Haystack 2019 Lightning Talk - State of Apache Tika - Tim Allison
0 likes368 views
This document discusses Apache Tika, a tool for extracting text and metadata from various file formats. It describes how Tika works and some challenges that may occur such as exceptions, unsupported formats, or memory issues. The document also mentions a tool called tika-eval that profiles Tika runs and exceptions. Future plans for Tika include improved CSV, ZIP file parsing and detection as well as more modularized statistics collection and language identification.
Haystack 2019 Lightning Talk - State of Apache Tika - Tim Allison
- 1. © 2019 The MITRE Corporation. All rights reserved. Apache Tika Tim Allison tallison@apache.org, @_tallison April 24, 2019 Haystack Conference Approved for Public Release; Distribution Unlimited. Case Number 18-3138-6
- 2. | 2 | © 2019 The MITRE Corporation. All rights reserved. Overview ▪ What is Tika ▪ tika-eval ▪ Running Tika safely ▪ Coming out in 1.21 and beyond
- 3. | 3 | © 2019 The MITRE Corporation. All rights reserved. Text/Metadata Extraction
- 4. | 4 | © 2019 The MITRE Corporation. All rights reserved. Things Can Happen ▪ Tired: – Exceptions – Unsupported file formats – Encrypted files – Garbled text – Missing text ▪ Wired: – OOM – Seg fault – Infinite loops – Multithreaded garbage collector pegging all CPU resources
- 5. | 5 | Stands up on Soap Box
- 6. | 6 | © 2019 The MITRE Corporation. All rights reserved. Upgrade from PDFBox 1.8.6->1.8.7
- 7. | 7 | © 2019 The MITRE Corporation. All rights reserved. Soap Box If your search system can’t tell the difference between those two…
- 8. | 8 | © 2019 The MITRE Corporation. All rights reserved. Soap Box If your search system can’t tell the difference between those two… You don’t have a search system.
- 9. | 9 | © 2019 The MITRE Corporation. All rights reserved. Soap Box If your search system can’t tell the difference between those two… 👍You’ve got a neat, little demo!👍 You don’t have a search system.
- 10. | 10 | Steps Off of Soap Box
- 11. | 11 | © 2019 The MITRE Corporation. All rights reserved. tika-eval ▪ Profile individual runs ▪ Compare two runs ▪ Exceptions by mime ▪ Out of vocabulary (OOV) statistics
- 12. | 12 | © 2019 The MITRE Corporation. All rights reserved. tika-eval: Eating our own dog food ▪ 3 million files (~1 TB) from Common Crawl and govdocs1 hosted on a public virtual machine, provided by Rackspace ▪ Code to profile a single run or compare two runs before release ▪ Evaluation methodology co-developed with and now co-run by open source colleagues (around the world) on the MSOffice parser project and the PDF parser project
- 13. | 13 | © 2019 The MITRE Corporation. All rights reserved. Tika 1.21 and beyond ▪ Tika 1.21 – csv/tsv detector and parser (Apache commons-csv) – Improved zip-based (.docx, .pptx, .xlsx) file detection and parsing ▪ Beyond – Modularize tika-eval and include stats within the extract for scalability and aggregation of stats w/in Solr/Elastic – Increase coverage/speed of zip-based file detection; can we move entirely to streaming detection? – Improve language coverage/lang id component w/in tika-eval ▪ Help! – What do you need? – How can you help us help you?