Skip to content

amiangshu/SentiSE

Repository files navigation

SentiSE

SentiSE is a sentiment analysis tool for Software Engineering interactions

SentiSE, a supervised learning based sentiment analysis tool that incorporates ten supervised learning algorithms and fourteen different optional pre-processing steps that are commonly used to improve the performance of sentiment analysis tools. We empirically evaluated each of the algorithms and preprocessing steps to determine the best configuration. We evaluated SentiSE using a large-scale labeled dataset of 13K comments from three different types of SE interactions.

Performance Evaluation: We compare SentiSE with other sentiment analysis tool available in software engineering domain. We use two dataset for this evaluation. Orcale1 with 13 k labeled dataset with 21% positive, 60% neutral and 19% negative data and Oracle2 with 30% positive, 40% neutral and 30% negative data. Table bellow shows the performance comparison:

Oracle tool Precision
(Positive)
Recall
(Positive)
F-measure
(Positive)
Precision
(Positive)
Recall
(Positive)
F-measure
(Positive)
Precision
(Positive)
Recall
(Positive)
F-measure
(Positive)
Accuracy Weighted
Kappa
Orcal1 SentiSE 85.63% 75.27% 80.11% 81.51% 92.78% 86.78% 81.03% 55.92% 66.16% 82.23% 0.681
Oracle1 SentiCR 81.81% 76.59% 79.04% 80.04% 92.77% 85.92% 82.71% 46.38% 59.40% 80.6655% 0.647
Oracle1 SentiStrength-SE 75.81% 81.45% 78.53% 84.68% 83.64% 84.16% 66.50% 63.42% 64.92% 79.32% 0.6587
Oracle2 SentiSE 88.83% 85.09% 86.92% 86.62% 91.52% 89.00% 85.87% 78.61% 82.07% 86.92% 0.788
Oracle2 SentiCR 84.32% 84.73% 84.50% 80.70% 92.08% 86.00% 86.45% 59.49% 70.40% 82.47% 0.716
Oracle2 SentiStrength-SE 79.56% 83.57% 81.52% 80.73% 84.15% 82.41% 80.41% 69.31% 74.45% 80.34% 0. 696




Usage Instructions: Downaload and import SentiSe Project. Run the build.xml file and generate the sentise.jar. SentiSE is a commandline base tool. Use the command java -jar sentise.jar -help to find all the commands avialable in sentiSE.



ScreenShot SentiSE-cli