Jigsaw Multilingual Toxic Comment Classification Kaggle Competition

Total Page:16

File Type:pdf, Size:1020Kb

Jigsaw Multilingual Toxic Comment Classification Kaggle Competition Jigsaw Multilingual Toxic Comment Classification Kaggle Competition Gabriela Ožegovic´ Sven Celin Graz University of Technology Graz University of Technology [email protected] [email protected] ABSTRACT to select specific types of toxicity and focus on them, since Great expansion of the internet has produced wide range of some sites might be fine with one type of toxicity (e.g. severe people to come online. All this different cultures have a lot of toxicity) and not others. things different from political views, religion, economics or In 2019, another competition on similar topic was held. In even their favourite singer or actress, and with this differences the "Unintended Bias in Toxicity Classification Challenge" people start to act irrationally and start to fight online. This [4], focus was on detecting the toxicity while minimizing things could happen in daily matter from internet bullying, the unintended model bias. The unintended model bias is online harassment or personal attacks. With this project idea happening while trying to detect the toxicity of comments was to limit toxic comments written by the users and flag them which contained the names of identities which are frequently as inappropriate. In this work, we will show how we flagged attacked or used in offensive ways. Training a model on this toxic comments from large dataset we have been provided by type of data ends with classifying a text as toxic just because Kaggle’s competition regarding this topic. they contain certain identity, even though it is not toxic itself. Author Keywords Text classification, Text mining, Toxic text classification, Word embeddings, word2vec, Feature extraction KAGGLE Kaggle is one of biggest data science and machine learning INTRODUCTION communities where users are publishing datasets and kernels All platforms that serves a lot of people will, in one point for everyone to see [11]. This web-based system allows data of their existence, have disagreements, abuse and harassment scientists and machine learning engineers to enter competi- from certain groups of people that disagree with their ideas. tions where they try to solve data science challenges. In the Only one comment is enough to derail an online discussion. newest instance of Kaggle they allowed usage of TPU and To counter that flow of non constructive comments this com- GPU cores on their cloud servers for challengers to use. That petition was made to yield the best algorithm for flagging sped up many machine learning algorithms and brought data toxic comments. As platforms struggle to effectively enable science closer for people that have no GPU or TPU cores in conversations in their comment sections, many limit or com- their computers. pletely shut down user comments. This project was made to focus machine learning models to identify toxicity in online conversations and flag them as rude, disrespectful or likely to make someone leave discussion. If this comments could Competitions be identified that would lead to safe and more collaborative Kaggle competitions sole purpose is to improve data science thread. and machine learning in particular field or in particular topic from the industry[1]. Every competition is made of overview Related work where competition maker write description of the work and In 2018, a competition was held on Kaggle called "Toxic data, evaluation, timeline, prizes and code requirements. Sec- Comment Classification Challenge" [3]. In that competition, ond, and the most important thing is the dataset that is provided competitors were asked to build models not only to recognize by the competition maker, in this part of the competition con- toxicity, but to also detect few types of toxicity. The types of testants can access all the data that competition requires to toxicity which had to be detected are: severe toxicity, obscene, make your model and assumptions. For contestant, notebooks threat, insult and identity hate. The goal was to enable users section is go to place when submitting their result. In this sec- Permission to make digital or hard copies of all or part of this work for personal or tion, contestant needs to upload (or make in online editor) their classroom use is granted without fee provided that copies are not made or distributed Python notebook and run it on Kaggle servers with all data for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this work owned by others than the provided. After execution of the notebooks in cloud servers author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or user is being placed in Leaderboard by their score number. In republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. Request permissions from [email