OCLAR: Logistic Regression Optimization for Arabic Sentiment Costumer Reviews
Marwan Al Omari N.A. · International Journal of Business Intelligence and Data Mining · 2021
The recognition and classification of sentiments in customer feedback are crucial for improving the service experience. Sentiment analysis (SA), as one of natural language processing (NLP) applications, evaluates customers' reviews by computing polarity text sequences. This paper extends the brute grid search methodology with pipeline architecture on OCLAR dataset. The architecture facilitates the search for inverse regularisation strength (IRS) parameter of logistic regression (LR). The experiments are evaluated on simultaneously different measures as like accuracy, area under the curve (AUC), etc. The experiments showed 0.10% improvement in AUC measure using bag-of-words (BoW) with n-gram levels of unigrams, bigrams, and trigrams, whereas 0.9% enhancement has been achieved by using term frequency and inverse document frequency (TF*IDF). In particular, the BoW classifier has achieved the largest improvement of 0.21% recall and 0.27% f-measure in negative predication, whereas it has achieved 0.03% precision and 0.01% f-measure in positive predication.