基于BP神经网络的产品性能满意度预测分析

Prediction and Analysis of Product Performance Satisfaction Based on Backpropagation Neural Network

为了解决传统性能设计缺乏复杂场景要素、忽略用户个性化行为等问题,提出一种用户体验评论数据驱动的产品性能满意度预测分析模型.通过分析在线评论数据,获知用户使用产品后的体验满意程度,在此基础上构建由产品结构配置和实际使用工况向性能满意度映射的神经网络模型,分析使用工况作用下的性能满意度影响因素及其影响方式.首先,结合领域本体知识概念和产品使用说明书,整理影响产品性能的相关结构配置要素.之后,以用户感知产品属性后发布的在线评论为数据源,借助自然语言处理技术进行评论内容的细粒度识别,获取产品的实际使用工况,包括使用的环境条件及用户的行为习惯;利用情感分析技术对用户在线评论内容的情感正负倾向进行标定,叠加到用户主观选择的产品评分上,作为用户对该项性能的满意度评分.接着,构建由结构配置要素、使用环境条件和用户行为习惯向性能满意度评分映射的BP神经网络模型,训练后的模型具有较好的预测功能,可以给定新的配置方案,进行不同使用工况下的满意度评分预测.最后,对不同性能结构配置组合方案进行用户满意度预测,分析其中的关键因子及因子间的交互效应.以汽车产品动力性属性为例,进行模型验证,对汽车动力性指标设计和改进提供参考和帮助.

This study proposes a predictive analytical model of product performance satisfaction degree driven by user experience feedback data to solve problems such as the lack of complex environmental elements and overlooking of users’ individual behavior in conventional product performance designs. Online feedback data were analyzed to obtain the user satisfaction degree after using a product. Based on the user satisfaction index,a neural network model was developed by mapping the product structural configurations and actual operating conditions to performance satisfaction degree for analyzing the factors and modes that influence the performance. First,structural configuration elements relevant to product performance were gathered from the concepts of domain ontology and product user manuals. Second,actual use conditions,including practical environmental conditions and individual behaviors,were collected by identifying the fine grit of feedback content using natural language processing with users’ online feedback after perceiving the product properties as the data resource. The satisfaction rating for a specific performance was determined by adding marks of positive or negative emotional inclination on users’ online feedback content using emotion analysis technique for rating products that users have independently selected. Third,a backpropagation(BP)neural network model was built by mapping structural configuration elements,practical environmental conditions,and individual behaviors to performance satisfaction degree. After training,the model depicted good predictive function to provide new configurations and predict satisfaction rating under various use conditions. User satisfaction degree was predicted for various structural configurations to analyze the key influence factors and the effect of the interaction of these factors. The vehicle dynamic quality was considered as the example to test the model,to provide reference and help for the design and improvement of automobile dynamic performance index.