STRATA
Skip to study
XLVII

SOURCE RECORD / 2006

Compression → classification

DocumentedR01 · 2026-11-17
Scope of this record

The journal reports the application and evaluation. No XLVII reproduction or claim of present-day superiority.

Statement

A published study evaluates dynamic Markov compression and prediction by partial matching for spam filtering. [source]

Equation

Unrecorded

The abstract connects compression models with classification; no mathematical equivalence is asserted here.

Observation

The 2006 study treats messages as character or binary sequences and evaluates DMC and PPM as spam classifiers. [source]

Progressive questions

What remains useful when the data distribution, baselines, and resource limits change?

Responses

The abstract reports an empirical comparison; its findings belong to the authors’ evaluation conditions. [source]

Interpretation

This is a documented reuse of compression machinery, not a claim that all prediction and compression problems are equivalent.

Conclusion

The historical application is recorded. Contemporary usefulness remains an open study.

Application

Link a new, separately versioned experiment rather than overwrite the source report.

Work through a related example.

IV.15 / Compression-based classification and contemporary reuse

Compare two class-conditioned descriptions.

Train two illustrative text models on separate, labeled collections and score a held-out message under each. Record which model assigns the shorter effective description and whether headers or repeated templates dominate that difference. Split related messages together when forming training and test sets so that near-duplicates do not make the evaluation artificially easy.

X.05 / Information theory, coding, and compression

Make a codebook part of the size calculation.

Suppose a short document uses four words repeatedly. Replacing each word with a short symbol can shrink the body, but the decoder also needs the symbol-to-word dictionary. Count both components and compare the total with the original. Repeat with a much longer document to see how shared overhead changes the result.

Follow earlier and later work.

Before

  • Applies · KNOWNDynamic Markov modelling

    Historical method/application relationship only; no equivalence or causal discovery claim.

  • Proposed connection · UNKNOWNInformation is not meaning

    Editorial research connection to investigate; not a verified correspondence, causal lineage, or mathematical equivalence.

After

Follow the chain
  1. Dynamic Markov modelling · 1 link away · KNOWN
  2. Compression, reopened · 1 link away · UNKNOWN
  3. Information is not meaning · 1 link away · UNKNOWN
  4. What survives a replication? · 2 links away · UNKNOWN
  5. Memory as a collective state · 2 links away · UNKNOWN
  6. Erasure has conditions · 2 links away · UNKNOWN
  7. Compute has a budget · 3 links away · UNKNOWN
  8. Prediction and residual error · 3 links away · UNKNOWN
  9. Reversibility and the record · 3 links away · UNKNOWN
  10. Attention without recurrence · 4 links away · UNKNOWN
  11. The chemical compass · 4 links away · UNKNOWN

Unfinished.

Present-day utility

A versioned benchmark with specified data, budgets, baselines, and error constraints.

Change the tested conditions explicitly; retain the earlier scope.

Bibliographic page-range mismatch

The author bibliography lists 2699–2720; the journal page lists 2673–2698. The journal’s range is used in this record, with the discrepancy retained. [source][source]

Check the journal issue and original article before closing the discrepancy. This is a citation discrepancy, not a contradiction in the scientific mechanism.

Placement.

Technology

A later application is documented; this is not evidence that the original work was discarded.

Sources.

  1. Publications of G. V. Cormack

    University of Waterloo · author bibliography · Journal publications: December 1987 entry, Computer Journal 30:6, 541–550.

    Bibliographic listing read; the original 1987 full paper was not inspected in this pass. Checked 2026-11-17.

  2. Spam Filtering Using Statistical Data Compression Models

    Journal of Machine Learning Research · Volume 7(97), 2673–2698 (2006); abstract.

    Journal metadata and abstract read; no independent replication performed. Checked 2026-11-17.

Record history
  1. 2026-11-17 / R1 / CREATED

    Source record created.