Research On User Pain Points In Social Commerce

With the development of social commerce,people usually share some shopping reviews after shopping on the e-commerce platform.These reviews contain consumers' views,attitudes and emotions about the goods,and can have an important impact on other consumers' purchase decisions.They are also an important basis for producers and businesses to obtain users' needs and improve product design or sales strategies.This thesis uses deep learning method,using large-scale review data for fine-grained emotional analysis to obtain user pain points,aiming to help businesses find user needs and achieve the purpose of precision marketing.In order to obtain fine-grained sentiment research user pain points,we uses two attribute level sentiment analysis methods,in order to obtain more comprehensive user pain points.The main research contents of the two methods are as follows:The first method uses attribute level sentiment analysis method of joint extraction to study user pain points.The selling point is defined as the attribute class in attribute level sentiment analysis,and the pre training model Bert is used to extract the embedded features of sentences and attribute words.At the same time,the shared layer is used to obtain the interactive information of attribute words and sentences,which improves the problem that the memory network only pays attention to a single semantic and ignores the semantic relationship between sentence sequences.Experiments were carried out on Semeval 2014's lap top and restaurant,and the accuracy of the model reached 80.25% and 84.82%respectively.In addition,the Chinese camera review data from semeval 2016 of Harbin Institute of technology is used for training to verify the effect of the fine-tuning model on the Chinese data set.Finally,we use the results of sentiment analysis to quantify user satisfaction and study the relationship between user pain points and selling points.The second method uses the attribute level sentiment analysis method of step-by-step pipeline to study user pain points.First,feature word pairs are extracted by sentence structure and inter word relationship.Then,sentiment analysis is carried out by using the pre training model of fusion feature word pairs.Finally,feature attributes and sentiment analysis results are combined to quantify user pain points.This method makes up for the low efficiency of the joint extraction method.The experimental results show that the accuracy and F1 value of sentiment analysis are 95.33% and 95.23% respectively.Using the comment data of tablet computer to analyze users' pain points,the results show that this method can effectively obtain users' pain points.In summary,this article uses two attribute-level sentiment analysis methods to analyze user pain points.The method of joint extraction is the current hot spot of attribute-level sentiment analysis research,which improves the accuracy of attribute-level sentiment analysis,but the model training is difficult and the training time is long.In order to improve the efficiency of user pain point research,the research also uses a step-by-step pipeline method to study user pain points.Finally,the results of sentiment analysis are used to quantify user pain points and help merchants with precision marketing.