MyMediaLite/Workshop projects

Some projects you could work on.

Visualization

• evaluation graphs
• dataset visualization

Input Formats

• SQLite

Parallelization

• parallel similarity computation
• 'naive' parallel learning for BPR-MF

Correlations

• Dice and Tyversky
• Jaccard index for binary feedback data
• similarity based on the Euclidean distance
• factor-based similarity

Algorithms

• determine the learn rate for SGD
• SGD learning for WRMF
• ALS-based matrix factorization
• fold-in for WRMF
• BPR-kNN
• WRMF with item bias
• k-means clustering of users and/or items
• Eigentaste 5.0
• integrating TrueSkill
• your favorite algorithm
• some algorithm from this year's conference or its workshops

Evaluation

• expected reciprocal rank
• Kendall's Tau
• Spearman rank coefficient
• add fold-in evaluation to the command-line tools
• your favorite missing evaluation protocol or measure

Item Recommendation

• attribute-based pre-filters
• popularity-based pre-filters
• time-based pre-filters
• post-filters

Under the Hood

• experiment with Math.Net Numerics matrix data types

Compatibility

• Comparing abstractions and APIs in LensKit, Mahout, and MyMediaLite
  • data models and storage
  • rating scales
  • recommenders
  • training and updates
  • correlations and similarities
  • evaluation protocols
  • handling of attributes and relations
  • portability and interfacing