Content-Based Models

Content-Based Models#

Recommendations from item/user text and metadata rather than (or in addition to) collaborative signals.

Production models (CI-tested)#

TFIDFRecommender#

TF-IDF similarity over item text documents.

Import

from corerec.engines.content_based import TFIDFRecommender

Tutorial

TF-IDF Tutorial

Fit API

Item IDs + document dict or corpus

 
from corerec.engines.content_based import TFIDFRecommender

docs = {
    101: "wireless bluetooth headphones noise cancelling",
    102: "running shoes lightweight marathon",
    103: "python machine learning tutorial",
}
model = TFIDFRecommender()
model.fit(items=list(docs.keys()), docs=docs)

# Similar items by content
similar = model.recommend(item_id=101, top_k=5)

Use cases: article/product catalogs, cold-start retrieval, hybrid pipelines (TF-IDF retrieval → ranker).

Sandbox models (experimental)#

Model

Tutorial

MIND-Content

MIND Content

Import path is documented in the tutorial (corerec.sandbox.*).

When to use#

  • Rich text/metadata per item

  • New items with no interaction history

  • Candidate generation stage before a ranker

See also#

Contents