How accurate is Germin8’s sentiment algorithm?
The accuracy of Germin8’s sentiment algorithm varies from tracker to tracker, depending on the industry. The sentiment tagged to all posts for a brand is usually between 70%-90% accurate.
If a post contains several topics, what kind of natural language/grammar markers are considered while tagging sentiment for each component of a topic phrase?.
We use various algorithms like text segmentation, ontology tagging and anaphora resolution to identify the text span of each topic. Following that, the algorithm calculates the sentiment of the topic by analysing the sentences in the text span.
Can we customize the sentiment rules for specific brands/trackers, where owing to either the brand name or words used by consumers with regards to the brand, the sentiment score can be made more accurate?
We do not support any user customization of sentiment rules. However, bulk verification rules can be set up to override the system tagged sentiment. Additionally, brand names get automatically ignored by the sentiment algorithm (so a brand name like Force India or Reliance won't impact the sentiment algorithm).