Introduction (5-10 minutes)
Media, Entertainment and Technology
The Data Points Virtual Summit features 12 interactive online sessions on Data Science, Analytics & Enterprise AI trends and best practices with a focus on the Media, Entertainment & Technology sectors. Join other data science leaders as they engage and learn from each other.
- 11:00 - 11:10 ET10 min intro
- 11:10 - 11:35 ET25 min session
A Plug-and-play Personalization Platform Using Reinforcement Learning - Aleksey Aleev & Vasilii Kovalchenko (Grid Dynamics)
Abstract: Personalization and recommendation models are widely adopted by technology and entertainment companies as highly efficient tools for digital experience improvement and conversion optimization. The traditional model development and productization process, however, requires significant engineering effort that increases the time to market and implementation costs.
In this session, targeted for product managers and software engineers, we'll discuss the design of a platform for in-app and in-game experience personalization that reduces the implementation complexity and improves the quality of recommendations using reinforcement learning.
- 11:35 - 11:40 ET5 min break
- 11:40 - 12:05 ET25 min session
Recipe Recommendations at The New York Times - Celia Eddy, Data Scientist (The New York Times)
Abstract: At the New York Times, we publish over 250 pieces of journalism every day and have a collection of over 20,000 recipes. In order to help readers find the content that is most engaging and relevant to them, the Algorithmic Recommendations team develops machine learning models to make content recommendations across the website, apps, and email newsletters. Our algorithms combine information about articles and recipes, reader behavior, and editorial judgment in order to recommend content that deepens user engagement. In this talk, I will focus on how we make recipe recommendations for NYT Cooking, and describe both the algorithms that our team has developed and the infrastructure that allows us to make real-time recommendations. Algorithm-driven recommendations have led to increased engagement and personalization on both the web and in apps and help our users find and cook the recipes that are most interesting to them.
- 12:05 - 12:10 ET5 min break
- 12:10 - 12:35 ET25 min session
Balancing the fraud equation: How DoorDash optimizes fraud model thresholds and actions - David Press, Data Scientist (DoorDash)
Abstract: Like all online businesses, DoorDash battles fraudsters who attempt to place orders with stolen credit cards. In this talk we’ll walk through how we leverage machine learning and experimentation to identify and block fraudsters while minimizing impact on the overwhelming majority of normal consumers.
First, we’ll introduce how we use machine-learning models to trigger additional verifications or ‘frictions’ targeted at blocking fraudulent activity. Next, we’ll outline how we choose the model’s threshold by minimizing a loss function, and how experimentation helps us estimate each term in the loss function: the costs of false positives, false negatives, and true positives. Finally, we’ll walk through a concrete example comparing the optimization of blocking orders versus applying a friction.
- 12:35 - 12:40 ET5 min break
- 12:40 - 1:05 ET25 min session
Ratings Transformation: Integrating Big Data - Danny Monistere, SVP of Policy and Guidelines (Nielsen)
Abstract: By combining advanced meter technology, big data and people-powered panels into its Local TV measurement service, Nielsen continues to provide in-depth ratings that are representative of what people are actually viewing. Unfortunately, big data has shortcomings and limitations that, if left uncorrected, could result in inconsistent or inaccurate data. By utilizing Nielsen’s gold standard technology and panels, and by adjusting for the inadequacies of big data, Nielsen provides unbiased and reliable measurement with greater stability and less variation in an increasingly fragmented media marketplace.
- 1:05 - 1:20 ET15 min break
- 1:20 - 1:45 ET25 min session
There are no shortcuts to any place worth going: The journey towards Enterprise AI - Carlos Ariza, Chief Data Scientist (Creative Artists Agency)
Abstract: Some may find it surprising that a talent agency has built an in-house data and analytics capability.
However, the landscape of entertainment has changed radically over the past five years, with data-savvy
companies like Netflix, Amazon and Spotify dominating distribution. This makes it critical to level the
playing field by building capabilities in data unification, validation and visualization, as well as Machine
Learning and AI.
In this talk, we’ll discuss our transformation from disconnected centers of analytics excellence to a
centralized data team, sharing some of the organizational challenges faced early on, and how building a
strong foundation of data quality allows us now to be very agile when applying predictive analytics to
create value for our clients.
