基于似然損失函數(shù)的組樣本排序?qū)W習方法
發(fā)布時間:2018-10-19 09:51
【摘要】:組樣本用于模型訓練,為排序?qū)W習方法的構(gòu)造提供一種新的思路.文中改進已有的組樣本排序?qū)W習方法,構(gòu)造組樣本損失函數(shù),用于排序?qū)W習模型的訓練.基于似然損失函數(shù),采用樣本偏序權(quán)重損失函數(shù)和最優(yōu)初始序列選擇方法,構(gòu)造基于神經(jīng)網(wǎng)絡(luò)的組排序?qū)W習方法,實驗證明文中方法能夠有效提高排序準確率.
[Abstract]:Group samples are used for model training, which provides a new idea for the construction of ranking learning methods. In this paper, we improve the existing group sample sorting learning method and construct the group sample loss function, which can be used to train the sorting learning model. Based on likelihood loss function, sample partial order weight loss function and optimal initial sequence selection method are used to construct group ranking learning method based on neural network. The experimental results show that the proposed method can effectively improve the accuracy of sorting.
【作者單位】: 大連理工大學人文與社會科學學部;大連理工大學計算機科學與技術(shù)學院;
【基金】:國家自然科學基金項目(No.61602078,61572102,61402075,61277370) 中國博士后科學基金項目(No.2016T90224,2015M581337) 中央高;究蒲袠I(yè)務(wù)費專項資金(No.DUT15RW401)資助~~
【分類號】:TP391.3;TP181
本文編號:2280780
[Abstract]:Group samples are used for model training, which provides a new idea for the construction of ranking learning methods. In this paper, we improve the existing group sample sorting learning method and construct the group sample loss function, which can be used to train the sorting learning model. Based on likelihood loss function, sample partial order weight loss function and optimal initial sequence selection method are used to construct group ranking learning method based on neural network. The experimental results show that the proposed method can effectively improve the accuracy of sorting.
【作者單位】: 大連理工大學人文與社會科學學部;大連理工大學計算機科學與技術(shù)學院;
【基金】:國家自然科學基金項目(No.61602078,61572102,61402075,61277370) 中國博士后科學基金項目(No.2016T90224,2015M581337) 中央高;究蒲袠I(yè)務(wù)費專項資金(No.DUT15RW401)資助~~
【分類號】:TP391.3;TP181
【相似文獻】
相關(guān)期刊論文 前1條
1 李小光;;混合損失函數(shù)支持向量回歸機的性能研究[J];西北大學學報(自然科學版);2011年02期
相關(guān)會議論文 前1條
1 吳佳金;楊志豪;林原;林鴻飛;;基于改進Pairwise損失函數(shù)的排序?qū)W習方法[A];第六屆全國信息檢索學術(shù)會議論文集[C];2010年
,本文編號:2280780
本文鏈接:http://sikaile.net/kejilunwen/zidonghuakongzhilunwen/2280780.html
最近更新
教材專著