Research On Mobile Online Reviews Of Pinduoduo Based On Prompt Learning And LDA

As mobile phones gradually become a necessity in life,the demand for mobile phones has been rising.Online shopping,with its advantages of better prices and the ability to shop anywhere,has attracted many people to buy mobile phones online,resulting in a large number of online reviews of mobile phones,which contain a lot of useful information.At the same time,Pinduoduo launched the "10 billion subsidy" campaign in 2019,which provided "quality insurance" and platform subsidies for goods in the "10 billion subsidy" zone,thus attracting many people to buy mobile phones on the platform.However,many people still have doubts about the quality of mobile phones on the platform due to the "fake goods" label.Therefore,it is of practical significance to study the online reviews of mobile phones on the Pinduoduo platform.By mining these reviews,we can understand consumers’ shopping feedback,so as to provide reference suggestions for consumers’ purchase decisions,Pinduoduo platform’s self-optimization and mobile phone manufacturers’ product upgrades.Based on this,this thesis conducts sentiment analysis and LDA thematic analysis of mobile online reviews on Pinduoduo platform.The content of mobile online reviews often involves multiple aspects,and it would be inaccurate to analyse only the overall sentiment of the review text,so this thesis performs aspect category sentiment classification for mobile online reviews obtained from the Pinduoduo platform.In order to save the cost of annotating sample data,we used the prompt learning method that performs well on zero-shot and few-shot tasks.The experimental results show that the prompt learning model achieved an accuracy of72.46% for zero samples,83.53% for a training set size of 1024,and it is better than the benchmark models AE-Bi LSTM and ATAE-Bi LSTM under any prompt template and training set size.Moreover,through cross-aspect and cross-domain migration experiments,it is shown that the prompt learning model has good generalization performance in new aspects and new domains.The trained prompt learning model was finally used to classify the unlabeled review text for sentiment,and the LDA topic clustering was performed on the classified positive and negative reviews respectively.Based on the analysis of the classification and clustering results,we found that consumers can buy mobile phones with guaranteed quality at super cost-effective in the "10 billion subsidy" zone of the Pinduoduo platform;however,there is still room for improvement in the customer service and delivery speed of the Pinduoduo platform.Based on the results of the analysis,we also provided some reasonable suggestions for mobile phone manufacturers and the Pinduoduo platform.