面向点击率预测的自注意力深度域嵌入因子分解机

TP391; 点击率(CTR)预测通过预测用户对广告或商品的点击概率,实现数字广告精准推荐.针对现有CTR模型存在原始嵌入向量未精化、特征交互方式偏简单的问题,本文提出自注意力深度域嵌入因子分解机(self-atten-tion deep field-embedded factorization machine,Self-AtDFEFM)模型.首先,通过多头自注意力对原始嵌入向量加权,精化出关键低层特征;其次,构建深度域嵌入因子分解机(FEFM)模块,设计域对对称矩阵以提升不同特征域之间的交互强度,为高阶特征交互优选出低阶特征组合;再次,基于低阶特征组合构建深度神经网络(DNN),完成隐式高阶...

Full description

Saved in:
Bibliographic Details
Published in工程科学与技术 Vol. 56; no. 5; pp. 287 - 296
Main Authors 李广丽, 叶艺源, 许广鑫, 张红斌, 吴光庭, 吕敬钦
Format Journal Article
LanguageChinese
Published 华东交通大学信息与软件工程学院,江西南昌 330013 01.09.2024
Subjects
Online AccessGet full text
ISSN2096-3246
DOI10.12454/j.jsuese.202201373

Cover

Abstract TP391; 点击率(CTR)预测通过预测用户对广告或商品的点击概率,实现数字广告精准推荐.针对现有CTR模型存在原始嵌入向量未精化、特征交互方式偏简单的问题,本文提出自注意力深度域嵌入因子分解机(self-atten-tion deep field-embedded factorization machine,Self-AtDFEFM)模型.首先,通过多头自注意力对原始嵌入向量加权,精化出关键低层特征;其次,构建深度域嵌入因子分解机(FEFM)模块,设计域对对称矩阵以提升不同特征域之间的交互强度,为高阶特征交互优选出低阶特征组合;再次,基于低阶特征组合构建深度神经网络(DNN),完成隐式高阶特征交互;然后,围绕精化后的嵌入向量,联合多头自注意力与残差机制堆叠多个显式高阶特征交互层,通过自注意力捕获同一特征在不同子空间上的互补信息,完成显示高阶特征交互;最后,联合显式与隐式高阶特征交互实现点击率预测.在Criteo和Avazu两大公开数据集上,将Self-AtDFEFM模型与主流基线模型在AUC和Lo-gLoss指标上进行对比实验;为Self-AtDFEFM模型调制显式高阶特征交互层层数、注意力头数量、嵌入层维度及隐式高阶特征交互层层数等参数;对Self-AtDFEFM模型进行消融实验.实验结果表明:在两大数据集上,Self-AtDFEFM 模型的AUC、LogLoss均优于主流基线模型;Self-AtDFEFM模型的全部参数已调为最佳;各模块形成合力以促使Self-AtDFEFM模型性能达到最优,其中显示高阶特征交互层的作用最大.Self-AtDFEFM模型各模块即插即用,易于构建和部署,且在性能与复杂度之间取得平衡,具备较高实用性.
AbstractList TP391; 点击率(CTR)预测通过预测用户对广告或商品的点击概率,实现数字广告精准推荐.针对现有CTR模型存在原始嵌入向量未精化、特征交互方式偏简单的问题,本文提出自注意力深度域嵌入因子分解机(self-atten-tion deep field-embedded factorization machine,Self-AtDFEFM)模型.首先,通过多头自注意力对原始嵌入向量加权,精化出关键低层特征;其次,构建深度域嵌入因子分解机(FEFM)模块,设计域对对称矩阵以提升不同特征域之间的交互强度,为高阶特征交互优选出低阶特征组合;再次,基于低阶特征组合构建深度神经网络(DNN),完成隐式高阶特征交互;然后,围绕精化后的嵌入向量,联合多头自注意力与残差机制堆叠多个显式高阶特征交互层,通过自注意力捕获同一特征在不同子空间上的互补信息,完成显示高阶特征交互;最后,联合显式与隐式高阶特征交互实现点击率预测.在Criteo和Avazu两大公开数据集上,将Self-AtDFEFM模型与主流基线模型在AUC和Lo-gLoss指标上进行对比实验;为Self-AtDFEFM模型调制显式高阶特征交互层层数、注意力头数量、嵌入层维度及隐式高阶特征交互层层数等参数;对Self-AtDFEFM模型进行消融实验.实验结果表明:在两大数据集上,Self-AtDFEFM 模型的AUC、LogLoss均优于主流基线模型;Self-AtDFEFM模型的全部参数已调为最佳;各模块形成合力以促使Self-AtDFEFM模型性能达到最优,其中显示高阶特征交互层的作用最大.Self-AtDFEFM模型各模块即插即用,易于构建和部署,且在性能与复杂度之间取得平衡,具备较高实用性.
