Preliminary data-based matrix factorization approach for recommendation

Highlights：

• This paper is the first to constrain the learning procedure in matrix factorization by using imputed data.

• This paper is the first to discuss the relationship between imputed data and recommendation quality.

• We propose an assumption based on the analysis of user and item preferences in matrix factorization model. This assumption lets the learning user and item preferences be correctly constrained, thus leading to a more accurate learned model.

• We design two learning model according to the assumption. One firstly makes the original preferences get close to preliminary preferences, and then creates the concatenated preferences. The other one firstly creates the concatenated preferences, and then makes the original, preliminary and concatenated preferences get close to each other.

• Exhaustive experiments are conducted on five datasets: MovieLens 100k, MovieLens 1M, Netflix, Filmtrust and Jester. Experiment results show that PDMF outperforms the state-of-the-art methods by more than 10% in recommendation accuracy.

摘要

•This paper is the first to constrain the learning procedure in matrix factorization by using imputed data.•This paper is the first to discuss the relationship between imputed data and recommendation quality.•We propose an assumption based on the analysis of user and item preferences in matrix factorization model. This assumption lets the learning user and item preferences be correctly constrained, thus leading to a more accurate learned model.•We design two learning model according to the assumption. One firstly makes the original preferences get close to preliminary preferences, and then creates the concatenated preferences. The other one firstly creates the concatenated preferences, and then makes the original, preliminary and concatenated preferences get close to each other.•Exhaustive experiments are conducted on five datasets: MovieLens 100k, MovieLens 1M, Netflix, Filmtrust and Jester. Experiment results show that PDMF outperforms the state-of-the-art methods by more than 10% in recommendation accuracy.