protected]. competitions there are often more than few thousand teams with maximum number of 5 contestants per team. We could © 2020 Copyright held by the owner/author(s). Publication rights licensed to ACM. say that this competitions attracts large number of people from ISBN 978-1-4503-2138-9. all over the world. DOI: 10.1145/1235 JIGSAW MULTILINGUAL TOXIC COMMENT CLASSIFICA- possibility of false negatives after the training, we decided to TION balance the dataset and downsample it. That seemed like the In this year’s competition, main challenge was to build mul- best solution, since the algorithm that uses unbalanced training tilingual models for toxicity classification with English-only set will most likely not learn to discriminate from the features. training data [5]. The ratio we opted for is 1:2. So, for each toxic comment there are 2 not-toxic ones. Jigsaw developed Perspective API, which uses machine learn- ing models to determine an impact a comment could have After that, we did some feature extraction processes, which are on the conversation [8]. It is served in multiple languages explained in the next chapter. In the end, we used a classifier of (English, Spanish, French, German, Portuguese and Italian). choice to train the model and predict the toxicity probabilities Recently, there are impressive multilingual capabilities, in- of test data. cluding few- and zero-shot learning, and the goal is to see it these results are applicable to toxicity classification. FEATURE EXTRACTION Feature extraction is a process of distillation of characteristics The host team shared two notebooks to help competitors start from the raw dataset and reducing it to manageable groups for running BERT. BERT (Bidirectional Encoder Representations processing [2]. Characteristic of datasets is that the number from Transformers) is a method of pre-training language rep- of features which they provide is not always appropriate for a resentations developed by Google, which presented state-of- task for which the same dataset is used. In the situation with the-art results in a wide variety of NLP tasks. datasets provided for this competition, there are no available The competitors had a week to work on their models and to features which could help determining the toxicity of a com- try get the best solution. The expected result is the probability ment. Only provided features are comment itself, it’s language that given comment is toxic, a value which is between 0 and 1. and whether or not is the comment toxic. For this reason, it is important to find the possible characteristics which are closely related to the toxicity level and extract them as separate fea- tures. We did some brainstorming, and though about in which way would a user type a comment if they are, e.g., angry. Extracted features and their significance will be explained in further sections. Comment length Figure 1. Competition When thinking about toxic comment and toxic behaviour in writing, we thought about the length of each comment. We Dataset assumed that people, who are willing to be toxic, and angrier Provided training data is English-only, and it is the same data and have less nerves than the ones who did not plan on being which was provided for last two competitions (mentioned in toxic. We can assume that they put less thought into their 1.1.). Both training datasets are given in unprocessed and comments, and often, possibly in heat of an argument, can processed format. The unprocessed format is just the data rush into them. which was given before, and the processed format is done For this reason we checked the length of comment, and found in a way that it is ready for input to BERT. Data consists of out that there is a difference between average length of toxic columns such as id, comment text, toxic (whether the toxic and non-toxic comments. Toxic comments ended up indeed is comment or not, or the probability of the comment to be being shorter than the non-toxic ones. The correlation is a toxic), and types of toxicity each one in separate column. negative value because the higher the comment length is, the Test data is consisted of comments from Wikipedia talk page lesser the toxicity is. in several different languages (Italian, Spanish, Turkish, Por- tuguese, Russian and French). Test data consists of few columns: id, comment content and language of the content. Other than that, competitors were provided validation data, Figure 2. Point-biserial correlation with toxicity and comment length which is, just as the test data, just in non-English languages. It consists of columns: id, comment text, language and toxic column. Punctuation count Similar to the previous idea about length of comments, here our idea was that people who are toxic are more prone to Problem approach using multiple punctuation signs, like "???" or "!!!". This is The first step was to inspect the data.
