Identification,Analysis And Monitoring Of Product Defects Based On Social Media Data

Product defects break the word-of-the-mouth of manufacturers.Safety defects may bring untold economic loss and threat lives of customers.Unsafety defects influence customer satisfactions and purchase intentions in a great degree.They may also induce lots of costs caused by customer claims.With the development of social media,data generated on social media platforms become an important information source for firms.For the advantages of comprehensiveness,promptness and authenticity,social media data have been used to detect product defects.But social media data are usually formatted in texts.They have enormous volumes and are generated at an incredible speed.These features all propose a daunting challenge for firms to ingest defect information from social media data.Recently,most related research focuses on discovering defect-related texts from social media data.Some researchers collect defect information from texts but they omit the exclusion of defectunrelated texts.Thus,their obtained information is inaccurate.Besides,social media data are generated every day.How to mine defect information in a long-term becomes a key problem.But there is no comprehensive research to date.Thus,this dissertation attempts to provide a methodology for identifying,analyzing and monitoring the defect information based on social media data.With methods in this dissertation,firms can derive product defect information accurately and promptly.The first part of the dissertation provides an important opinion that defect-related text needs to be discovered from the whole corpus before collecting defect information when identifying defect discussed by customers.The omission of this finding in extant studies induces biased results.To solve this deficiency,this dissertation develops two approaches.One is Product Defect-Related and Defect-Unrelated Information Derivation Model(PDR-DURIDM)which can distinguish between defect-related and defect-unrelated texts.Another method is the integrated method of Bidirectional Encoder Representations from Transformers and Product Defect Discovery and Information Derivation Model(BERT-PDDIDM).Experimental results show that proposed methods are more effective in defect information derivation than classical model.PDR-DURIDM deals with defect-related and defect-unrelated texts simultaneously while BERT-PDDIDM owns outperformance in defect-related text discovery.Their collected defect information is more accurate and abundant that the information extracted by extant models.The second part improves the classical quality management tools to analyze the defect information extracted from social media data.The improved approaches analyze defect information and evaluate the importance of product defects,defective components and defect causes.Using a case study,the effectiveness of the developed methods in the importance assessment has been proved.The improved methods use the derived information from social media data as inputs and alleviate the subjectivity brought by expert evaluations.The importance of defect information will provide more abundant managerial insights for reasonable remedial decision-making.To deal with numerous texts more effectively and automatically and monitor defect information in social media data,the third part develops a defect knowledge graph(DKG)and a defect early warning system(DEWS)based on the DKG.DKG transfers social media texts into defect information.It identifies the defects that never happened before with the help of expert systems.Besides,the supplement of the product component knowledge graph also solves the problem of coarse-grained descriptions of defective components.Experimental results validate that DKG addresses social media data in the tasks of defect management more effectively and accurately.Based on DKG,the developed DEWS achieves the real-time monitoring of text contents and prediction of the number of defect-related texts.DEWS has been applied in the case study of automobiles successfully.With DEWS,firms will react to defects more promptly.This dissertation proposes accurate and effective methods to identify and extract defect information from social media data.Meanwhile,the importance evaluation method of defect information is also researched by improving classical quality management tools.These developed approaches overcome the deficiency of extant studies which only concentrate on defect-related text identification or defect information extraction and enrich extracted defect information.Furthermore,the proposed DKG and DEWS identify,derive,analyze and monitor defect information hidden in social media data.This dissertation enriches the research of defect management and helps firms perform product defect management more accurately and promptly.It provides more reliable and comprehensive managerial insights for decision-making and thus shows the great potential practical value in product defect management.