The sentiment analysis app allow you to analyze user opinion, converting text to numerical ratings.
This app includes:
From a command shell, you can train it on your own data. The training produces a (naive bayes classifier) model file used for the sentiment command, but training also outputs a lookup table of a few thousand terms and their expected sentiment rating, so the vocabulary can be tweaked and used with lookup commands to also do some basic sentiment analysis.
A note about accuracy: Over the twitter data, assuming 50-50 positive and negative tweets, it's about 70% accurate. Assuming the actual distribution of tweets, which seems to be about 75% positive, it's about 85% accurate. The most important part is not the accuracy on any given text but on the aggregate -- to see how sentiment trend over time or how "google" vs "apple" compare with other.
handle non-utf-8 more gracefully
Fixed bug in tokens command. Reorganized 'local' into 'default'
updated app so commands are exported properly to other apps
updated 'tokens' to not give error messages unappropriately. updated 'heat' terms to return fewer false positives of swearing and violent terms.
updated package
Fix a major bug where only the first batch of results (e.g., 50) were getting sentiment but all other events were empty.
Performance speed up.
Added support for distributed environments. Previously model files were not being distributed to indexers, cause an error and sentiment failing to work.
As a Splunkbase app developer, you will have access to all Splunk development resources and receive a 10GB license to build an app that will help solve use cases for customers all over the world. Splunkbase has 1000+ apps from Splunk, our partners and our community. Find an app for most any data source and user need, or simply create your own with help from our developer portal.