Jethro's Braindump

Recommender Systems

Are We Really Making Much Progress? (Dacrema, Cremonesi, and Jannach, n.d.)

The authors analyzed various recent publications on recommendation systems techniques, and found that these have:

  1. Weak baselines
  2. Establish weak methods as baselines
  3. Are outperformed by simple, sometimes non-neural approaches

The simple approaches that work well include ItemKNN, a collaborative-filtering approach that uses k-nearest neighbours and item-item similarities:

\begin{equation} s_{ij} = \frac{r_i \dot r_j}{\lvert r_i \rvert \lvert r_j \rvert + h} \end{equation}

To alleviate these issues:

  1. Use appropriate evaluation methods
  2. Evaluate on appropriate datasets (size is important)
  3. Release reproducible code

Bibliography

Dacrema, Maurizio Ferrari, Paolo Cremonesi, and Dietmar Jannach. n.d. “Are We Really Making Much Progress? a Worrying Analysis of Recent Neural Recommendation Approaches.” http://arxiv.org/abs/1907.06902v1.