A combined deep learning method for internet car evaluation

Deming Li,Menggang Li,Gang Han,Ting Li

DOI: https://doi.org/10.1007/s00521-020-05291-x

2020-08-27

Neural Computing and Applications

Abstract: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.

computer science, artificial intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in the Internet car evaluation system, due to the small number of abnormal comment results in the sample processing process, it is difficult to cover the entire comment sample distribution. To overcome this challenge, the author proposes a comprehensive method based on topic feature extraction, long - short - term memory network (LSTM) and deep convolutional generative adversarial network (DCGAN), aiming to achieve effective evaluation of Internet car development by using a large number of normal comment event samples for training.

Specifically, the paper proposes the following points to solve the above problems:

1. **Topic Feature Extraction**: Use the LDA model to extract topic features from user comments. These features can represent data more regularly for use in subsequent stages of input.

2. **LSTM Sentiment Analysis**: Train and predict the sentiment tendency of input samples through the LSTM model. LSTM can capture long - term dependencies in sequential data and is suitable for text sentiment classification tasks.

3. **DCGAN Abnormal Sample Processing**: Considering that there are few abnormal comment results in the sample processing process and it is difficult to cover the entire comment sample distribution, the paper introduces DCGAN. DCGAN can achieve the evaluation of Internet car development by only using a large number of normal comment event samples for training without using too many abnormal samples. DCGAN consists of a generator (G) and a discriminator (D). The generator is used to generate new data samples, while the discriminator is used to distinguish between real data and generated samples.

Through the above methods, the paper aims to provide a systematic Internet car evaluation system, which not only provides references for consumers to buy cars, but also provides a basis for car companies to optimize their products.