<th id="hbr4m"><td id="hbr4m"></td></th>
    1. <code id="hbr4m"><xmp id="hbr4m">

        <optgroup id="hbr4m"><xmp id="hbr4m"><code id="hbr4m"></code>
        <listing id="hbr4m"><thead id="hbr4m"><bdo id="hbr4m"></bdo></thead></listing>

        <code id="hbr4m"><menu id="hbr4m"></menu></code>

        學術報告:Data Monetization with Machine Learning




        報告人:談飛 博士



            談飛博士,于20195月獲得美國新澤西理工計算機博士學位,博士論文獲得計算機系Joseph Leung獎?,F于美國Verizon Media集團下的雅虎研究院(紐約曼哈頓)任職研究科學家,主要研究興趣是數據挖掘,機器學習應用和自然語言處理。他在相關的主流會議和期刊上發表十余篇學術論文,包括IEEE TNNLS (影響因子11.68), IEEE ICDM, SIAM SDM, IJCAI, Data Mining and Knowledge Discovery, Physical Review E, Europhysics Letters等。他也是一項在申請的美國專利共同持有人。他曾在Adobe(硅谷圣何塞總部)和雅虎研究院實習工作多次以及作為研究助理訪問香港理工大學。



            Machine learning has being harnessed to refine big data and render it value like never before. In this talk, we will explain three data monetization cases through machine learning. Specifically, in online lending, how to represent two competing risks, charge-off and prepayment, in funded loans is a fundamental problem behind ROI maximization. We develop a hierarchical grading framework to integrate them both qualitatively and quantitatively. In addition, in digital marketing, we propose to treat content understanding as to elucidate their causal implications in driving user responses. A flexible and adaptive doubly robust estimator is introduced to identify the causality between related features and user responses from observational data. In online communities, abusive language has profound impacts on their integrity. We explore two byte-level quantization schemes for character representation. The primitive representation empowers models to capture signals underlying multi-byte characters of online posts elegantly.