| Learning to rank is a technique that uses machine learning techniques to sort search results.It has played a major role in many application fields including information retrieval and data mining,and has received much attention in recent years.Learning to rank assumes that each training instance is associated with a reliable tag,and it provides an excellent automation framework for feature combinations that can query for dependency features,such as assigning scores to documents through existing search engines,as well as querying individual features.This paper investigates two basic types of auxiliary information and introduces them into corresponding learning algorithms.The probabilistic ordering model used in the study first combines the generalized linear model and the Plackett-Luce(P-L)model to deal with the instance-based ranking problem of multi-category labels.The goal is to train to learn a ranking function that trains and uses the maximum likelihood estimation method to estimate the label ordering and iteratively trains the ranking function,which produces a complete sort on the entire set of labels.Then,the subsequent sorting function is optimized by two new algorithms of the ListMLE algorithm with improved auxiliary information,which can be iteratively trained the rank function through the ground-truth labels and the level of expert knowledge of the annotator.In addition,learning ordering can be explored from the crowdsourcing label.The improved algorithm has been tested on both synthetic and actual data.The results show that the improved method is significantly better than the average method and the existing crowdsourcing g regression method.For top-k technology with a large dataset of data,the use of top-k technology for ranking learning is helpful in terms of time efficiency through comparison of full sorting and top-k technology sorting.The result is comparable to the learning result of a fully ordered tag with real values. |
PDF Full Text Request