Tag Cloud


你是第 位访问者

Zhiling Luo

Zhiling Luo (Bruce)

Algorithm Expert
DAMO Academy (达摩院) / XLab (ALIME)

Alibaba Group

IEEE Member, ACM Professional Member



Research Interests: NLP, Graph Computing, General Machine Learning.

Research Gate Linkedin dblp baidu

News: our paper "Adapted Graph Reasoning and Filtration for Description-Image Retrieval" is accepted and will appear on SIGIR 2021.


I'm an algorithm expert in Alibaba Group, DAMO Academy.

Before joining Alibaba group, I was an assistant professor in Computer Science at Zhejiang University.

My research topic is about the advanced automatic dialogue system in online Customer Service, called ALIME.

I serve as the Programme Committee of IJCAI(2018-2020), AAAI(2019-2021), ACL(2020), EMNLP(2020) and IEEE ICWS(2019-2020). Besides, I'm the regular reviewer of IEEE TKDE, TSC, CIM and Neurocomputing.

I offered a course "Artificial Intelligence" (code 2122026004) for graduated students in the College of Computer Science at Zhejiang University.

I received my Ph.D. degree in Computer Science in 2017 and my B.S. in Computer Science from Zhejiang University in 2012.

I visited Georgia Institute of Technology (GT) in a visiting scholar project (Nov 2015- Nov 2016) under the guidance of Prof. Ling Liu.


我的研究主要集中在图论和多模态相关技术。我们研发的对话机器人(阿里小蜜),目前是使用量最大的中文客服机器人之一。 特别的,我团队负责多模态内容中台的建设,用统一的架构,支持优质的内容生产、管理和发放。我们自动生产的亿级别内容达到了亿级别每日曝光。在手机淘宝,天猫,饿了么,盒马,天猫精灵,都有我们的算法应用。自研的模态抽取算法和检索算法均达到SOTA水平。文本生成算法,视觉检测算法方面达到阿里巴巴集团领先水平;除了算法效果之外还构建有完整的高效内容生产工程化链路,支持千万级商品的T+1更新,累计产生的内容超过1亿。我们的算法被用在阿里巴巴十多个部门和业务,服务百万商家和数千万消费者。

目前我是IJCAI,AAAI,ACL,EMNLP等会议的PC,IEEE TKDE, TSC, CIM, Neurocomputing的常规审稿人。


此外也提供两种实习岗位: 1. 暑期实习生:面向次年毕业的硕士和博士生,主要集中在6-8月; 2. 研究型实习生:面向博士生和优秀的硕士生。具体邮件详聊


Shiqian Chen, Zhiling Luo, et. al. (2021) Adapted Graph Reasoning and Filtration for Description-Image Retrieval [C] ACM SIGIR 2021.

[Best Paper Award] Meng XI, Zhiling Luo, et. al. (2020) A Latent Feelings-aware RNN Model for User Churn Prediction with only behavior data [C] IEEE SMDS 2020.

Yunzhou Shi, Zhiling Luo, et. al. (2020) G2T: Generating Fluent Descriptions for Knowledge Graph [C] ACM SIGIR 2020.

Yu Chen, Zhiling Luo, Sha Zhao, Ying Li and Jianwei Yin (2020) Adversarial Attacks on Graphs by Adding Fake Nodes [C] AAAI 2020 (Workshop DLGMA). Full paper

Bo Lin, Wei Luo, Zhiling Luo, Bo Wang, Shuiguang Deng, Jianwei Yin, Mengchu Zhou (2020) Bradykinesia Recognition in Parkinson's Disease via Single RGB Video. [J] ACM Trans. Knowl. Discov. Data (TKDD) 14(2): 16:1-16:19

Zhiling Luo, Yinghua Cui, Sha Zhao, Jianwei Yin. (2020) g-Inspector: Recurrent Attention Model on Graph. [J] IEEE Transactions on Knowledge and Data Engineering (TKDE). 2020. Code, Full paper

Sha Zhao, Feng Xu, Yizhi Xu, Xiaojuan Ma, Zhiling Luo, Shijian Li, Anind Dey, and Gang Pan. (2019) Investigating Smartphone User Differences in Their Application Usage Behaviors: An Empirical Study. [J] CCF Transactions on Pervasive Computing and Interaction, 2019

Sha Zhao, Yizhi Xu, Xiaojuan Ma, Ziwen Jiang, Zhiling Luo, Shijian Li, Laurence T. Yang, Anind Dey, and Gang Pan. (2019) Gender Profiling from a Single Snapshot of Apps Installed on a Smartphone: An Empirical Study. [J] IEEE Transactions on Industrial Informatics, 2019 (IF = 7.377) (JCR Section I).

