This is the Salience 6.2 release, primarily featuring improvements to entities, sentiment and queries.
· By default, Salience will now search for '#term' and '@term' when you query on 'term'.
· + and - can be prepended to queries to only match on negated or nonnegated mentions.
· NOTNEAR and NOTWITH query operators for excluding hits based on proximity between phrases.
· Salience ships with HSDTrainer tool, to generate potential new hsd entries from a set of documents divided into positive, negative and neutral sets.
· Options can now be set via data file instead of programmatically.
· Better English entity recognition accuracy.
· User entities can now be matched by query metadata tokens.
· Better theme, entity and topic sentiment assignment.
· Sentiment phrases that led to a topic hit's sentiment are now reported.
· Phrases that normally bear sentiment but don't in a particular case (e.g. appearing in a question) are reported in sentiment phrase lists.
· Better handling of sentiment phrases that appear with various modifiers ('not great', 'very great', 'extremely great!' all appearing in a single document.
· Finer grained options to control operation of sentiment algorithm.
· Support for stemming of #terms and @terms.
· Option to allow themes to overlap each other.
· HSD sentiment values of 0 now turn off a phrase rather than universally decreasing the magnitude of sentiment, as this is a more intuitive behavior.
· Updated emoji support to Unicode 9.0 standard.
· Stopwords are now case insensitive.
· Some minor improvements to various datafiles.
· Fixed bug in collection processing where sentiment could be taken from the wrong sentences when calculating theme sentiment.
· Salience was incorrectly throwing an exception when options were set if an HSD with invalid queries was being used.
· Removed "intensifiers" with a multiple of 1, since they were causing confusion.
If you have any questions, please email [email protected]
Updated 8 months ago