Abstract
The growing usage of social media as a platform to interact and source for information have enriched and lighten analytics for generating insight and improving business value. Recently, companies direct their attention to mining useful information from the big social data of their customers to understand their customers better and maximise profit. However there is need to utilise the right approach to avoid waste of resources and time.
Sentiment analysis (SA) is a developing area of research; the technique is useful in classifying words or phrases into polarity such as positive or negative, good or bad. Organisations employ SA techniques to understand the opinion or intentions of their customers toward a subject such as product or service. Industries like airline, retail, transportation and hospitality use SA to drive their business, determine brand image, monitor stock market, discover trend and manage crises. However, the natural language, ambiguity and colloquial nature of text made sentiment analytics complicated. Nigeria has her official language as English but due to over 400 languages spoken across the country Pidgin English became popular. Nigeria social media users communicate in Pidgin English and/or Standard English that makes the analytics challenging.
This study aims to compare the performance of three different sentiment classification lexicons AFINN, BING and NRC by applying them to twitter data of Nigeria bank customers. The bank customer live tweets were extracted for duration of one (1) month using python programming language and analysed in R programming language. Their opinion, sentiment and attitude were mined towards their ATM service experience. The study compared results of the lexicons against manually classified tweets selected randomly. Thus, reported the lexicon performances in terms of precision, recall and f1 score. In general, the three lexicons did not perform greatly especially in terms of sarcasm and Pidgin English classification however AFINN performed better than others. The study provides beginners with background knowledge of lexicon based approach to sentiment classification and also identifies the need for creation of Pidgin English sentiment lexicon or improve existing ones.
Sentiment analysis (SA) is a developing area of research; the technique is useful in classifying words or phrases into polarity such as positive or negative, good or bad. Organisations employ SA techniques to understand the opinion or intentions of their customers toward a subject such as product or service. Industries like airline, retail, transportation and hospitality use SA to drive their business, determine brand image, monitor stock market, discover trend and manage crises. However, the natural language, ambiguity and colloquial nature of text made sentiment analytics complicated. Nigeria has her official language as English but due to over 400 languages spoken across the country Pidgin English became popular. Nigeria social media users communicate in Pidgin English and/or Standard English that makes the analytics challenging.
This study aims to compare the performance of three different sentiment classification lexicons AFINN, BING and NRC by applying them to twitter data of Nigeria bank customers. The bank customer live tweets were extracted for duration of one (1) month using python programming language and analysed in R programming language. Their opinion, sentiment and attitude were mined towards their ATM service experience. The study compared results of the lexicons against manually classified tweets selected randomly. Thus, reported the lexicon performances in terms of precision, recall and f1 score. In general, the three lexicons did not perform greatly especially in terms of sarcasm and Pidgin English classification however AFINN performed better than others. The study provides beginners with background knowledge of lexicon based approach to sentiment classification and also identifies the need for creation of Pidgin English sentiment lexicon or improve existing ones.
Original language | English |
---|---|
Publication status | Published - 1 May 2019 |