Defining sentiment rules

Define sentiment rules to augment the existing sentiment model, or to add sentiment processing for language models without built-in sentiment

Rules defined in a senti_list file or a senti_rules string follow the same format. In either case, rules may be separated by newline characters or by commas. Support for comma-delineated lists of rules was added in V‑Blaze version 7.5.

The general format for sentiment rules is {sentiment} {word} where {sentiment} is one of three special characters {+,o,-} corresponding to positive, neutral, and negative sentiment; and {word} is the word to be associated with the sentiment value.

For example, the following newline-separated list assigns sentiment value to six words.

+ fantastic
+ awesome
o okay
o basic
- awful
- rude

The preceding list assigns positive sentiment to the words fantastic and awesome, neutral sentiment to the words okay and basic, and negative sentiment to the words awful and rude.

This list format is valid for both senti_rules strings and senti_list files.

Note: With the 7.5 ASR engine release, the range that limits utterance-phrase capacity for sentiment detection was expanded. Previously, only utterances containing 2,3,4, or 5 words were evaluated for sentiment. Now, utterances containing as few as 1 or as many as 10 words are evaluated for sentiment.

If two sentiment rules define different values for the same word, the last value specified in the request is used for scoring. For example, given two sentiment files senfile1 and senfile2, final sentiment rules vary when the files are specified in different a order; in each case, the last rule specified in the request takes effect.

senfile1:                       
+ great                         
+ perfect		                

senfile2:
o great
- terrible

senfile1,senfile2:              
o great                         
+ perfect                       
- terrible                      

senfile2,senfile1:
+ great
+ perfect
- terrible