Sha Zhao, Shijian Li, Julian Ramos, Zhiling Luo, Ziwen Jiang, Anind Dey, and Gang Pan (2019) User Profiling from Their Use of Smartphone Applications: A Survey. [J] Pervasive and Mobile Computing, 2019 .

Sha Zhao, Zhiling Luo, Ziwen Jiang, Haiyan Wang, Feng Xu, Shijian Li, Jianwei Yin, and Gang Pan. (2018) AppUsage2Vec: Modeling Smartphong App Usage for Prediction. [C] The 35th IEEE International Conference on Data Engineering (ICDE2019) .

Zhiling Luo, Ling Liu, Jianwei Yin, Ying Li, Zhaohui Wu (2018). Latent Ability Model: A Generative Probabilistic Learning Framework for Workforce Analytics[J] IEEE Transactions on Knowledge and Data Engineering(TKDE) ( Full paper (pre-print version)).

Sha Zhao, Feng Xu, Zhiling Luo, Shijian Li, and Gang Pan (2018). Demographic Attributes Prediction Through App Usage Behaviors on Smartphones [C] ACM Workshop on The 2018 International Joint Conference on Pervasive and Ubiquitous Computing(AppLens 2018).

Ying Li, Meng Xi, Yuyu Yin, Zhiling Luo, Honghao Gao, Jianwei Yin (2018). MeCo-TSM: Multi-Entity Complex Process-Oriented Service Modeling Method. [C] 2018 IEEE International Conference on Web Services (ICWS), 82 - 90.

Jianwei Yin, Bangpeng Zheng, Shuiguang Deng, Yingying Wen, Meng Xi, Zhiling Luo, Ying Li (2018). Crossover Service: Deep Convergence for Pattern, Ecosystem, Environment, Quality and Value. [C] 2018 IEEE 38th International Conference on Distributed Computing Systems (ICDCS), 1250 - 1257.

Junxiang Wang, Jianwei Yin, Shuiguang Deng, Ying Li, Calton Pu, Yan Tang, Zhiling Luo (2018). Evaluating User Satisfaction with Typography Designs via Mining Touch Interaction Data in Mobile Reading. [C] ACM SIGCHI 2018 Full paper

Shenglin Zhang, Ying Liu, Weibin Meng, Zhiling Luo, Jiahao Bu, Sen Yang, Peixian Liang, Dan Pei, Jun (Jim) Xu, Yuzhi Zhang, Yu Chen, Hui Dong, Xianping Qu, Lei Song (2018). PreFix: Switch Failure Prediction in Datacenter Networks. [C] ACM SIGMETRICS 2018 Full paper

Zhiling Luo, Ling Liu, Jianwei Yin, Ying Li, Zhaohui Wu (2017). Deep Learning of Graphs with Ngram Convolutional Neural Networks[J] IEEE Transactions on Knowledge and Data Engineering(TKDE) 29, 2125 - 2139. Code, Full paper

Sha Zhao, Yifan Zhao, Zhiling Luo, Runhe Huang, Shijian Li, Gang Pan(2017). Characterizing a User from Large-scale Smartphone-sensed Data. [C] UbiComp/ISWC'17. Full paper

Zhiling Luo, Ying Li, Jianwei Yin, Honghao Gao, Yuyu Yin (2017). Service Pattern Evaluation: Studying Profitability from Perspective of Resource. [C] IEEE International Conference on Cognitive Computing. Full paper

Zhiling Luo, Ying Li, Ruisheng Fu, Jianwei Yin (2016). Don't Fire Me, a Kernel Autoregressive Hybrid Model For Optimal Layoff Plan. [C] 2016 IEEE Big Data. Full paper

Ying Li, Zhiling Luo, Binbin Fan, Jianwei Yin (2016). Who is the Most Diligent Employee? A Discriminate Model for e-Government Service Time Analyzing. [J] Journal of Industrial Integration and Management 2016 01:01 Full paper

