Study On User-Generated-Content Based Product Knowledge Discovery Method

Product knowledge is benefit for enterprise to create and keep advantages.Recently,in rapidly developed social media platform,customer posted large amount of User-Generated Content(UGC)which involves usage experience and personal opinion about product.These kinds of information contain a lot of product knowledge which is valuable for product defect management,product marketing and product development.Unfortunately,the characteristics of UGC,such as volume,variety and unstructured,bring difficulty and challenge to analyze UGC and discover product knowledge.This dissertation focus on three kinds of product knowledge in social media,namely product defect knowledge,product competitiveness and lead user requirement knowledge.Then we design corresponding knowledge discovery framework and study knowledge discovery methods.In the discovery of product knowledge discussed above,sentiment analysis is the key technique of judge and measure customer attitude and plays essential role in product knowledge discovery.Therefore,we need to do further research of domain-specific sentiment analysis before studying product knowledge discovery methods.The specific research content has been shown as follow.(1)Domain-specific sentiment lexicon generation methodTo generate domain-specifc sentiment lexicon,domain-specific sentiment lexicon generation method has been studied through integrating PMI and word embedding information based on the princinple that the similar words have high co-occurrence probability.Using domain-specific sentiment lexicon labeled by experts,PMI and word embedding information via linear regression to judge the sentiment orientation.The domain-specific sentiment lexicon can be generated after iterative operation.In experiment,the effectiveness of method proposed has been proved by comparing with general sentiment lexicon.(2)UGC-based product defect knowledge discoveryTo solve the problems in product defect knowledge on social media,UGC-based product defect knowledge discovery framework has been designed,contextual feature and multi-view learning-based product defect UGC identification method and word embedding-based product attribute description normalization method has been studied.First,analyzing the characteristics of origin UGC and reply on style,social network and so on.Then extracting corresponding UGC feature,especially contextual feature,and analyzing its effectiveness.In product defect identification model,multi-view learning has been employed to solve UGC high-dimension problem.To improve performance of multi-view learning,studying the view creation method based on nature attribute and feature combination.Second,studying the product attribute description normalization method due to various description of product attribute.Finally,summarizing product defect attribute with high frequency based on clustering analysis to generate product defect knowledge.The experiment results based on large amount of true social media data proved the feasibility and effectiveness.(3)UGC-based product competitiveness knowledge discoveryTo solve the problems in product competitiveness knowledge discovery on social media,UGC-based product competitiveness knowledge discovery framework has been designed,competitor identification method based on product comparison UGC and competitiveness analysis based on customer satisfaction have been studied.First,Product and brand feature,linguistic feature,keyword and reply feature have been extracted based on product comparison UGC characteristics.Additionally,imbalanced data classification algorithms have been employed to construct product comparison UGC identification model due to the sparseness of product comparison UGC.The competitor can be extracted from identified product comparison UGC.Finally,analyzing customer satisfaction,unsatisfaction and integrated satisfaction based on domain-specific sentiment analysis,and discover the advantage and disadvantage based on comparison analysis.The experiment results based on large amount of true social media data prove the feasibility and effectiveness.(4)UGC-based Product innovation knowledge discoveryTo solve the problems in product innovation knowledge discovery,lear user-based knowledge discovery framework has been designed,SamplingBagging-based lead user requirement UGC identification method and text summarization-based lead user requirement summary generation method have been studied.First,analyzing the characteristic of lead user requirement UGC and extracting its feature.Lead user requirement UGC identification is extremely imbalanced data classification task,which aggravates the disadvantages of existing classification methods and reduces performance.In this dissertation,OverSampling and UnderSampling are combined using Bagging to overcome the problems above.Finally,text summarization has been used to analyze lead user requirement UGC,which can provide enterprise brief lead user requirement UGC summarization information and help them get lead user requirement topic rapidly.The experiment results based on large amount of true social media data prove the feasibility and effectiveness.UGC-based product knowledge is valuable in decision making of product defect management,product marketing and product innovation and development.The discovery methods of three kinds of product knowledge and domain-specifc sentiment lexicon generation method have been studied respectively,which solves the key scientific problems of product knowledge discovery on social media.