ABSTRACT

In this research study word labelling and sentiment analysis has been studied for the Romanized Sindhi text. As the use of Sindhi script in the text communication has been increased on various platforrms by peoples of different regions especially in Sindh, Province of Pakistan. But they feel difficulty to write Sindhi script on various like mobile massages, face book massages, tweets on twitter, WhatsApp massages and etc. Romanization of Sindhi script is one of the best options to resolve the problems of users of Sindhi scripts on different platforms. Sindhi language has its script (52 letters of alphabet) and written by the right-handed. For written purposes of Romanized text in text communication, there are many issues and challenges (Written script with Romanized style as the English style, Space Issues in Romanized script, some Characters are not suitable in Romanized Sindhi, Some Characters are not suitable in Romanized Sindhi, Paragraph, Row, Characters Issues, Punctuation. Row Break and Font Style) were found during the Romanization of Sindhi text. So for the solution of these issues and challenges rules were created for the Romanization of Text. Romanized Sindhi text increased in rapid rate for the communication purpose. that's why the sentiment classification and word labelling (Tokenization and Parts-of-speech tagging) of Romanized Sindhi text has been studied in this research study. Algorithm and Rule based model (Hybrid approach) were used for the Word labelling of Romanized Text. Word Labelling and Sentiment classification of Romanized Sindhi Text was performed on the online Python tool on three thousand sixty three words of Romanized Sindhi text were tested. Results of sentiment analysis of Romanized Sindhi Text were compared with Ground Truth Values. Sentiment Analysis of Romanized Sindhi text was performed on five hundred ninety three sentences and the sentiments of sentences are in the shape of Positive, Neural and Negative. As per sentiment analysis result on Python tool 97% of total sentences have the neutral sentiments, 2% of the total results of sentiments analysis having negative sentiments and only 1% of the total sentences of having the positive sentiments. All the results of sentiment analysis were evaluated On the basis of Ground Truth Values and the 84% of the sentences have the neutral sentiments. Accuracy of the sentiment analysis Was measured by using equation and the parameters were calculated form the confusion matrix. "This confusion matrix was created on the values of Ground truth and accuracy was achieved 87.02%. This research study provides the solution for the problem/issues came in communication (text) and the data given in this research work is useful for computer/mobile users. Also it removes communication (text) gap between peoples of different regions of World on different platforms of communication.
Keywords: Communication, Romanized Sindhi Text, Rule Base Model, Online
Python Tool. Word Labelling, Sentiment Analysis, Ground truth values.