Improved ALS Recommendation Algorithm Based on Interest-Value Model
Zhaoyang Wang, Jianjun Huang, Hongyuan Sun · Journal of Physics Conference Series · 2022
Abstract In the era of big data, all kinds of information around, how to effectively get the information others want and achieve recommendation is the original purpose of recommendation algorithm, traditional recommendation has collaborative filtering, content-based, model-based recommendation algorithm. Matrix factorization recommendation algorithm solves the problems of data sparsity and low space utilization brought by dimensional explosion. ALS(Alternating Least Squares) algorithm is one of the matrix factorization and the only one recommendation algorithm implemented in Spark, which is in line with the computation method of big data distributed computing. However, in the traditional ALS, there are still some shortcomings: the determination of the number of Implicit features does not give a clear definition, the information factor is lost after matrix factorization, which leads to the poor interpretability of the decomposed matrix, and the performance parameters of the ALS model are not optimal in terms of performance. Based on the above shortcomings, the IVM-ALS(ALS Based on Interest-Value Model) based on interest-value is proposed, an interest-value model is introduced, a solution to solve the Implicit feature number is introduced, and the feasibility of the feature matrix of the subdivided item is enhanced, so as to meet the actual requirements., And improve the algorithm performance than traditional ALS.