/master/cligen/textUt.... as well as in https://github.com/c-blake/suggest ).

Or, you could do a little program like ndup/sh/ndup to create a "mirrored file tree" of such content-based slices then you could use any true duplicate-file finder (like https://github.com/c-blake/bu/blob/main/dups.nim) on the little signature system to identify duplicates and go from path suffixes in those clusters back to the main filesystem. Of course, a single KV store within one or two files would be more efficient than thousands of tiny files. There are many possibilities.

It's all in the Nim programming language, but if you prefer reading code or running diffs then you might get a vague sense of (some) low level nuts & bolts from: https://github.com/c-blake/nimsearch

You both might be interested in this little Nim program "framed" to frame & digest text for near-duplicate detection:

    https://github.com/c-blake/ndup/blob/main/framed.nim