Recommended publications
  • Machine Learning Is Driving an Innovation Wave in Saas Software
    Machine Learning is Driving an Innovation Wave in SaaS Software By Jeff Houston, CFA [email protected] | 512.364.2258 SaaS software vendors are enhancing their soluti ons with machine learning (ML) algorithms. ML is the ability of a computer to either automate or recommend appropriate actions by applying probability to data with a feedback loop that enables learning. We view ML as a subset of artificial intelligence (AI), which represents a broad collection of tools that include/leverage ML, such as natural language processing (NLP), self-driving cars, and robotics as well as some that are tangent to ML, such as logical rule-based algorithms. Although there have been multiple “AI Winters” since the 1950s where hype cycles were followed by a dearth of funding, ML is now an enduring innovation driver, in our opinion, due in part to the increasing ubiquity of affordable cloud-based processing power, data storage—we expect the pace of breakthroughs to accelerate. automation, customer support, marketing, and human resources) through both organic and M&A These innovations will make it easier for companies initiatives. As for vertical-specific solutions, we think to benefit beyond what is available with traditional that a new crop of vendors will embrace ML and business intelligence (BI) solutions, like IBM/ achieve billion-dollar valuations. VCs are making big Cognos and Tableau, that simply describe what bets that these hypotheses will come to fruition as happened in the past. New ML solutions are demonstrated by the $5B they poured into 550 amplifying intelligence while enabling consistent and startups using AI as a core part of their solution in better-informed reasoning, behaving as thought 20162.
    [Show full text]
  • Economic and Social Impacts of Google Cloud September 2018 Economic and Social Impacts of Google Cloud |
    Economic and social impacts of Google Cloud September 2018 Economic and social impacts of Google Cloud | Contents Executive Summary 03 Introduction 10 Productivity impacts 15 Social and other impacts 29 Barriers to Cloud adoption and use 38 Policy actions to support Cloud adoption 42 Appendix 1. Country Sections 48 Appendix 2. Methodology 105 This final report (the “Final Report”) has been prepared by Deloitte Financial Advisory, S.L.U. (“Deloitte”) for Google in accordance with the contract with them dated 23rd February 2018 (“the Contract”) and on the basis of the scope and limitations set out below. The Final Report has been prepared solely for the purposes of assessment of the economic and social impacts of Google Cloud as set out in the Contract. It should not be used for any other purposes or in any other context, and Deloitte accepts no responsibility for its use in either regard. The Final Report is provided exclusively for Google’s use under the terms of the Contract. No party other than Google is entitled to rely on the Final Report for any purpose whatsoever and Deloitte accepts no responsibility or liability or duty of care to any party other than Google in respect of the Final Report and any of its contents. As set out in the Contract, the scope of our work has been limited by the time, information and explanations made available to us. The information contained in the Final Report has been obtained from Google and third party sources that are clearly referenced in the appropriate sections of the Final Report.
    [Show full text]
  • Youtube-8M Video Classification
    YouTube-8M Video Classification Alexandre Gauthier and Haiyu Lu Stanford University 450 Serra Mall Stanford, CA 94305 [email protected] [email protected] Abstract metadata. To this end, Google has released the YouTube- 8M dataset [1]. This set of over 8 million labeled videos al- Convolutional Neural Networks (CNNs) have been lows the public to train algorithms to learn how to associate widely applied for image classification. Encouraged by labels with YouTube videos, helping to solve the problem these results, we apply a CNN model to large scale video of video classification. classification using the recently published YouTube-8M We use multilayer convolutional neural networks to clas- dataset, which contains more than 8 million videos labeled sify YouTube-8M videos. The input to our algorithm is with 4800 classes. We study models with different archi- pre-processed video data taken from individual frames of tectures, and find best performance on a 3-layer spatial- the videos. We then use a CNN to output a score for each temporal CNN ([CNN-BatchNormzliation-LeakyReLU]3- label in the vocabulary. We briefly tried LSTM and fully- FullyConnected). We propose that this is because this connected models, but found that convolutional models per- model takes both feature and frame information into ac- form better. count. By optimizing hyperparameters including learning rate, we achieve a Hit@1 of 79.1%, a significant improve- 2. Related Work ment over the 64.5% for the best model Google trained in Large-scale datasets such as ImageNet [2] have played their initial analysis. Further improvements may be achiev- a crucial role for the rapid progress of image recognition able by incorporating audio information into our models.
