- Understanding the sources and different types of user-generated data [k2]
- Gathering user-generated and other data needed to build a recommender system (e.g., user-item interactions, user side information, item side information, context data) [k3]
- Understanding the key concepts and techniques underlying recommender systems, such as user-item interactions, similarity metrics, matrix factorization, and hybridization methods [k2]
- Understanding the most common recommendation strategies (e.g., collaborative filtering, content-based filtering, context-aware recommendation), their application domains, and types of data they operate on [k2]
- Evaluating recommender systems / content ranking models via accuracy- and beyond-accuracy metrics; deciding which metric is suited for which problem [k5]
- Critically reflecting on ethical concerns (e.g., privacy, biases, or fairness issues) in user-generated data and recommender systems [k4]
|
- Sources of user-generated data
- Tasks, applications, and use cases that leverage user-generated data
- Differentiating retrieval, browsing, and recommendation
- Foundations of recommender systems (e.g., explicit versus implicit feedback on items, similarity metrics, user-item rating matrix)
- Key concepts and methods for collaborative filtering, including memory-based and model-based variants
- Key concepts and methods for content-based filtering, focusing on the text modality, including natural language processing methods (casefolding, stemming, lemmatization, stopping, topic modeling)
- Key concepts and methods for graph-based recommenders, including graph-based transitivity, and applying techniques from social network mining (edge and node measures)
- Strategies to create hybrid recommenders, including parallelized, monolithic, and pipelined designs
- Different evaluation strategies for recommender systems (offline testing, online testing, user studies); most important accuracy metrics (e.g., F1, NDCG, MRR, MAP) and beyond-accuracy metrics (e.g., diversity, novelty, coverage)
- Personalization, user-awareness, and context-awareness in recommender systems
- Psychology-informed recommender systems (cognition-inspired, personality-aware, and affect-aware recommenders)
- Biases and fairness in recommender systems, including categorization of biases, relation to fairness, calibration metrics, and strategies to mitigate harmful biases (pre-, in-, and post-processing)
|
