Tatiana da Silva Gama’s scientific contributions

Product reviews are valuable resources that assist shoppers in making informed transactions by reducing uncertainty within the purchase process. However, user-generated content is not always secure or adequate. The goal of customer review moderation is to ensure both a secure environment for all parties participating and the integrity of the review information. Content moderation is a difficult task even for human moderators, and in some circumstances, due to the enormous volume of reviews, manual content moderation is not practical. In this paper , we present the experiments carried out using automated machine learning (AutoML) for moderating product reviews on one of Brazil's largest e-commerce platforms. Our machine learning-based solution is faster and more accurate than the previously used content moderation system, performed by a third-party company system dependent on human intervention. Overall, the results showed that our model was 31.12% more accurate than the third-party company system and it had a fast development due to the use of AutoML techniques.

We introduce RePro, a corpus of e-commerce product reviews in Brazilian Portuguese labeled with sentiment and topic information. We carried out a careful annotation process, whose aim is to introduce an easily available and open benchmark for opinion mining related tasks, namely sentiment analysis and topic modeling tasks. This work describes the corpus design and annotation process as well as the preliminary results of classification tasks. These preliminary results can be used as baselines for future work. RePro contains 10,000 humanly annotated reviews, based on data from the largest Brazilian e-commerce platform, which produced the B2W-Reviews01 dataset.

Clarifying doubts can become decisive when shopping on e-commerces platforms. Considering the relevance of user generated content, this work aimed to develop an internal hybrid system, composed of machine learning models along-side a rule-based module, to moderate customers’ questions and sellers’ answers in one of the biggest marketplaces in Brazil.