    [Show full text]
  • Student Resume Book
    Class of 2019 STUDENT RESUME BOOK [email protected] CLASS OF 2019 PROFILE 46 48% WOMEN CLASS SIZE PRIOR COMPANIES Feldsted & Scolney 2 Amazon.com, Inc Fincare Small Finance Bank American Airlines Group General Electric Co AVERAGE YEARS American Family Insurance Mu Sigma, Inc PRIOR WORK Bank of New York, Qualtrics LLC EXPERIENCE Melloncorp Quantiphi Inc Brookhurst Insurance Skyline Technologies Services TechLoss Consulting & CEB Inc Restoration, Inc Cecil College ThoughtWorks, Inc Darwin Labs US Army Economists, Inc UCSD Guardian United Health Group, Inc PRIOR DEGREE Welch Consulting, Ltd CONCENTRATIONS ZS Associates Inc & BACKGROUND* MATH & STATISTICS 82.61% ENGINEERING 32.61% ECONOMICS 30.43% COMPUTER SCIENCE & IT 19.57% SOCIAL SCIENCES 13.04% HUMANITIES 10.87% BUSINESS 8.70% OTHER SCIENCES 8.70% *many students had multiple majors or OTHER 8.70% specializations; for example, of the 82.61% of students with a math and/or statistics DATA SCIENCE 2.17% background, most had an additional major or concentration and therefore are represented in additional categories. 0 20 40 60 80 100 Class of 2019 SAURABH ANNADATE ALICIA BURRIS ALEX BURZINSKI TED CARLSON IVAN CHEN ANGELA CHEN CARSON CHEN HARISH CHOCKALINGAM SABARISH CHOCKALINGAM TONY COLUCCI JD COOK SOPHIE DU JAMES FAN MICHAEL FEDELL JOYCE FENG NATHAN FRANKLIN TIAN FU ELLIOT GARDNER MAX HOLIBER NAOMI KADUWELA MATT KEHOE JOE KUPRESANIN MICHEL LEROY JONATHAN LEWYCKYJ Class of 2019 HENRY PARK KAREN QIAN FINN QIAO RACHEL ROSENBERG SHREYAS SABNIS SURABHI SETH TOVA SIMONSON MOLLY SROUR
    [Show full text]
  • Implementation of Machine Learning with Google Cloud Platform
    Cloud Computing Implementation of Machine learning with Google cloud platform Referent : Prof. Dr. Christian Baun Submitted by: Ammar Albaalbaki(1267651) Anish Joys Yesuadimai Michael(1280214) Eigenständigkeitserklärung Ich versichere wahrheitsgemäß, die Arbeit selbstständig angefertigt, alle benutzten Hilfsmittel vollständig und genau angegeben und alles kenntlich gemacht zu haben, was aus Arbeiten anderer unverändert oder mit Abänderungen entnommen wurde. __________________ Ammar Albaalbaki Ich versichere wahrheitsgemäß, die Arbeit selbstständig angefertigt, alle benutzten Hilfsmittel vollständig und genau angegeben und alles kenntlich gemacht zu haben, was aus Arbeiten anderer unverändert oder mit Abänderungen entnommen wurde. ------------------------------ Anish joys yesuadimai michael Table of Contents 1 Introduction 1 1.1 Motivation 1 1.2 Purpose 1 1.3 The aim of the work 2 2 Fundamental 4 2.1 Google Cloud 4 2.2 Machine Learning 5 2.2.1 Supervised Learning 6 2.3 Image Processing 7 2.4 Webhosting 8 2.4.1 Basic of response and request 9 2.4.2 Describe PHP 9 2.4.3 Putty to connection with host 10 3 Implementation 11 3.1 Log in Google Cloud 11 3.2 Google Cloud Storage 11 3.3 Datalab for implementing image processing 14 3.4 Create Web Host 19 4 Evaluation and results 23 5 Summary and outlook 24 5.1 Summary 24 5.2 outlook 24 List of Figures I List of tables II References III Page I 1 Introduction Cloud computing is a very hot and booming topic in the current era, lots of companies have already moved to cloud computing technologies and many companies are urging to move to cloud.