Jianwei Yin, Zhiling Luo, Ying Li, Zhaohui Wu (2016). Service Pattern: An Integrated Business Process Model For Modern Service Industry. [J] IEEE Transactions on Services Computing. Code, Tech report, Full paper

Zhiling Luo, Ying Li, Jianwei Yin (2015). A Framework for Transmission Cost Aware Service Selection.[C] 2015 IEEE 22nd International Conference on Web Services (ICWS) 503-510. Code, Data, Full paper

Ying Li, Zhiling Luo, Jianwei Yin, Lida Xu, Yuyu Yin, Zhaohui Wu (2015). Enterprise Pattern: Integrating the business process into a unified enterprise model of Modern Service Company. [J]. Enterprise Information Systems, 2015: 1-21. Full paper

Jianwei Yin, Zhiling Luo, Ying Li, Binbin Fan, Weiwei Liu, Zhaohui Wu (2014). Towards a Service Pattern Model Supporting Quantitative Economic Analysis. [C] 2014 IEEE World Congress on Services (SERVICES) 95-102. Full paper

Zhiling Luo, Ying Li, Jianwei Yin (2013). Location: a feature for service selection in the era of big data. [C] 2013 IEEE 20th International Conference on Web Services (ICWS) 515 - 522. Full paper

Ying Li, Zhiling Luo, Jianwei Yin. (2013). A location-aware service selection model. [J] International Journal of Services Computing. IEEE International Journal of Services Computing (IJSC) 1, 52-63 Full paper


[Active] Human Behavior and Ability Analytics (2016- )

This research consists two parts: the understanding of human behavior and the estimation on ability. The former is on its way and supported by Netease Inc (Hangzhou). The latter work is published on TKDE 2018.

[TKDE2018] We introduce a novel dual generative model, called LAM, for workforce analytics. The LAM development is novel in three aspects. First, we introduce the concept of latent ability variables to model hidden relations between employees and activities in terms of job performance, such as the set of skills provided by an employee and the set of skills required by an activity, and how well they matchup in employee-activity assignment. Second, we construct the latent ability model (LAM) by learning latent ability parameters from the employee-activity log data using expectation-maximization and gradient descent. Finally, we leverage LAM to build inference and prediction models for employee performance prediction, employee ability comparison, and employee-activity matchup quality estimation

[JIIM2016] We build a discriminate model to extract the subjective factor, e.g. diligence, and objective factor from the working behaviors in logs. This model can be trained by Maximum Likeness Estimation to get the parameters which denote the factors.

[BigData2016] We address the problem of making an optimal layoff plan with the least influence on the executing of the business process. The key challenge is estimating the process throughput under a layoff plan. We overcome this challenge by two steps: regressing the activity throughput by the stuff number and inferring process throughput by the maximum flow or minimum cut algorithm on the Directed Acyclic Graph of process. In the regressing step, a kernel autoregressive hybrid model is proposed, whose MSE is 30% lower than SVM. After that, an augmenting path based algorithm is introduced to make an optimal layoff plan.

[Active] Deep Learning @ Graph Mining (2017- )

This research is to leverage the popular deep learning techniques in graph mining. NgramCNN involves Convolution Nerual Network in grpah classification task, which is published at TKDE2017.

[TKDE2017] NgramCNN is a deep convolutional neural network developed for classification of graphs based on common substructure patterns and their latent relationships in the collection of graphs. Our NgramCNN deep learning framework consists of three novel components: (1) The concept of $n$-gram graph block to transform each raw graph object into a sequence of $n$-gram blocks connected through overlapping regions. (2) The diagonal convolution layer to extract local patterns and connectivity features hidden in the $n$-gram blocks by performing $n$-gram normalization before conducting deep learning through the network of convolution layers. (3) The extraction of deeper global patterns based on the local patterns and the ways that they respond to overlapping regions by building a $n$-gram deep convolutional neural network.

TCSS-Transmission Cost aware Service Selection (2013-2016 )

This research is to study the technologies in selecting optimal services to minimize the transmission cost of the composite service. My paper, Location: a feature for service selection in the era of big data, accepted by ICWS 2013, firstly introduced the service's location and used it to represent the transfer rate between services.

