Incremental matrix factorization for simultaneous learning from parallel data streamsJakomin, Martin (Avtor) Bosnić, Zoran (Mentor) Curk, Tomaž (Komentor) machine learningmatrix factorizationdata streamsdata fusionincremental learningrecommender systemssynthetic data generatorMatrix factorization techniques have proven to be useful and reliable for solving largescale machine learning problems. The data sparsity and cold-start problems found in real-world applications, such as recommender systems, can be indirectly alleviated by considering multiple heterogeneous data sources, while at the same time the successful utilization of data fusion resolves in a higher predictive accuracy. However, incrementally handling models upon multiple data streams remains a crucial and only partially solved problem. This work presents one way of fusing multiple data streams through matrix factorization. Our proposed method models heterogeneous and asynchronous data streams and provides predictions in real time. As a result of incremental updating, the proposed method successfully adapts to changes in data concepts, while application of data fusion improves prediction accuracy and reduces effects of the cold-start problem. Using the proposed methodology we develop a streaming recommender system and show how prediction accuracy can be substantially increased by considering multiple data sources. Nevertheless, evaluating data fusion, recommender and other incremental algorithms, such as our presented method, is inherently difficult due to the scarcity of obtainable data sources. In order to address this problem, we conjointly propose a synthetic data generator, capable of generating multiple temporal and inter-dependent data streams of relational data. Data streams generated in this way successfully mimic real-life datasets in terms of statistical data properties and comparable performance of various machine learning models. Proposed methodologies help in development of solutions for collective modeling of streaming data in real-time. Apart from recommender systems, the versatility of matrix factorization further allows for its use in solving several other machine learning problems, such as dimensionality reduction, clustering and classification.20192019-12-09 15:25:02Doktorsko delo/naloga113169sl