    [Show full text]
  • Machine Learning in Google Drive
    Senior Staff Software Engineer, Google Google I/O Extended Boulder 8-May-2018 #io18extended ...like space :) #spaceishard Spaceflight is unforgiving and complicated. But it’s not a miracle. It can be done #intelligenceishard ● Fundamentals of machine learning are complicated; ML is easily misapplied ● Doing “toy” things with ML is easy. ○ Doing useful things is a lot harder How to get there from here? ● Start small, but real ● Measure. Be rigorous. Make sure you’re helping ● Have a compelling UX ● Launch. Iterate. Improve. ● Never forget the user #io18extended Drive Quick Access Main Idea: Prominently show the user's documents and files they likely want to open right now Benefit 1: Save Users Time ● Quick Access gets users to their files 50% faster… and with less cognitive friction Benefit 2: Enable users to make better business decisions ● Show users relevant documents to their pending business decisions, including documents they may not be aware of. (The right information at the right time.) #io18extended Drive Quick Access: Mobile (Android and iOS) #io18extended What we’ve learned: Quick Access Feature Intelligence works ● Training and using Machine Learning models results in improved metrics beyond a simple “Most Recently Used” across all metrics Quick Access saves users time ● About 50% of opens come from QA, each one saves 50% on “finding time” Starting Point for Future Work ● Quick Access has proven out our machine learning infrastructure for Drive and provides a framework for future intelligence features. #io18extended Quick Access
    [Show full text]
  • The Big Analytics Book Project Is an Aggregation of Articles That Best Represent Thoughts by Some of the Leading Minds in the Data Analytics Industry
    The Big Analytics Leaders Collaborative Book Project For, Of, and By the Data Analytics Leaders and Influencers Compiled by Deeksha Joshi & Vishal Kumar Contents SECTION A: STRATEGY FOR THE DATA ANALYTICS LEADERS ................................................... 8 1. Rise of Data Capital by Paul Sonderegger ........................................................................................ 9 2. Investing in Big Data by Bill Pieroni .............................................................................................. 13 3. From Data to Insight: Seven Tips for a Great Data Strategy by Anne Russell ........................ 16 4. Data is Not Always a Substitute for Strategy by Steven Sinofsky ............................................. 20 5. Understand How Your Company Creates Value by Cary Burch ............................................... 26 6. Act Like an Analyst, Think Like a Strategist by Dr. Justin Barclay .......................................... 28 7. The Last Mile of Intuitive Decision Making by Vishal Kumar ................................................. 31 8. Three BIG Reasons Companies are Failing with Big Data Analytics by Ian St. Maurice ..... 36 9. How to Leverage Power of Big Data Analytics in Making Strategic Decisions by Ankush Chopra ........................................................................................................................................................... 40 10. How to Hire the Best Analytics Leaders: And it’s not the way you think! by Tony Branda 43 11. Big Decisions
    [Show full text]
  • Esercitazione 4 - Data Studio
    Esercitazione 4 - Data Studio Sistemi per la gestione delle basi di dati 1. Login Connect to Google Data Studio, login with your Google Account or create a new free Google Account. ● https://datastudio.google.com 1 2. Welcome report Click on “Welcome to Data Studio” and follow the tutorial. ● https://datastudio.google.com/reporting/0B5FF6JBKbNJxOWItcWo2SVVVeGc ● learn the basics of the Data Studio tool by copying the “Welcome report” and following the step- by-step instructions provided ● page 8, “Track report usage with Google Analytics”, can be safely skipped 3. Template report Preliminary steps to clone a template report Start from a provided template report to create new data visualizations. ● Return to the Data Studio home page ● Click on “All templates” to open the template gallery ● Choose the “World Population Data” template 2 ● Click on the “Use Template” button to confirm the selection, and accept the Terms and Conditions if you haven’t yet ● Click on the “Create Report” button to confirm the data source selection “[Sample] World Population Data 2005 - 2014” 3 ● If prompted, allow Google Drive access to Data Studio You have just created a new private report, based on the “World Population Data” template 4 Analyze the World Population data source ● Click on the “Add a page” button ● Add the following analyses to the new page of the report ● (Query A) – Add a table in the report to select the following data. Considering only year 2013, select the top-10 countries with the highest “internet %”, and their “population” and “internet users”. ● (Query B) – Add a pivot table in the report to select the following data.