- 1:45 - 1:50 ET5 min break
- 1:50 - 2:15 ET25 min session
A multi-faceted approach to optimizing digital advertising spend towards desired outcomes using attention measurement and machine learning - Sergei Izrailev, Head of Data Science and Data Analytics (Yieldmo)
Abstract: In this talk, we first introduce attention metrics derived from users' micro-interactions with the ads and show how these metrics relate to typical online advertiser KPIs. We will then describe how a combination of different tactics, including user-based targeting on addressable media and impression-based real-time optimization of non-addressable media, are applied to direct the campaign spend and to achieve the advertiser goals. Finally, we will describe the technology behind the optimization and how we measure the incremental improvement of performance with A/B testing.
- 2:15 - 2:20 ET5 min break
- 2:20 - 2:45 ET25 min session
Sidney Madison Prescott, Global Head of Intelligent Automation (Spotify)
- 2:45 - 2:50 ET5 min break
- 2:50 - 3:15 ET25 min session
Solving Customer Churn: A Prescriptive Modelling Approach - Jake Jian, Data Science Lead (Grid Dynamics)
Abstract: Customer retention is the key pillar for growing a subscription-based business model. In fact, many advanced machine learning techniques have been attempted in order to provide more accurate churn prediction. However, a good churn model requires not only prediction accuracy. It also requires the explanatory power that can help to identify and improve upon areas where customer service is lacking. In this presentation, we will show a prescriptive modelling approach to solve customer churn problems.
- 3:15 - 3:30 ET15 min break
- 3:30 - 3:55 ET25 min session
Consumer Segmentation, Targeting, and Positioning using Unsupervised Clustering and Distance Algorithms in Advertising - Das Dasgupta, Chief Data Officer (Saatchi & Saatchi)
Abstract: Marketers balance brand awareness campaigns with lower funnel targeted campaigns to move KPI's on Consideration and Conversion. In a targeted campaign, it is critical to understand the 'needs and wants' of consumers based on their demographics, and map these to the product's perception and preferences across segments of consumers, so that appropriate positioning can be done to address these. We show a three-step analytical framework that uses respondent level survey data to generate distinctive segments using hierarchical or K-Means algorithms, use the McKinsey-GE matrix to target the right consumers based on profitability and growth, and finally use a matrix of perceptions on the product and preference sets of consumers in the chosen targets to develop strategic messaging and positioning the product so it is relevant to their needs and wants.
- 3:55 - 4:00 ET5 min break
- 4:00 - 4:25 ET25 min session
iOS14 or: How I Learned to Stop Worrying and Love the SKAdNetwork - Jacob Claussen, Manager, Marketing Analytics (Zynga)
- 4:25 - 4:30 ET5 min break
- 4:30 - 4:55 ET25 min session
A nonlinear embedder framework with interpretable features for content personalization - Taeyoung Choi, Data Scientist (Condé Nast)
Abstract: Condé Nast publishes thousands of new articles each week globally. Once published, editorial teams will often revise these articles with up-to-date information and connections to new cover stories. From a recommendation system's perspective, this imposes unique constraints because the item inventory is continually changing, and the items have relatively short life spans. By the time relevant data has been collected, and the model trained, the learned latent vectors can quickly become stale. A successful recommendation system for this type of media business must represent users and their respective content interests as soon as data around them is made available. To address these challenges effectively, we've grown from classical collaborative filtering approaches and developed a nonlinear variant of word2vec. In this talk, we'll cover how the recommendations system at Condé Nast learns various entity latent vectors that are both easily interpretable and can capture multiple user interests through its nonlinearity component.
- 4:55 - 5:00 ET5 min break
- 5:00 - 5:25 ET25 min session
Content-Driven Advertising using First Party Data - Amit Bhattacharyya, Head of Data Science (Vox Media)
Abstract: For quite some time, the advertising industry has assumed that targeting based on online demographic data equates to better ad performance. However, with recent changes in data privacy laws and implementation, the entire industry will have to change the way advertisers can and should be connecting to users. It behooves all of us to find sustainable solutions that free us from relying on third-party data.
Making context the centerpiece of ad targeting gives an opportunity to reduce the reliance on demographic databases and put more emphasis on relevant intent signals from when the user is consuming content.
We describe a clustering model that connects users to the content they are consuming to identify context-based audience segments. In turn, these segments are used to train a classification model that is used in real time to inform advertising without relying on 3rd-party cookies.
- 5:25 - 5:30 ET5 min outro
- Customer analytics and personalization
- B2C and B2B price and promotion optimization
- Marketing spend optimization and attribution
- Risk scoring and fraud detection
- AI/ML for IT Operations
- Data Science platforms and MLOps
- Deep learning and reinforcement learning for enterprise applications
- Computer vision for digital experience platforms and manufacturing
- Data platforms for self-service analytics
- and more...