Você está na página 1de 4

Seattle Machine Learning and Ranking PhD Intern Teams @ Facebook

Facebook Engineering is a dynamic fast-paced organization, and it consists of more than 40 smaller
engineering teams (which means there are far too many teams to list in this document). Each team is small,
which allows each of our engineers to move fast and have a huge impact at the same time. To help facilitate
the team selection process, we’ve grouped the engineering teams together into broader families. Please
read through these descriptions and let your recruiter know which of these groups sound interesting to you.

Ads Measurement Science - (Seattle)


Measurement Science builds scientiific methodologies and machine learning models to estimate the true
value of the ads - how many people has seen the ads, with what frequency, how many sales can be
attributed to ads, what is the incremental value generated by ads, and so on. Our models work on large set
of data and methodologies require strong statistical foundation.

Ads Ranking - (Seattle)


Our mission is to maximize the value of advertising delivered to advertisers and Facebook users, through
building scalable and highly automated machine learning infrastructure, innovating cutting-edge machine
learning algorithms, and formulating machine learning optimization problems to enable next-generation
ads products and achieve business goals. The expertise of our engineering team covers disciplinary research
and engineering areas including distributed systems, large scale distributed learning, embedding algorithms
and neural nets, large scale recommender systems and multi-variable optimization.

Business Integrity Optimization - (Seattle)


The Business Integrity ML team helps to keep only high quality content created by advertisers, page
owners, merchants, etc. reaching our users in order to ensure a satisfactory and sustainable experience.
Our team creates new and improves hundreds of existing production Machine Learning (ML) models to
ensure scalability of our content review process and minimize the amount of negative experiences within
Ads, Pages, Marketplace, Groups, Messenger, etc. on Facebook, by automatically detecting policy violating
and low quality content. Business Integrity Optimization offers the unique challenge of dealing with a large
variety of ML problems (classification, regression, ranking, ...) applied to different type of components (e.g.,
text, video, image, URL, ...) and the unique opportunity of combining it with human computation (hundreds
of reviewers) within our ML workflows.

CareML - (Seattle)
The CareML team is responsible for the detection, prevention, and remediation of objectionable content on
Facebook (Nudity, Pornography, Hate Speech, Violence, and Gore). We work to keep Facebook a clean and
welcoming platform for all. We build state of the art algorithms for images, text, and video as well as build
infrastructure that is truly "Facebook scale". Examples of projects (1) Automatically reporting suspicious
group posts to group admins for approval. This helps keeps groups clean and on topic. (2) Improving video
classification of suggestive videos through audio analysis.

Distribution Science - (Seattle)


Distribution science builds first class ML technologies and products to drive meaningful interactions
between people and businesses. This is critical to helping SMBs succeed on FB and also optimize people's
time spent on FB.
Feed Experience - (Seattle)
The mission of the News Feed Experience team is to help people tell the stories of their lives, and have
conversations around them. The NFX ML and Backend team in Seattle builds large scale machine learning
systems for News Feed Experience, with start-of-the-art deep learning techniques, including ranking
content in different surfaces, deep learning based recommendations systems, and more. This is your
opportunity to work on a Facebook core product that is used by more than 1B users every day.

Feed Integrity - (Seattle)


Our mission is to connect people to the highest quality content in News Feed -- while addressing problems
of spam, abuse, deception and incivility. "Feed Integrity" is the parent group that many of our sub-teams
roll up to. We plan on growing aggressively throughout 2018, with the department more than doubling in
size.

Misinformation : Fake news and misinformation is one of the most serious problems facing our community
and elections. Our team build systems and develops machine learning models to fight the spread of fake
news. We aim to reduce the prevalence of fake and misleading information in US and other countries by
discovering the actors that promulgate it, the specific content which spreads it, and the engagement
patterns which typify its spread through our community.

Controls: helping our users make choices so that they can better control the kinds of content they see in
their feed -- for example allowing them to express how much we should crack down on Clickbait, or how
much to trust Fact Checkers such as Snopes and Politifact.

Web Experience and Quality: The mission of the team is to ensure that our 2 billion+ users have a world
class experience when they click on any link displayed in their news feed. We do this by modeling and
understanding the web, acquiring Integrity landing page signals; and developing metrics to better
understand user experience “after the click”. The work involves building classifiers that predict the
probability that a particular landing page represents a given kind of bad experience (hate/destructive
conflict, adult, violence, spam, etc.), and then using the resulting scores as Integrity inputs into Feed
Ranking. We would like to understand how bad content *inside Facebook is monetized outside *Facebook.
The team will build and evolve web graph models, and apply classifiers to the content and images on web
pages.