    [Show full text]
  • Analysis on Youtube Trending Videos
    International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 07 Issue: 08 | Aug 2020 www.irjet.net p-ISSN: 2395-0072 Analysis on YouTube Trending Videos Swati Gayakwad1, Rajas Patankar2, Dashrath Mane3 1PG Student, Vivekanand Education Society’s Institute of Technology, Dept. of MCA, Mumbai, India 2PG Student, Vivekanand Education Society’s Institute of Technology, Dept. of MCA Mumbai, India 3Assistant Professor, Vivekanand Education Society’s Institute of Technology, Dept. of MCA, Mumbai, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - Online video streaming platforms are heavily Provide content creators and other digital businesses used nowadays. Websites such as YouTube offers content with information about features required to make a creators a great platform to share their knowledge, ideas and video that trends and reaches to maximum people. interesting information to their viewers. For a video to reach to maximum people, YouTube offers a trending page on To accelerate a channels growth and increase their website that shows videos which are trending at that subscriber count by making trending videos. particular time. Other than few viral videos that achieve high view count which are predictable to end up in trending section, By increasing subscriber count and views this can help rest of the videos cannot be predicted. Corporate companies you youtuber's to earn more money, sponsorships and are using social media for improving their businesses, the data opportunities to excel in their social media carrier mining and analysis are very important in these days. This paper deals with analysis of YouTube Data on Trending 2. METHODS OF ANALYSIS Videos. The analysis is done using user features such as Views, Comments, Likes, and Dislikes.
    [Show full text]
  • Google Playstore Application Analysis and Prediction”
    “Google Playstore Application Analysis and Prediction” A PROJECT Submitted in partial fulfillment of the requirements for the award of the degree of BACHELOR OF TECHNOLOGY IN COMPUTER SCIENCE AND ENGINEERING Under the supervision of Dr. Hemraj Saini (Associate Professor) By Shubham Ruhela (151364) to JAYPEE UNIVERSITY OF INFORMATION TECHNOLOGY WAKNAGHAT, SOLAN – 173234 HIMACHAL PRADESH, INDIA CERTIFICATE I hereby declare that the work presented in this report entitled “Google Playstore Application Analysis and Prediction” in partial fulfillment of the requirements for the award of the degree of Bachelor of Technology in Computer Science and Engineering submitted in the department of Computer Science & Engineering and Information Technology, Jaypee University of Information Technology, Waknaghat is an authentic record of my own work carried out over a period from August 2018 to May 2019 under the supervision of Dr. Hemraj Saini, Associate Professor, department of Computer Science & Engineering. The matter embodied in the report has not been submitted for the award of any other degree or diploma. Shubham Ruhela (151364)…………. This is to certify that the above statement made by the candidate is true to the best of my knowledge. Dr. Hemraj Saini (Associate Professor) Department of Computer Science & Engineering Dated: i ACKNOWLEDGEMENT We take upon this opportunity endowed upon us by grace of the almighty, to thank all those who have been part of this endeavor. First & Foremost, we want to thank our supervisor 'Dr. Hemraj Saini‟ for giving us the correct heading and legitimate direction in regards to the subject. Without his dynamic association and the correct direction this would not have been conceivable.
