Resumo

The word-of-mouth posted online impacts consumer decision making, an example of this phenomenon is the growing of complaint sites, which concentrate several comments and experiences of the diverse sectors, allowing the consumer a unique place for gathering information and opinions helping the purchase decision. Considering the importance of complaints in the relationship between client and company, it is important to emphasize the importance of a department that knows how to receive and send to both sides the necessary information to solve the problem in a timely manner and avoid recurrence.

This research aims to present the first step to develop an automation tool for a customer complaint mediation department in a Brazilian retail company. In order to do so, we based on Design Science Research (DSR), as it searches through data investigation, collected by standardized research techniques, to develop a tool that can help a specific problem in a business context. In addition, it is also characterized by the use of the concepts of quantitative analysis - the use of quantification both in the information collection and in their treatment using statistical and mathematical techniques.

Considering the connectivity of the 21st century, there are some websites specialized in evaluations and complaints with great adherence of customers, which concentrate a good amount of negative feedback information, allowing companies to hand complaints and turn a negative into a positive opinion. A complaints mediation department is a strategic sector for retailing companies, it obtains information that allows to answer the external questions and explain possible internal bottlenecks, providing suggestions for improvements in operation, enhancing corporation’s results.

After data pre-processing, the first analysis consisted of understanding the lexicon of the database. Being a polarized base, once it was collected from a complaint’s website, a negative reaction is common throughout all the data file, mainly because the data consist of a set of strictly negative feelings, mostly with complaints, so the word "no" was removed from the database, providing a more efficient analysis. We performed the LSA analysis and noticed that less than 50 from the 253 dimensions were responsible for most of the data variability, meaning a high concentration in few categories.

Using computational software that allows the processing, classification and segmentation of 1,175 customer comments from the Brazilian website Reclame Aqui, we develop a semi-supervised classification tool. We identified that most of the categories of analysis were related to delay of delivery. With LSA analysis, we conclude that in all first-25 dimensions, there was the presence of at least one of the top-4 words ("Day," "Product," "Delivery," and "Order"). With all analysis compiled, we concluded that those four words were representative of the main problems reported by the customers.

With the large volume of available data related to customer printing on the various variables that involve a purchase, it becomes increasingly important to have a department that knows how to receive and send the information needed to solve the problem on both sides. in a timely manner and avoid recurrence. In this scenario, this research aims to present the first step to develop a text analytic tool that enables the automation of a complaint mediation sector in a Brazilian retail company that sells home appliances and furniture.

Ashton, T., Evangelopoulos, N., & Prybutok, V. 2014. Extending monitoring methods to textual data. Quality & Quantity, 48,4 Graham, J., & Havlena, W. 2007. Finding the “Missing Link” Advertising’s Impact on Word of Mouth. Journal of Advertising Research, 47,4 Kulkarni, S. S., Apte, U. M., & Evangelopoulos, N. 2014. The use of latent semantic analysis in operations management research. Decision Sciences, 45,5 Weitzl, W. J., & Hutzinger, C. 2019. Rise and fall of complainants’ desires: the role of pre-failure brand commitment and online service recovery satisfaction. Computers in Human Behavior,