A combined deep learning method for internet car evaluation

In recent years, the Internet has become a trend in the development of the global automotive industry. Numerous Internet companies have joined the automobile manufacturing industry. At the same time, people generally search for information about cars on the Internet as an important reference to purchase decisions before buying them. As a high-value commodity, almost all consumers use search engines to find out the price, reputation and other information about their favorite models before they buy. On the other hand, online reviews contain a large amount of information about what consumers are saying about products, and they influence the purchasing decisions of potential consumers. It is observed that current reviews of automobiles can include several dimensions: corporate brand attention, corporate development and user reputation. In order to provide reference for users and car manufacturers, this paper established a systematic model of Internet car evaluation system based on topic feature extraction, the long short-term memory (LSTM) and the deep convolutional generative adversarial networks (DCGAN). Firstly, the model uses feature extraction and LSTM for sentiment analysis of user evaluations; secondly, considering anomalies in the sample processing, which makes it difficult to cover the distribution of the entire review sample, we proposed a way to train without using too many anomalous samples using DCGAN. The results show that this method can achieve an effective systematic evaluation of Internet cars using only a large sample of normal review events. The results can be used as a reference for people to buy a car and for car companies to optimize their products.