machine learning - Recommender approach and algorithms for cold-start -
we looking @ building recommender system our brand-new learning management system. there bunch of users , items (learning modules) onboarded, no ratings yet - typical cold start problem.
to begin with, thinking of using simple item-based similarity using item attributes (tags, category, etc.) idea switch more robust collaborative filtering ratings start coming in.
questions:
- is approach? there recommended ml pattern handle such cold-start conditions?
- to realise item-based similarity, right algorithm? say, cosine similarity. however, please note there no "matrix". should try use standard ml algorithm or maybe roll our own?
your approach good. start unsupervised learning algorithm such 'k-nearest neighbors classifier'. if team doesn't know first thing ml, recommend read tutorial http://www.astroml.org/sklearn_tutorial/general_concepts.html . uses python , great library called scikit-learn. there andrew's ng course (https://www.coursera.org/learn/machine-learning/) although not cover recommendation systems.
i go pearson correlation algorithm (https://en.wikipedia.org/wiki/pearson_product-moment_correlation_coefficient) , suffices me problems. problem approach is linear. have read orange data mining tool provides many correlation measures. using find 1 best data. advice against using own algorithm.
there older question provides further information on matter: how can implement recommendation engine?
Comments
Post a Comment