Affective Polarization: “Affective polarization” refers to people's negative feelings toward others who are
different from themselves (e.g., political affiliation). Research indicates that affective polarization has been
increasing world-wide over the past few years. While there are many contributors to polarization, we want
to understand its relationship to social media use and we have a shared responsibility to reduce its negative
effects. We also want to study the separate but related concepts of filter bubbles (only seeing/showing
items that somebody already agrees with) and echo chambers (forming homogenous communities). The
goal of our team is to build a suite of metrics that can measure the extent of polarization of our users, and
to take steps to reduce it. We are experimenting with several different measures to do so - e.g. boost
content that leads to positive cross-cutting interactions, ranking improvements to help present a balanced
view of a hot-button issue, UI displays to highlight commonality, down-ranking toxic comments, and more.
We will also be researching and testing several open questions: how does polarization affect the user
experience; how do products like Groups, Pages You May Like, and Friend recommendations influence
polarization; can Facebook change polarization with future products and features; and how reactive any
polarization metric will be to our changes.
Pages Integrity - (Seattle)
Pages Integrity is Facebook's main abuse and threat prevention team centered on counteracting external
risks to Facebook from the use of Facebook Pages. Our team is building ML/AI technologies, infra solutions
and product use cases to safeguard our platform from bad page actors. The internship project will focus on
the following 2 pillars:
Enforcement: Not everyone has the best intentions when using Facebook Pages. Our job is to identify bad
actors and enforce against them as quickly as possible. We do this by enhancing internal tools that our Page
Operations partners use, generating signals that alert us to possible risks to Facebook, and collaborating
with stakeholders to take the most appropriate action
Machine Learning: Different types of badness exists within the humongous Facebook Pages Graph e.g.
spam, scams, offensive/pornographic content, impersonation etc. Identifying bad pages actors involves
building a diverse set of signals derived from the Page including content (text, photo, videos etc.), links,
connections to other pages and users and page-admin behavior. The ML pillar focuses on building ML and
ranking systems to predict different kinds of badness in Pages. The pillar also supports signals and features
that feed into the rest of the pillars described above.
We have major plans to grow the team in MPK and Seattle. We're are looking for ML/backend generalists
and specialists, and WWW and mobile engineers with a heavy focus on product work.

Search Ranking - (Seattle)


People use Facebook search over 2B times a day with a broad set of intents around searching and learning
about other people and the world around. They also come to Facebook to search over trillion pieces of
shared content (e.g. posts, photos, videos, links) to find what people are talking about any topic right now,
a personalized feed of content to satisfy a particular intent, or just a specific post they had seen before. The
goal of the search ranking team is to assist the users to complete their intents, and to provide the most
relevant and personalized set of results for these intents and unlock this wealth of information that exists
on Facebook. We develop ranking algorithms that are a three-way optimization between the query,
searcher, and the content. We are investing and growing in almost every area that generates signals and
contributes to this optimization – query and intent understanding, image and video search, news
understanding, needle search, learning from click data, building better relevance models, text search over
posts. The ability to find unconnected people and other entities in the Facebook graph is key to the
Facebook product (research background in embedding, deep learning, NLP, recommendation systems,
social graph modeling and related areas highly desirable). We use state of the art machine learning
algorithms to combine hundreds of these features into the production ranking models to generate the
relevant search results at scale. Join us to improve the quality of our search results, add search intents that
we don't support today, and to grow usage.

Signals + Identity Prediction - (Seattle)


Our mission is to optimize our understanding in people using machine learning and statistical modeling.
Identity is the unique Facebook advantage and the core asset to power Ads delivery, targeting and
measurement. Ads Identity Seattle team is working on cutting edge Machine Learning techniques to
strengthen people's identity on and off Facebook. We increase the coverage of Facebook Identity with
predictions outside of Facebook, build non-Facebook people Identity with traits predictions and cross
device graph clustering. We build audience recommendation models to automatically create targeting for
advertisers according to their objectives. We infer time dependent user intent from massive amount of on
and off Facebook data. The expertise of this team includes prediction, classification, statistical modeling,
distributed systems, and recommender systems.
Available Start Dates for 2019
Seattle
Start January 8 – End March 30
Start May 6 – End July 26
Start May 13 – End August 2
Start May 20 - End August 9
Start May 28 – End August 16
Start June 3 - End August 23
Start June 10 – End August 30
Start June 17 – End September 6
Start June 24- End September 13

Você também pode gostar