POME

This folder contains an html file "patterns.html" whcih described the patterns we used in POME, including the example sentences. Please ensure your internet connection is working when you open the file.

Click to view the patterns online.

File "data_rq1_rq2.csv" contains the dataset we manually labeled for RQ1 and RQ2. File "data_rq3_pome.csv" and "data_rq3_opiner.csv" contain the data used for RQ3 (collected from POME and Opiner, respectively), including the aspect and sentiment labeled by evaluators.

Inside "results", folder "results_rq1" contains the results of RQ1. Inside you can find the performance of 10 different machine learning approachs using seven different set of features. The results without SMOTE are in the folder "Without SMOTE", and the results with SMOTE are in the folder "With SMOTE". Result for pattern-matching approach is in the "pattern matching.csv" file.

File "results_rq2.csv" contains the raw results for RQ2, namely the polarity detection results of POME and other 6 state-of-the-art sentiment analysis tools we compared with in the paper.

This folder contains the source code of the core component of POME, namely the aspect and sentiment detector. In the source code folder, there is an example file for demonstrating the usage.

This folder contains the source code of labeling app we used during our study. In the source code folder, there is a README file for explaining the usage.