Comparison Of Algorithms For Cross Feature Processing In Recommendation System

With the coming of big data era,recommendation system becomes more and more important.It is an important research direction of recommender system algorithm to find the cross features related to label from a large number of input features.At present,statistical machine learning algorithm is the main method to mine cross features in the industry.So,this paper focuses on the mechanism of different algorithms for cross feature mining in recommender systems.In this paper,we use cross feature rules to generate datasets for experiments.The experimental objects include some mainstream algorithms in the field of recommendation system,aiming to explore the performance principle and interpretability of cross feature mining.The experimental results show that some of the current mainstream algorithms are difficult to find the optimal solution in the hypothesis space when they solve the problem of mining cross features.This problem is especially serious in decision tree algorithm,so this kind of algorithm is not suitable for mining cross features.This paper innovatively explores how to solve the problem of mining cross features in recommendation system from three aspects of performance principle and interpretability(most of the research focuses on performance).We give a clear and reasonable mathematical model of cross feature problem.This paper provides a reference for the industry to mine the cross feature and the academic community to understand the algorithm mechanism.