The Automobile Quality Problem Discovery Method Research In The Context Of Chinese Social Media

The continuous development of the domestic auto industry and the continuous improvement of relevant laws and regulations automotive environment put forward higher requirements to the level of quality automotive product. At the same time, automotive product quality problems arise in many ways will be a huge loss to the enterprise. With the popularity of Web2.0, there has been a lot of user generated content about the quality of automotive products published on social media. And the social media provides automobile manufacturer a new way to get feedback about product quality from users, but existing research on social media has little involved in social media product quality problems discovery, especially lack of research in the context of Chinese social media. Besides, there still challenges in text classification and topic clustering of automotive quality problem discovery in the context of Chinese social media. To address these issues, this thesis mainly finished the following work:(1) Analyzed the characteristic of text feature set and social media, and construct a framework for automotive quality problem discovery in the context Chinese social media.(2)Construct the automotive product quality problems oriented text feature set based on existing common text features and characteristics of Chinese automotive social media. On the basis of the improved algorithm Tri-training on collaborative training, this thesis design Chinese social media automotive quality problem classification model based on semi-supervised learning.(3) For existing deficiencies in text classification or text clustering algorithm to identify and organize those posts, proposed the modeling method of automotive product quality problems, as well as topic clustering methods.(4) Designed with quality real events occurring background experiments, the Chinese social media automotive quality problem classification model and the topic clustering model is validated classification results in a small amount of labeled data.At the practical application level, the research of this paper can help automotive companies perform quick decision-making based on automotive quality problem. At the theoretical level, this paper expand the research scope of social media mining and could be the foundation of further research like product innovation demand founding.