Abstract_FL Objective Click-through rate(CTR)prediction realizes accurate recommendation of digital advertisements by predicting the user's click probabil-ity on advertisements or commodities.However,current CTR prediction models have the following key issues.First,the raw embedding vectors have not been fully refined.Second,the corresponding feature interaction method is too simple.As a result,the performance of the models is heavily restricted.To alleviate these issues,a novel CTR model named self-attention deep field-embedded factorization machine(Self-AtDFEFM)is proposed. Methods First,a well-known multi-head self-attention mechanism is employed to capture the implicit information of the raw embedding vectors on different sub-spaces,and the corresponding weight is calculated to further refine the key low-level features.Second,a novel field-embedded factorization machine(FEFM)is designed to strengthen the interaction intensity between different feature fields by the field pair symmetric mat-rix.The key low-order feature combinations are fully optimized by the FEFM module for the subsequent high-order feature interaction.Third,a deep neural network(DNN)is built based on the low-order feature combinations to complete implicit high-order feature interaction.Finally,both the explicit and implicit feature interactions are combined together to implement CTR prediction. Results and Discussions Extensive experiments have been performed on the two public available datasets,namely Criteo and Avazu.First,the proposed Self-AtDFEFM is compared with numerous state-of-the-art baselines on the AUC(area under curve)and LogLoss metrics.Second,all parameters of Self-AtDFEFM was tuned,and the parameters included the number of the explicit high-order feature interaction layers,the number of the attention heads,the embedding dimension,and the number of the implicit high-order feature interaction layers.Further,ablation experi-ments of our model were completed.The results of the experiments showed that:the Self-AtDFEFM model outperformed mainstream baseline models on the AUC and LogLoss metrics;all parameters of Self-AtDFEFM have been adjusted to their optimal values;each module form a kind of joint force to improve the final CTR prediction performance.Notably,the explicit high-order feature interaction layer plays the most important role in Self-AtDFEFM. Conclusions Each module of Self-AtDFEFM is plug-and-play,that is,the Self-AtDFEFM is easier to build and deploy.Hence,Self-AtDFEFM achieves a good trade-off between prediction performance and model complexity,making it highly practical.
Author 李广丽
许广鑫
张红斌
吴光庭
叶艺源
吕敬钦
AuthorAffiliation 华东交通大学信息与软件工程学院,江西南昌 330013
AuthorAffiliation_xml – name: 华东交通大学信息与软件工程学院,江西南昌 330013
Author_FL ZHANG Hongbin
LYU Jingqin
XU Guangxin
LI Guangli
WU Guangting
YE Yiyuan
Author_FL_xml – sequence: 1
  fullname: LI Guangli
– sequence: 2
  fullname: YE Yiyuan
– sequence: 3
  fullname: XU Guangxin
– sequence: 4
  fullname: ZHANG Hongbin
– sequence: 5
  fullname: WU Guangting
– sequence: 6
  fullname: LYU Jingqin
Author_xml – sequence: 1
  fullname: 李广丽
– sequence: 2
  fullname: 叶艺源
– sequence: 3
  fullname: 许广鑫
– sequence: 4
  fullname: 张红斌
– sequence: 5
  fullname: 吴光庭
– sequence: 6
  fullname: 吕敬钦
BookMark eNotj01LAkEch-dgkJmfoK_QbjP_eVnnWNIbCF3qvMzOzkoWKzRInsMySsquWRhGLwQdIiSyzzOrfouEOv3gOTwPvwWUS-upQWiJYJ8A42yl5tdsw1jjAwbAhAY0h_KApfAoMDGPitbuR5gKRjkXPI_WpvcD170Zn3y79s_4qj0dtLLh5fi2NWm_ZZ-vWevaXfSyrw83enb9vht23OmT6z249647P5u8PGZ3o0U0l6hDa4r_W0B7G-u75S2vsrO5XV6teJbgQHqEUQLaRIKD1oGRmErGmC4loCKhBOEl0DwGBVIrHmipZUQSykwygwxiQgto-c97rNJEpdWwVm8cpbNiaHXcbEZhVR80Z7cZ5hgk_QXNtWfv
ClassificationCodes TP391
ContentType Journal Article
Copyright Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
Copyright_xml – notice: Copyright © Wanfang Data Co. Ltd. All Rights Reserved.