The original TCSS can be approximately represented as an integer optimization problem with an acceptable approximate ratio, based on service's location. The approximate problem can be solved efficiently and effectively by selecting the shortest path in the graph of candidate services.

In my recent researches, I introduce a more practical model using the service logs to predict the network bandwidth in TCSS. I prove that the approximate ratio of TCSS is a constant closed to 1, with a reasonable assumption about the distribution of the transfer rate. I build a complete service environment in lab with 500 services by Python and collect their logs in a period of time. Then I measure the practical transmission time of services selected by different approaches. The experiment results demonstrate that services selected by my approach cost the least transmission time. I summarize this work and wrote a paper accepted by ICWS 2015 , NY USA.

SPDL -Service Pattern Description Language (2013-2015)

Services are organized as the process in the information systems built by SOA. My research focus on studying the pattern of the process in the perspective of resource. I introduce a formalized language, Service Pattern Description Language(SPDL). SPDL emphasizes the life cycle of the resource in process, compared with other process modeling languages (BPMN, BPEL and etc.).

This research are accepted by IEEE SERVICE 2014. I attended this conference and made a presentation about SPDL in Aug. 2014, Anchorage, Alaska, US. Photo

In recent works, I introduce the pattern of the process in SPDL and propose a process pattern matching approach based on Type Theory. Besides, we build a software, SPDLEditor, to support process designing by SPDL. The source code can be download here.

Business Model Analyzing (2012-2012)

We analyzed the data from the enterprise business model of Modern Service Industry. We collected the business model data of 62 enterprises from 355 public MSC in the Growth Enterprise Market (GEM), the Secondboard Market which is very similar to the NASDAQ Stock Market, in China. I combined the clustering approach of K-means with DaviesBoundin index in data processing.

We wrote a book, named Business Model Innovation of Modern Service Company, A Value Network Perspective, published in 2013. In this book, I wrote two chapters (all 5 chapters) to describe my approach in detail.

UTS -Universal Transferring Standard (2011-2011)

I attended the Student Research Training Program (SRTP) and led a team consisting of ten people in a project called UTS in 2011. UTS, short for Universal Transferring Standard, is a service bus providing a united interface of different news sources (e.g. Weibo and media publishers) for users. We built a service program using SSH of Java EE and a client for users by C#.

This project won the first prize in the Innovation Entrepreneurship Competition sponsored by Qware Tech. Co. Ltd. Photo

This project is supported by the Innovation Projects of Zhejiang Province 2011.


23875 services event logs

The dataset contains 23875 event logs of 500 service which are deployed on 10 servers (50 services per server).
Each line in the log files is an event. For example:

2014-10-27 14:24:12,760 - INFO: [RecvDataS]
2014-10-27 14:24:26,198 - INFO: [RecvDataE] 1697KB
2014-10-27 14:24:26,199 - INFO: [TransactionS] trans92
2014-10-27 14:24:28,201 - INFO: [TransactionE] trans92
2014-10-27 14:24:28,201 - INFO: [NextIP]
2014-10-27 14:24:28,201 - INFO: [SendDataS]
2014-10-27 14:24:30,301 - INFO: [SendDataE] 1686KB

In this segement, the service recived the data 1697KB as the input and started a transation, called trans92 and sent 1686KB data to
Sample code is here (Python and Matlab).
If you use this dataset, please use the following reference in citing the dataset:

  • Zhiling Luo, Ying Li, Jianwei Yin, "A Framework for Transmission Cost Aware Service Selection" in 22nd International Conference on Web Services (ICWS 2015).
  • Zhiling Luo, Ying Li, Jianwei Yin, "Location: a feature for service selection in the era of big data" in IEEE 20th International Conference on Web Services (ICWS 2013).


RBF-AR(p) regression

RBF-AR(p) is a hybrid kernel auto regressive model. As a regression tool, it can handle the auto-regressive factor y and the non-auto-regressive factor x at the same time.

The whole package consists of several files each is a function.

The core functionality is supported by following functions:

  • gartrain: The basic trainer
  • gartune: The parameter tuner
  • garpredict: The predicter


  Contact Me

Xixi Campus
Alibaba Group
Hangzhou, China


Last change: April 1st, 2021