Corpus development for brand image and sentiment analysis with social media analytics

From a methodological point of view sentiment dictionaries created for general purposes do not adequately reflect the brand concepts of the brand image study area, the absence of a corpus of sentiments in the framework of brand image in Turkish and the very limited number of studies that reveal brand image perceptions with emotion analysis reveals the gap in this area. In this context, the first aim of the research is to create a Turkish corpus specific to the brand field required for sentiment analysis in order to determine the image perception of brands in the minds of internet users. The second aim is to compare the sentiment classification performance of the brand image corpus developed with the general sentiment dictionaries. In addition, as the third aim of the study, the obtained domain is to reveal the factors that shape the brand image dimension for the two retail brands with the special corpus. In order to achieve the first aim, 113 Turkish theses related to brand image in higher education institution (YÖK) and 150 scientific articles were scanning, a corpus of sentiment containing 14228 sentiment terms was developed using the R programming tool with the corpus approach. The sentiment classification success rate of the Brand Image corpus was obtained as a result of comparing the precision, recall, F-measure and accuracy rates and sentiment classification success rates with the three big general dictionaries SentiWordNet, SentiTürkNet and NRC. The obtained analysis findings showed that the classification performance ratios of the developed corpus were more successful in both positive classification and negative classification compared to general purpose dictionaries. In order to achieve the third purpose, brand image dimensions, which are examined in terms of two retail brands, are discussed on the basis of brand personality. The data set used in the application was extracted from the Twitter platform. 65328 tweet comments of brand X and 117289 tweet comments of brand Y were used in sentiment analysis. According to the analysis findings, it was seen that the X brand was perceived in the "Down to eart" and "cheerful" brand image sub-dimension with the highest value, while the Y brand was perceived in the "Excitement" and "Sincerity" and "Sophistication" brand image dimensions with the highest value.