Abstract: Online review, which is access to the public, contains a variety of valuable information about product quality. Analyzing and mining of these information is very helpful for the enterprises to understand customer needs and preferences. Currently, the customer preference model based on online reviews considers the number and sentiment of the reviews, but ignores the impact of product parameters on consumers' purchasing decisions. Therefore, a customer preferences model that fuses feature sentiment and product parameters is proposed. First, the product features discussed in the online reviews must be extracted, and the sentiment polarities of the product features should be identified, thereby generating the feature sentiment by measuring the score of positive and negative sentiment. Then, an econometric model combining product parameters provided by the enterprises is built to reflect the influence of the feature sentiment and product parameters on product sales. Finally, in order to verify the effectiveness of our model, an online review text of 39 automotive products (SUV models) from autohome.com has been obtained, and sales tracking has been kept for 8 months. The results show that our model has better performance than the methods of information gain and TF-IDF. It is found that most product features and some product parameters have significant impacts on product sales. Besides, product parameters can enhance customers' perceptions on product quality. In addition, our conclusions explain the impacts of feature sentiment and product parameters on product sales, which provide a theoretical basis for marketing strategy and making product design decisions.

Key words: online reviews, customer preferences, feature sentiment, product parameters