Sentiment analysis precision - how can Neticle finetune precision for sentiments and labeling?

We are constantly teaching the system so that the array of recognized sentiment phrases and labels keeps increasing. The NLP engine is trained both in a supervised and an unsupervised way. Our team of testers manually checks the system on a weekly basis and corrects any possible mistakes. They also teach the engine any new phrases that it has come across. In addition, they incorporate our customers’ feedback to further improve sentiment precision and labeling.

Furthermore, changing the sentiment is also possible directly on the dashboard, allowing our clients to easily modify it on their own. The figure below illustrates where to find this option. For any specific mention, you can shift the sentiment in a positive direction by clicking the up arrow, while clicking the down arrow will change it in a negative direction.

We improve all languages simultaneously: the changes in one language are automatically translated into all the other languages Neticle handles. If the new sentiment rule passes the system’s verification process, it immediately becomes available in all our languages.

This constant teaching and finetuning keeps precision over 90% in all available languages.