Salience 6.4.0 Release 2018-11-16 (r2501)

This is the Salience 6.4 release, introducing sentiment improvements, speed up of entity sentiment, and more.


  • Improvements to english sentiment dictionary (fewer terms that were frequently false hits).
  • Improved patterns for determining which words in a sentence are negated smaller/faster custom trained models.
  • Sentiment defined by a query is now applied to topics and entities defined by queries
  • Theme and entity sentiment are now faster to calculate. This is especially noticable for very long documents.
  • Wrappers now all expose lengths and offsets in characters as well as bytes. A unicode character (e.g. an emoji) is 1-3 bytes long, but will be treated as a single character in the character lengths/offsets.
  • With SimpleSentiment option enabled, stemmed terms no longer match the full term and subwords (e.g. inside hashtags) are not given sentiment
  • Improved multithreaded performance in Python.

Bug fixes

  • Fixed support for *{n,m} count patterns in relationships (where you specify min and max counts for a token).
  • Fixed bug calculating entity/theme sentiment when AllSentimentPhrases option was enabled.
  • Fixed options with missing names in dump_environment.
  • Salience expressions (e.g. entity patterns) were not correctly parsing negative numbers.
  • Token length in characters was incorrect for abbreviated words (U.S.A. would be reported as 5 characters instead of 6). Length in bytes was fine.
  • Fixed memory leak when reloading industry packs.
  • Metaentities were not being returned.

Upgrade Directions

  1. You must recompile any wrappers (e.g. java_salience.dll/.so) you've built. If you use a wrapper we ship with (e.g. java), things should be fine. If you built it (e.g. you ran "python install") you'll have to rerun that. There have been no backwards incompatible changes, so any consuming code will continue working as before.

  2. You must update all first tier language directories you use. Downloads are available for all languages on the support portal. Since that's a lot to download we've also uploaded a file with the files that have changed in all non-English languages. Unfortunately, since we don't require you to place your language directories in a specific location or give them a specific name, we can't automate the unpacking of these. If your directories are all in one location and named like it-data, de-data, etc. then unzipping this file in the parent directory should work. If not, you'll have to copy the changed files and directories over (or use the full installations).

  3. If you use Basic Languages, the runner for that has been updated. If you reinstall any of the basic languages (Hebrew, Arabic, etc.) you'll be all set for all of them.

If you have any questions, please email [email protected]