Related Pins at Pinterest: The Evolution of a Real-World Recommender System

Related Pins is the Web-scale recommender system that powers over 40% of user engagement on Pinterest. This paper is a longitudinal study of three years of its development, exploring the evolution of the system and its components from prototypes to present state. Each component was originally built with many constraints on engineering effort and computational resources, so we prioritized the simplest and highest-leverage solutions. We show how organic growth led to a complex system and how we managed this complexity. Many challenges arose while building this system, such as avoiding feedback loops, evaluating performance, activating content, and eliminating legacy heuristics. Finally, we offer suggestions for tackling these challenges when engineering Web-scale recommender systems

lambda-text-extractor

<span style="color: #24292e;">lambda-text-extractor</span> is a Python 3.6 app that works with the AWS Lambda architecture to extract text from common binary document formats.

Features

Some of its key features are:

  • out of the box support for many common binary document formats (see section on Supported Formats),
  • scalable PDF parsing using OCR in parallel using AWS Lambda and asyncio,
  • creation of text searchable PDFs after OCR,
  • serverless architecture makes deployment quick and easy,
  • detailed instruction for preparing libraries and dependencies necessary for processing binary documents, and
  • sensible unicode handling

Supported Formats

<span style="color: #24292e;">lambda-text-extractor</span> supports many common and legacy document formats:

  • Portable Document Format (<span style="color: #24292e;">.pdf</span>),
  • Microsoft Word 2, 6, 7, 97, 2000, 2002 and 2003 (<span style="color: #24292e;">.doc</span>) using Antiword with fallback to Catdoc,
  • Microsoft Word 2007 OpenXML files (<span style="color: #24292e;">.docx</span>) using python-docx,
  • Microsoft PowerPoint 2007 OpenXML files (<span style="color: #24292e;">.pptx</span>) using python-pptx,
  • Microsoft Excel 5.0, 97-2003, and 2007 OpenXML files (<span style="color: #24292e;">.xls</span>, <span style="color: #24292e;">.xlsx</span>) using xlrd,
  • OpenDocument 1.2 (<span style="color: #24292e;">.odm</span>, <span style="color: #24292e;">.odp</span>, <span style="color: #24292e;">.ods</span>, <span style="color: #24292e;">.odt</span>, <span style="color: #24292e;">.oth</span>, <span style="color: #24292e;">.otm</span>, <span style="color: #24292e;">.otp</span>, <span style="color: #24292e;">.ots</span>, <span style="color: #24292e;">.ott</span>) using odfpy,
  • Rich Text Format (<span style="color: #24292e;">.rtf</span>) using UnRTF v0.21.9,
  • XML files and HTML web pages (<span style="color: #24292e;">.html</span>, <span style="color: #24292e;">.htm</span>, <span style="color: #24292e;">.xml</span>) using lxml,
  • CSV files (<span style="color: #24292e;">.csv</span>) using Python csv module,
  • Images (<span style="color: #24292e;">.tiff</span>, <span style="color: #24292e;">.jpg</span>, <span style="color: #24292e;">.jpeg</span>, <span style="color: #24292e;">.png</span>) using Tesseract, and
  • Plain text files (<span style="color: #24292e;">.txt</span>)