The purpose of sentiment analysis is to understand author’s emotion and attitude in a specific text. The attitude reflects the author’s personal emotions during writing, or the sentiments intended to be conveyed to readers through the text. Automated sentiment analysis is to extract subjective sentiment information from the text, with a combination of Natural Language Processing, Text Mining, computer language and other technical methods. The automated sentiment analysis model provided on the platform is developed by the DiVoMiner® team exclusively, and this technology is in patent application process.

Thanks to the team’s continuous upgrade of the algorithm model and rich corpus support, the tested accuracy of specific corpus reaches 0.7~0.9 (within the industry acceptance range). It is recommended that user determines the applicability of the model according to the research purpose. The test results are as follows:

Test Index	Online News	Weibo	Hotel Comments
Accuracy	0.70	0.85	0.80
F1-score	0.76	0.86	0.84
Precision	0.64	0.81	0.80
Recall	0.91	0.92	0.88

Tip: The test results above are based on the internal data set. And “accuracy” is a quantized value by general test index such as accuracy, F1-score, precision and recall for a specific test set. Limited to the test set, the “accuracy” range will fluctuate, and currently there is no strict industry standard.

Index description：

1. Accuracy: the proportion of correctly predicted samples to the total samples.
2. F1-score: the harmonic average of Precision and Recall.
3. The percentage of correctly predicted data to the overall predicted data.
4. Recall: the percentage of data that is correctly predicted to the actual data.

Reference:

Powers, David M W (2011). “Evaluation: From Precision, Recall and F-Measure to ROC, Informedness, Markedness & Correlation” (PDF). Journal of Machine Learning Technologies. 2 (1): 37–63.