DBID 2B.
4A8
92I
93N
PSX
TCJ
DOI 10.12454/j.jsuese.202201373
DatabaseName Wanfang Data Journals - Hong Kong
WANFANG Data Centre
Wanfang Data Journals
万方数据期刊 - 香港版
China Online Journals (COJ)
China Online Journals (COJ)
DatabaseTitleList
DeliveryMethod fulltext_linktorsrc
DocumentTitle_FL Self-attention Deep Field-embedded Factorization Machine for Click-through Rate Prediction
EndPage 296
ExternalDocumentID scdxxb_gckx202405029
GrantInformation_xml – fundername: (国家自然科学基金); (国家自然科学基金); (江西省重点研发计划重点项目); (教育部人文社会科学研究项目); (江西省自然科学基金面上项目); (江西省社会科学研究规划基金项目); (江西省高校人文社科基金项目); (江西省教育厅科技项目); (江西省教育厅科技项目); (江西省研究生创新专项资金项目)
  funderid: (国家自然科学基金); (国家自然科学基金); (江西省重点研发计划重点项目); (教育部人文社会科学研究项目); (江西省自然科学基金面上项目); (江西省社会科学研究规划基金项目); (江西省高校人文社科基金项目); (江西省教育厅科技项目); (江西省教育厅科技项目); (江西省研究生创新专项资金项目)
GroupedDBID -0C
-SC
-S~
2B.
2RA
4A8
5VR
92I
92M
93N
9D9
9DC
AFUIB
ALMA_UNASSIGNED_HOLDINGS
CAJEC
CQIGP
GROUPED_DOAJ
PB1
PB9
PSX
Q--
R-C
RT3
T8S
TCJ
U1F
U5C
ID FETCH-LOGICAL-s1079-14312ceb652cc7e9039444c8f2ab6a61582c5d2a29ca57c9c9b1f34ef5d242d13
ISSN 2096-3246
IngestDate Thu May 29 03:53:57 EDT 2025
IsPeerReviewed true
IsScholarly true
Issue 5
Keywords field-embedded factorization machine
click-through rate prediction
multi-head self-attention
深度神经网络
feature interaction
点击率预测
多头自注意力
deep neural network
特征交互
域嵌入因子分解机
Language Chinese
LinkModel OpenURL
MergedId FETCHMERGED-LOGICAL-s1079-14312ceb652cc7e9039444c8f2ab6a61582c5d2a29ca57c9c9b1f34ef5d242d13
PageCount 10
ParticipantIDs wanfang_journals_scdxxb_gckx202405029
PublicationCentury 2000
PublicationDate 2024-09-01
PublicationDateYYYYMMDD 2024-09-01
PublicationDate_xml – month: 09
  year: 2024
  text: 2024-09-01
  day: 01
PublicationDecade 2020
PublicationTitle 工程科学与技术
PublicationTitle_FL Advanced Engineering Sciences
PublicationYear 2024
Publisher 华东交通大学信息与软件工程学院,江西南昌 330013
Publisher_xml – name: 华东交通大学信息与软件工程学院,江西南昌 330013
SSID ssib036435565
ssib050593459
ssib041261190
ssib030194745
ssib051371919
ssj0003313526
ssib027967859
Score 2.3949916
Snippet TP391; 点击率(CTR)预测通过预测用户对广告或商品的点击概率,实现数字广告精准推荐.针对现有CTR模型存在原始嵌入向量未精化、特征交互方式偏简单的问题,本文提出自注意力深...
