2018 - Bidding Machine: Learning to Bid for Directly Optimizing Profits in Display Advertising
2018 - Audience Size Forecasting
2017 - Attribution Modeling Increases Efficiency of Bidding in Display Advertising
2017 - Profit Maximization for Online Advertising Demand-Side Platforms
2015 - Smart Pacing for Effective Online Ad Campaign Optimization
Czech notes
Ad requesty jsou rozdeleny do vice skupin, dle predikovaneho vykonu (CTR, CR). Kazda skupina ma vypocten pacing rate. Algoritmus se snazi bidovat na skupiny s nejvyse predikovanou hodnotou pokud stiha odtacet budget. Adaptivne upravuje pacing rate dle toho jak odtaci. Kdyz nestiha odtacet, zvysuje pacing rate i pro skupiny s mensim vykonem, napr. s nizsim CTR prediction.2013 - Real Time Bid Optimization with Smooth Budget Delivery in Online Advertising
Czech notes
Maximalizace dosazeni daneho cile (CTR, CR, eCPC, eCPA) pri dodrzeni rozlozeni budgetu v ramci celeho behu kampane. Budget lze rozkladat na casove sloty dle cile, napr. utracet nejvice v hodinach kdy se nejvice nakupuje na shopu, ci klika. Metoda stavi na metrice pancing rate, tedy urceni idealniho poctu bidu ze vsech adrequestu, odpovidajicich cileni, abychom odtaceli spravne budget. Dale hledame hladinu predikovane CTR, CR kdy ma smysl jeste bidovat a s jakou cenou.2012 - Bid Optimizing and Inventory Scoring in Targeted Online Advertising
Czech notes
Modifikace zakladni ceny bidu dle predikce pravdepodobnosti konverze dle inventory. Ma cenu bidovat s dvakrat vetsi cenou pro uzivatel, kteri maji dvakrat vetsi pravdepodobnost konverze. Inventory, kontext zobrazeni stranky ma vliv na konverzi. Typ cteni (clanek o SQL nebo bulvar), viditelnost reklam, atd. meni pravdepodobnost konverze. Zkouseli vice bidovacich strategii, 1 - pomerne menit bid price dle predikovane CVR, 2 - agresivni pristup, zarezavat uzivatele s malou pravdepodobnosti konverze (<0.8), bidovat dvakrat vice pro dobe uzivatele CVR>1.2 a 3 - baseline, bidovat konstani cenu. Druha stategie mela nejvetsi CVR, ale take vetsi CPA (mensi marze pro DSP). Vhodne pro nove klienty, kteri porovnavaji vykon s jinymi systemy.
2011 - Real-Time Bidding Algorithms for Performance-Based Display Ad Allocation
2014 - Frequency Capping in Online Advertising
2015 - Evaluating and Optimizing Online Advertising Forget the click, but there are good proxies
2016 - Improving Advertisement Recommendation by Enriching User Browser Cookie Attributes
2011 - Scalable Distributed Inference of Dynamic User Interests for Behavioral Targeting
2009 - Audience Selection for On-line Brand Advertising: Privacy-friendly Social Network Targeting
2007 - Demographic Prediction Based on User’s Browsing Behavior
2018 - Adaptive look-alike targeting in social networks advertising
2016 - Audience Expansion for Online Social Network Advertising
2016 - A Sub-linear, Massive-scale Look-alike Audience Extension System
2016 - Score Look-Alike Audiences
2015 - Effective Audience Extension in Online Advertising
Social Networks: Finding Highly Similar Users and Their Inherent Patterns
2008 - Contextual Advertising by Combining Relevance with Click Feedback
2015 - From 0.5 Million to 2.5 Million: Efficiently Scaling up Real-Time Bidding
2013 - When Does Retargeting Work? Information Specificity in Online Advertising
2017 - Optimized Cost per Click in Taobao Display Advertising
2017 - Cost-sensitive Learning for Utility Optimization in Online Advertising Auctions
2017 - Modeling Delayed Feedback in Display Advertising
2017 - Advertisement Click-Through Rate Prediction Based on the Weighted-ELM and Adaboost Algorithm
2014 - Practical Lessons from Predicting Clicks on Ads at Facebook
2013 - Ad Click Prediction: a View from the Trenches (Google)
2012 - Multimedia Features for Click Prediction of New Ads in Display Advertising
Czech notes
Resi cold start CTR predictoru pro nove ady. Extrahuje vlastnosti z adu (img, flash) jako jsou barvy, svetlost, stupne sede, kontrast, barevnost, texturea, text pres OCR, objekty na obrazku, jejich mnozstvi a dle techto vlastnosti predikuji CTR.
2012 - The Impact of Visual Appearance on User Response in Online Display Advertising
2012 - Estimating Conversion Rate in Display Advertising from Past Performance Data
Czech notes
Uzivatele, publishery a advertisery rozdeluji do taxonomii. Predikuji CVR na zaklade historickych statistik. Jake maji CVR podobni uzivatele na podobnych webech. Pro predikci vyuzivaji logistickou regresi.
2010 - Estimating Rates of Rare Events with Multiple Hierarchies through Scalable Log-linear Models
2010 - Personalized Click Prediction in Sponsored Search
2010 - Predicting the Click-Through Rate for Rare/New Ads
2007 - Predicting Clicks: Estimating the Click-Through Rate for New Ads
Czech notes
Predikce CTR dle klicovych slov (termu) v sponsored search advertisingu. Snazi se najit kampane cilene na stejne termy, pripadne na podmnozinu, ci nadmnozinu.
2016 - Webpage Depth-level Dwell Time Prediction
Czech notes
Pomoci Factorization Machine modelu predikuji jak hluboko na strance uzivatel zascroluje a jakou dobu tam vydrzi. Model dobre pracuje s ridkymi daty a umoznuje zpojit kontextualni data. Zkousi ruzne kontextualni data (viewport, geo, delku dokumentu dle poctu slov, ..) a jako velmy silnou feature se ukazal hlavne viewport. Model porovnavaji s regreasi a FM vychazi lepe. Pro vyhodnoceni pouzivaji Root-Mena-Square Deviation a agresivnejsi Logistic Loss.
2015 - Viewability Prediction for Online Display Ads
Czech notes
Pomoci Probabilistic Latent Class Modelu predikuji jak hluboko n stance uzivatel zascruluje (bez casu). Porovnavaji s SVD, Logistic Regresi. Testuji ruzne featury (device type, geo, day of week, hour of day, view rate per user). Vyhodnocuji pomoci Root-Mean-Square-Deviance a F1 score.