Research On The Data-driven Requirement Analysis And Mining Methods For Complex Products

Product design and development is the core part of complex product manufacturing,in which the user requirement analysis and confirmation is of great significance.Traditional survey-based requirement analysis methods have many limitations,such as small data size,high data collection cost,strong subjectivity in requirment results and so on.With the development of information technology and mobile Internet,massive online reviews related to product usage experience are published on various Internet platforms,which makes it possible for enterprises to obtain requirements for product design from the Internet in a timely,quick,and automatic manner.However,the Internet data increases exponentially and shows low value density,there remains many problems in practical applications of data-driven requirement analysis and mining technologies.Under such circumstances,a data-driven complex product requirement analysis and mining method containing review spam detection,requirement identification,and requirement evolution analysis and mining is studied in this dissertation.The main contents include:1)A process framework of data-driven complex product requirement analysis and mining is designed.To address the problems and challenges of complex product requirement analysis and mining,online customer review data is used as the requirement data source.Regarding the fuzziness,complexity,personalization and dynamics of user requirements,a process framework that decouples review spam detection,requirement identification,and requirement analysis is designed.It divides the requirement analysis and mining activities into a number of high cohesion and low coupling modules,including review spam detection,requirement information elicitation,requirement classification,requirement ranking,requirement difference identification,critical requirement identification and requirement evolution analysis,providing a reasonable,feasible and applicable top-level framework for data-driven complex products requirement analysis and mining.2)A review spam detection method based on aspect features is proposed.To address the low value density and poor authenticity of online customer revies,a review spam detection model using aspect features is proposed to obtian reliable and high-quality review data for further requirement analysis and mining.First,the product aspect words are extracted from review content via a Bi-LSTM(Bi-directional Long Short-Term Memory)model,and are then classified into different clusters through K-means clustering algorithm.Next,a number of novel aspect features are proposed and fed into the XGBoost(e Xtreme Gradient Boosting)algorithm for identifying spam reviews.Finally,the effectiveness of the review spam detection model is evaluated on open labelled review datasets.3)A requirement identification method for complex products based on“Elicitation-Classification-Ranking” is proposed.Previous requirement identification results usually show limitations in covering a large number of product features and reflecting the hierarchical relationships between complex product features.To address that problem,in this dissertation,the requirement identification task is divided into three sub tasks,i.e.,requirement information elicitation,requirement classification and requirement ranking.First,a requirement information elicitation model using Bi-LSTM is proposed to automatically extract feature words,negative words,and opinion words.Then,a requirement classification model using hierarchical clustering algorithm is proposed to classify the extracted feature words into different clusters with a hierarchical tree.Finally,an automatic requirement ranking model using the weight function on user satisfaction index is proposed to improve the objectivity and automation level of requirement ranking.4)A “Static-Dynamic” combined requirement analysis and mining method for complex products is designed.Considering that previous studies have rarely analyzed the requirement evolution process and thus are difficult to support the future requirement evolution prediction,a “Static-Dynamic” combined requirement analysis and mining method is designed in this dissertation.Regarding static requirements,the requirement difference identification model based on statistical tests,the data-driven static SA-Kano model,and the critical requirement entity identification model based on heterogeneous information network are proposed to realize multi-angle user requirement analysis.Regarding dynamic requirements,the user requirement evolution analysis model,the data-driven dynamic SA-Kano model,and the requirement evolution law clustering model based on similarity measure and community discovery are proposed to realize the multi-dimensional user requirement evolution law mining.5)Taking new energy vehicle as an example,the application research of datadriven complex product requirement analysis and mining is carried out.Taking a new energy vehicle as an example,the relevant online review data is obtained using web crawlers.Data preprocessing is conducted to screen out reliable and high-quality review data,after which the model validation and case study of the requirement identification method based on “Elicitation-Classification-Ranking”,and the “Static-Dyanmic”combined requirement analysis and mining method are carried out.