SourceID wanfang
SourceType Aggregation Database
StartPage 287
Title 面向点击率预测的自注意力深度域嵌入因子分解机
URI https://d.wanfangdata.com.cn/periodical/scdxxb-gckx202405029
Volume 56
hasFullText 1
inHoldings 1
isFullTextHit
isPrint
journalDatabaseRights – providerCode: PRVAON
  databaseName: Directory of Open Access Journals
  issn: 2096-3246
  databaseCode: DOA
  dateStart: 20220101
  customDbUrl:
  isFulltext: true
  dateEnd: 99991231
  titleUrlDefault: https://www.doaj.org/
  omitProxy: true
  ssIdentifier: ssj0003313526
  providerName: Directory of Open Access Journals
link http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwnV3Na9VAEA9tvXgRRcVvenBPEk02u9nscfe9hOLBUwu9lWRfUqnwBPsKpUepVlS0Xq1SqfiB4EGkiPVf8ZrX9r9wZpO-RNpDFR5hmZ3MzG_nbXYm2Q_HuR6mqd9LaeqyQhiXCSHclPd6bpEXacqMxC2tcLbFnXBqht2e5bNj479bs5aWBtlNs3LkupL_8SrQwK-4SvYfPDsSCgQog3_hCh6G67F8TGJJZJcoSmJOpEekT2JBIkq0RArk-VpbSoxlYAbOiJE4JBpqbZVUlhIhg1K2KiAqwgLQo8TKUURqWwUCfaRoRVRolSb4QwqwdSwzJ4rbKk2UhwXVRduwCrSEqEuBrgAFyg6pXxQsHCDiqAUlCDSjMhL4pV-LQr2M6AhBoZEK52pUotRolrIldC0Lx8bQSX2T7jYsHOFpa1AkERIiBKhewwKmxnhbI0WiJUq3peiOBSpsq1CrGuzqtN-nUDaaMFb1AKu9MtDaBdZjAQQw1IHTT5QFzBD8X8gTonyLHOxIWm2B4A4o2uI63JQjOQDD_htox6L2rRcjZNZJbZsUFgrc2LkRBBjEN4MGhZTUhSA5bI9w1dbtdU_m7eGqDnaqyIdWZwsfGlQp48yOqgt4HiTuLUsp7lQZNDHEaGbnouktL2dz8-beMjauxz0qx50TFMZcr_XCA572VEgIlJpgGkYayVpnLwQQK3PefINnPiT7fvNNl-NhlK2dkDhY5Mt6c0KMw4LAx8Mf8HTJg0aptx1DQLcOw7Hr9vpF2p9vhZjTp51TdW44qaqOfsYZW7l71tH7b7fK9Ve7D3-Wa792X6ztb60Ot5_tvl7dW_sy_P55uPqyfLox_PGt3PlYbm6W28_LRx_KjXfl1_XyyeO9T--Hb3bOOTNJPN2ZcuuDT9xF3xPShRzGpybPQk6NEbn0cPU6M1FB0yxMIQeJqOE9mlJpUi6MNDLzi4DlBRAZ7fnBeWeif7-fX3AmZZAzaSARMFGIwXiW8UwYDnlXwEVB84sOqTHP1Q-2xbmjfHjpmHyXnZNNp7riTAweLOVXIWwfZNes9_8Ah6mykA
linkProvider Directory of Open Access Journals
openUrl ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=%E9%9D%A2%E5%90%91%E7%82%B9%E5%87%BB%E7%8E%87%E9%A2%84%E6%B5%8B%E7%9A%84%E8%87%AA%E6%B3%A8%E6%84%8F%E5%8A%9B%E6%B7%B1%E5%BA%A6%E5%9F%9F%E5%B5%8C%E5%85%A5%E5%9B%A0%E5%AD%90%E5%88%86%E8%A7%A3%E6%9C%BA&rft.jtitle=%E5%B7%A5%E7%A8%8B%E7%A7%91%E5%AD%A6%E4%B8%8E%E6%8A%80%E6%9C%AF&rft.au=%E6%9D%8E%E5%B9%BF%E4%B8%BD&rft.au=%E5%8F%B6%E8%89%BA%E6%BA%90&rft.au=%E8%AE%B8%E5%B9%BF%E9%91%AB&rft.au=%E5%BC%A0%E7%BA%A2%E6%96%8C&rft.date=2024-09-01&rft.pub=%E5%8D%8E%E4%B8%9C%E4%BA%A4%E9%80%9A%E5%A4%A7%E5%AD%A6%E4%BF%A1%E6%81%AF%E4%B8%8E%E8%BD%AF%E4%BB%B6%E5%B7%A5%E7%A8%8B%E5%AD%A6%E9%99%A2%2C%E6%B1%9F%E8%A5%BF%E5%8D%97%E6%98%8C+330013&rft.issn=2096-3246&rft.volume=56&rft.issue=5&rft.spage=287&rft.epage=296&rft_id=info:doi/10.12454%2Fj.jsuese.202201373&rft.externalDocID=scdxxb_gckx202405029
thumbnail_s http://utb.summon.serialssolutions.com/2.0.0/image/custom?url=http%3A%2F%2Fwww.wanfangdata.com.cn%2Fimages%2FPeriodicalImages%2Fscdxxb-gckx%2Fscdxxb-gckx.jpg