Nonparametric Bayes pachinko allocation model
Model Families AIMech:NonparametricBayesPachinkoAllocationModel DRAFT
Aliases: NPB PAM, nonparametric PAM
A Bayesian pachinko topic-model family using coupled hierarchical Dirichlet-process allocations to infer occupied supertopics and shared subtopics, with subtopic allocation groups determined by latent supertopic assignments.
Identity Notes & Concept Scope
Scoped to Li, Blei and McCallum's UAI2007 nonparametric four-level PAM construction. Its countably unbounded prior supports data-dependent finite topic occupancy while retaining root, supertopic, subtopic and word levels. A two-level allocation franchise selects shared supertopics; a three-level franchise selects shared subtopics conditional on the selected supertopic. This differs from fixed-count PAM, internal-word-emission hPAM, a single-document-path nested Chinese restaurant process and a generic HDP with externally fixed groups. No normalized candidate identity matched all 442 pinned AIO IDs, labels and scoped synonyms or 435 preceding native IDs, names and aliases. Full-checkout searches included ignored and hidden files in both checkouts; two matching lines per checkout came from one unique prior discovery-ledger content with a source URL and pending extension cohort. All matching excerpts were read. All 205 semantic label/synonym hits and 18 annotated AIO nodes with outgoing edges were byte-verified unchanged from prior readings. Twenty-six contextual YAMLs comprise 24 unchanged prior readings and two fully read final round104 native targets verified against saved review inputs. No candidate record is present within this bounded pinned-snapshot and maintained-corpus search.
Claims & Grounded Evidence
3 Claims
Mechanism
#coupled-allocation-franchises
Each token selects a document-local entryway; a newly created entryway selects a globally shared category representing a supertopic. Conditional on that category, the token selects a local document/category table. A new table selects a category-specific corpus-shared menu, and a new menu selects a globally shared dish representing a subtopic. Existing auxiliary objects retain their assigned category, menu or dish. A dish has a shared categorical word distribution with a Dirichlet base. Super-topic assignments therefore dynamically define groups for the lower franchise.
Mechanism
#collapsed-franchise-inference
The source proposes jointly sampling a token's entryway, category, table, menu and dish assignments, combining franchise predictive factors with a corpus dish-word predictive factor. Customer, entryway, table and menu counts play different roles; a correct leave-one-out calculation must consistently remove emptied auxiliary objects. Positive concentration parameters receive Gamma priors and are resampled by a cited HDP procedure. The word factor follows the dish identity in Table 1, despite nearby prose loosely calling that object a menu.
Mechanism
#occupied-sparsity-boundary
Reported sparse connectivity records which supertopic/subtopic pairs were occupied in finite assignments. It need not impose structural zero transition probabilities: a new local table and category menu can select a globally existing dish not previously used by that supertopic. In a synthetic positive-concentration example, such an unobserved edge retains predictive probability 1/24. Learning occupied topic counts within the fixed four-level organization is distinct from learning arbitrary DAG topology or guaranteeing the true finite number of topics.
⚠️ Curated Limitations & Negative Results
- Nonparametric means an unbounded prior on available components, not absence of hyperparameters or guaranteed recovery of an objectively correct topic count. Finite chains, concentration priors and auxiliary allocation bookkeeping affect inference.
- The source's general correlation rhetoric does not establish universal DAG expressivity, literal independence of Dirichlet topic proportions, or exact CTM equivalence. The record follows the explicit four-level franchise construction.
- Synthetic recovery and selected comp5, Rexa and NIPS studies provide bounded evidence. Likelihood is estimated from a finite mixture of generated documents; non-significant pairwise differences do not prove model equivalence. The comp5 HDP baseline uses document group information unavailable to the other models.
- An unused edge can remain possible under the prior. Proposed pre-pruning and large-scale extensions are not evaluated exact algorithms in this source. No author code or benchmark was executed.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:Model
Model
|
broad_match | This source-defined latent-variable model family is narrower than the pinned Model class. |
AIO:ProbabilisticTopicModel
Probabilistic Topic Model
|
related_match | The model supports probabilistic topic modeling; the pinned target is task-scoped, so related match preserves the model/task distinction. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Wei Li, David Blei, Andrew McCallum
|
primary paper
full_text
|
sha256:a7c2ca36e349…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-12T00:36:41Z
Updated: 2026-10-12T00:36:41Z