基于在线评论数据的产品需求趋势挖掘

随着经济水平的提高和物质生活的丰富，消费者的需求变化也越来越快。能否迎合市场需求的变化是企业产品成功的关键。随着社交媒体的发展，消费者为了分享购物体验发表了许多在线评论信息，其中蕴含着消费者的需求变化。本文在产品特征提取和属性情感分析的基础上，构建了垃圾评论识别模型。然后，利用时间序列分析模型预测下阶段的产品属性关注度和情感计算。最后结合历史数据的变化趋势，分析产品属性的重要性和市场满意情况。利用汽车论坛上的汽车评论数据对本文提出的研究模型进行了验证。研究结果可以为企业制定营销策略以及产品改进与创新提供决策支持。

Trend Mining of Product Requirements from Online Reviews

Acquiring and meeting customer needs is the primary concern for product development. The changing market environment and the increasing people's consumption level make customer requirement information become personalized, fragmented and versatile. The product requirements acquired by traditional methods are not objective, timely and comprehensive, and are insufficient to support a customer-centric product development strategy. With the development of social media technology, consumers have published a lot of online reviews in order to share their shopping experience, which contains consumers' requirements for products. How to get product features and their emotional polarity from online reviews? How to identify spam comments in online reviews? How to acquire customer requirements for products using useful reviews? These are some new problems worth studying.In this paper, the process of acquiring product requirements from online reviews in social media environment is studied. Attribute recognition and sentiment analysis methods together with spam comments recognition model based on support vector machine and product requirement trend mining model based on time series analysis are presented. Firstly, unsupervised extraction technique based on crowdsourcing is built to identify product attributes from online reviews. The product attribute sentiment dictionary is constructed, and the adjacent-based method is used to determine the emotional polarity of the product attribute. Then, the product feature extraction is carried out based on the information quality and information gain respectively. For the binary classification problem of spam comments recognition, a spam comments recognition model based on support vector machine is proposed. Next, the non-parametric exponential smoothing model Holt-Winters is used to examine the requirement trend for the products in the next stage, and the Mann-Kendall test method is used to detect the trend of attention and positive and negative emotional changes of each attribute. Finally, the validity of the research model is verified by the review data on the automobile forum, and three automobile product attributes are selected to analyze the importance of product attributes and market satisfaction.The results show that the method of attribute recognition and sentiment analysis constructed in this paper can effectively identify product attributes and judge the emotional polarity of product attributes. The established spam comments recognition model can effectively eliminate spam comments. The proposed time series analysis model can predict the customer requirements for product. These results are useful to provide decision support for companies to undertake marketing strategies and product improvement and innovation.