Skip to content

Content-Based Recommenders

The Content-Based Recommenders module of WarpRec focuses on generating recommendations by analyzing item attributes and comparing them with user preferences. These models rely on side information (e.g., tags, descriptions, metadata) to build user profiles and match them with similar items. They are particularly useful in cold-start scenarios or when collaborative signals are sparse.

In the following sections, you will find the list of available content-based models within WarpRec, together with their respective parameters.

API Reference

For class signatures, parameters, and source code, see the Content-Based API Reference.

Summary of Available Content Models

Category Model Description
Vector Space VSM Classical content-based model using TF-IDF and cosine similarity.

Vector Space

Vector Space models represent items and users as vectors in a multi-dimensional space derived from item features. Similarity is calculated based on the geometric distance or angle between these vectors.

VSM

VSM (Vector Space Model): A classical content-based recommender that represents items and users in a shared vector space. Items are encoded using TF-IDF or binary features, while user profiles are typically aggregated from consumed items. Recommendations are generated by computing the similarity (e.g., cosine) between user and item vectors. This model requires side information to function properly.

For further details, please refer to the paper.

models:
  VSM:
    similarity: cosine
    user_profile: binary
    item_profile: tfidf