基于语义量化匹配的汽车外观显著性意象挖掘

目的 为解决传统感性设计研究中意象实验耗时大以及小样本偶然性等问题，依托现有网络评价文本信息提取了用户意象认知。方法 首先，爬取大规模汽车外观评论文本，构建语义分析词汇库，构建word2vec词向量模型;然后，基于模型获取词库内部的语义联系，计算高频关键形容词之间的语义离散性，以构建代表性意象词空间;最后，通过语义量化匹配将评论映射到意象词空间，得到大规模用户对各车型的显著性意象表征，明确了指定意象词汇下的汽车外观匹配结果。结果 运用该方法挖掘汽车外观显著性意象与基于人工评价的实验结果无显著性差异且具有高度相关性，证明了该方法的有效性。结论 以该方法挖掘用户意象认知，运用了现有的大批量用户反馈知识，提高了意象分析效率，有助于决策者快速理解消费者对汽车外观的感性知识，在设计迭代中可使产品更符合市场期望;对比相关研究，基于语义量化匹配的方式无需对超高维向量进行降维和聚类，避免了以往研究因特征降维而可能导致的词向量语义联系的损失，以得到更为准确的意象挖掘结果。

Mining of Salient Image of Automobile Appearance Based on Semantic Quantification Matching

The work aims to solve the problems of time-consuming image experiments and accidental small samples in traditional perceptual design research by extracting user image cognition based on existing network evaluation text information. Firstly, a large number of automobile appearance comment texts were crawled to construct a semantic analysis vocabulary library and a word2vec word vector model. Then, based on the model, the semantic connections within the word library were obtained to calculate the semantic discreteness between high-frequency key adjectives to construct a representative image word space. Finally, through semantic quantification matching, the comments were mapped to the image word space to obtain significant image representations of various automobile models from a large number of users, and the matching results of automobile appearance under specified image words were clarified. The application of this method to mine significant automobile appearance images showed no significant difference compared with experimental results based on manual evaluation, and had high correlation, demonstrating the effectiveness of this method. This method is used to explore user image cognition through a large number of user feedback knowledge, which improves the efficiency of image analysis, helps decision makers quickly understand consumers'' perceptual knowledge of automobile appearance, and can make products more in line with market expectations during the design iteration. Compared with related research, the method based on semantic quantification matching does not need to reduce the dimensionality and clustering of high-dimensional vectors, which eliminates the loss of semantic connections between word vectors that may be caused by feature reduction in previous studies, so as to obtain more accurate results of image mining.