Hello s.s Mahtab

The accuracy is a measure related to the total number of elements correctly identified in relation to a total amount of data in the searches / analyzes you are making. In other words, is a percentual of the amount of the things that gone right in your process.

Translating this into this area of sentiment analysis, t could be something like this:

Accuracy = (all correct measurements) / (total data)

Take a read on these two questions, that are aligned with your one:

https://www.quora.com/How-do-I-measure-performance-of-a-sentiment-analysis-system

https://stackoverflow.com/questions/45088949/how-can-i-calculate-precision-and-recall-for-sentiment-analysis-multi-class-clas

In both links there are answers that can help you.

Regards.

An any other technique, the accuracy can be evaluated in terms of a data that are already known

I think, it is classification accuracy of the customers sentiment. If it is, then the accuracy will be the ratio of the number of truly classified sentiments sample and the total sample used ( for testing or training)

Join us in Granada at SAMSN 2019

http://emergingtechnet.org/SAMSN2019/

Having a labeled (annotated) training dataset will help. During model construction we usually use an annotated dataset. We divide the dataset for training and test purposes into two parts ( sometimes it may be three also for validation). The accuracy can be checked by comparing annotated test records. However rather than using only accuracy rate F-measure, TP ( True Positive), FP (False Positive) will also help. Common accuracy ratios have been also given in this study : Thesis Applying Machine Learning and Natural Language Processing Te...