    [Show full text]
  • State of Data Infrastructure and Analytics Report 2021
    THE STATE OF DATA INFRASTRUCTURE & ANALYTICS REPORT 2021 Work-Bench is an enterprise-focused THE STATE OF DATA INFRASTRUCTURE venture capital firm based in New York City. & ANALYTICS REPORT Work-Bench was founded with a unique, thesis-driven approach that flips traditional The State of Data Infrastructure & Analytics Report is a collection of the top enterprise venture capital upside down by technological trends and tools that have been dominating the modern data first validating Fortune 500 IT pain points infrastructure landscape over the past year. through our extensive corporate network and then investing in the most promising As an early stage investor at Work-Bench, I’ve been keeping an active pulse companies addressing these challenges. on the data ecosystem by speaking with data practitioners and startup founders who are laying the groundwork in the space as well as corporate www.work-bench.com executives from our network to develop my understanding of the top challenges and opportunities in data. @Work_Bench [email protected] This report is a culmination of these conversations and illustrates the next generation of tools that we think are best poised to tackle the next phase in data infrastructure. AUTHORED BY: PRIYANKA SOMRAH � Additionally, listen to the recording of The State of Data Infrastructure & Analyst Analytics Webinar, where I walk through the report and dive a layer deeper. [email protected] 2 The Rise of Data Engineering 4 Trends Shaping The Modern Data Stack 11 TABLE OF Data Catalog, Lineage & Discovery 13 CONTENTS The Last Mile Problem in ETL 19 The Rise of Operational Analytics 28 Augmenting Stream & Batch 31 Processing 3 THE RISE OF DATA ENGINEERING & WHAT IT MEANS FOR THE MODERN DATA STACK THE RISE OF DATA ENGINEERING IS BRIDGING THE TOOLING GAP IN THE MODERN DATA ECOSYSTEM Today, there are more data engineering jobs We are on the cusp of a new era in data where than ever before! innovation in data engineering is shaping the future of data-driven organizations and their initiatives.
    [Show full text]
  • Data Quality 2
    P.S., Want a summary of ML advancements? ml-surveys Table of Contents 1. Data Quality 2. Data Engineering 3. Data Discovery 4. Classification 5. Regression 6. Forecasting 7. Recommendation 8. Search & Ranking 9. Embeddings 10. Natural Language Processing 11. Sequence Modelling 12. Computer Vision 13. Reinforcement Learning 14. Anomaly Detection 15. Graph 16. Optimization 17. Information Extraction 18. Weak Supervision 19. Generation 20. Validation and A/B Testing 21. Model Management 22. Efficiency 23. Ethics 24. Practices 25. Fails Data Quality 1. Monitoring Data Quality at Scale with Statistical Modeling Uber 2. An Approach to Data Quality for Netflix Personalization Systems Netflix 3. Automating Large-Scale Data Quality Verification (Paper) Amazon 4. Meet Hodor — Gojek’s Upstream Data Quality Tool Gojek 5. Reliable and Scalable Data Ingestion at Airbnb Airbnb 6. Data Management Challenges in Production Machine Learning (Paper) Google 7. Improving Accuracy By Certainty Estimation of Human Decisions, Labels, and Raters (Paper) Facebook Data Engineering 1. Zipline: Airbnb’s Machine Learning Data Management Platform Airbnb 2. Sputnik: Airbnb’s Apache Spark Framework for Data Engineering Airbnb 3. Introducing Feast: an open source feature store for machine learning (Code) Gojek 4. Feast: Bridging ML Models and Data Gojek 5. Unbundling Data Science Workflows with Metaflow and AWS Step Functions Netflix Data Discovery 1. Amundsen — Lyft’s Data Discovery & Metadata Engine Lyft 2. Open Sourcing Amundsen: A Data Discovery And Metadata Platform (Code) Lyft 3. Using Amundsen to Support User Privacy via Metadata Collection at Square Square 4. Democratizing Data at Airbnb Airbnb 5. Databook: Turning Big Data into Knowledge with Metadata at Uber Uber 6.
    [Show full text]