Mining Sentiment-Dependent Linguistic Patterns from Automotive Reviews for Product Defects
Bin Wang, Guilei Zhu, Zhu Zeng · 2022
Due to the universality and rapidity of information dissemination on social media, it is of guiding significance for automobile manufacturers to improve product design and optimize quality management to timely discover the defect information of automobiles from social media. At present, the research on social media defect recognition has mined less defect information and mostly takes negative comments as product defects. To solve this problem, we put forward a comment representation model based on sentiment-dependent linguistic features, which effectively uses the domain context. In reality, the distribution of the data set is biased in some way. To avoid the major defect, we use the clustering-based under-sampling method. The experimental results show that the model can effectively identify car defects in Chinese social media, and has a high accuracy and recall rate.