基于多任务学习的机票价格预测模型  

Airfare price prediction based on multi-task learning

作　　者：卢敏[1,2] 贾玉璇[1] LU Min;JIA Yu-xuan(College of Computer Science and Technology,Civil Aviation University of China,Tianjin 300300,China;The Key Laboratory of Smart Airport Theory and System,Civil Aviation Administration of China,Tianjin 300300,China)

机构地区：[1]中国民航大学计算机科学与技术学院,天津300300 [2]中国民用航空局民航智慧机场理论与系统重点实验室,天津300300

出　　处：《计算机工程与设计》2023年第8期2459-2464,共6页Computer Engineering and Design

基　　金：中央高校基本科研业务费专项资金基金项目(3122014D032)。

摘　　要：针对现有机票销售模型忽视不同票价等级需求问题,考虑到机票需求的影响,提出一种多任务学习模型预测机票价格。在机票价格预测中引入辅助任务机票需求预测,建立多任务学习网络,通过共享相关任务在日、周、半月、月等水平上的多尺度需求特征,分析不同周期需求特征的影响。在六千万条记录的真实数据集上的实验结果表明,较之基准算法,该模型在准确率和F1分数方面提高了将近6%,验证了多任务学习模型的有效性。To solve the problem that the existing ticket pricing strategy ignores the demand for different fare classes,a multi-task learning model for airfare price prediction was proposed to consider the impact of ticket demand.The ticket demand prediction was introduced as an auxiliary task to build a multi-task learning network in airfare price prediction.The impact of different periodic demand characteristics was analyzed by sharing the multi-scale demand characteristics of related tasks at the daily,weekly,semi-monthly,and monthly levels.Multiple experiments were conducted on real-word dataset with sixty millions of records.The results show that the improvements over the baselines in terms of accuracy and F1-score are nearly 6%,verifying the effectiveness of the multi-task learning model.

关 键 词：机票价格预测 机票需求 多尺度需求特征 多任务学习 卷积神经网络 残差网络 分类器模型 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]