Research And Implementation Of Ranking Model Based On Deep Learning

Ranking problem is one of the key issues in information retrieval and recommendation system.The accuracy and rationality of the ranking results directly affect the quality of retrieval and recommendation.Therefore,it is of great value to study the ranking model and the optimization method of ranking results.Traditional ranking methods are available including determined by artificial scoring formula,based on the importance and based on the correlation and so on.However,these methods have some disadvantages such as relying on experts' experience,the lack of objectivity and without considering the factors synthetically.In recent years,deep learning has become a hot topic of academic research and has made a series of remarkable achievements.Compared with the traditional machine learning algorithm models,the deep neural network model has stronger fitting ability for complex functions and stronger feature extraction and presentation learning ability.With the development of machine learning technology,using machine learning algorithm model to solve the ranking problem has gradually become the mainstream practice in academia and industry.In this context,a ranking model based on deep learning is proposed.The first chapter describes the background and significance of the research.The second chapter introduces the background theories and related technologies of deep learning.In the third chapter,this paper introduces the commonly used ranking methods,and analyzes the key problems existing in the ranking model.The fourth chapter describes the paper's main research results.Fusion methods such as Boosting and Stacking are applied to combining three kinds of algorithms,including the logistic regression,deep neural network and gradient boosting tree to a fusion model for scoring,and improve the ranking effect of the model.The ranking model adopts the layered structure design,which takes into account the accuracy and efficiency of the ranking.This paper presents a method of data encoding of historical behaviors based on recurrent neural network model,which solves the disadvantages of the previous methods such as feature vector length is not fixed and the information loss problem.The fifth chapter gives the detailed design and implementation of the model,and proposes a subway unilateral transaction processing method based on the ranking model.The experimental results show that the fusion model has higher prediction accuracy compared with other algorithms.At the same time,the subway unilateral transaction processing method based on the ranking model is more reasonable than the existing methods,verifying the validity and practicability of this ranking model.