| China's mobile phone market has huge capacity and broad development space,but at the same time,market competition is becoming increasingly fierce,and smartphone shipments continue to decline.Major mobile phone manufacturers urgently need to find breakthroughs through user's specific needs to improve product quality.With the development of the Internet,more and more consumers choose to buy mobile phones online,and share satisfaction and complaints about mobile phone features and platform services through online platforms,resulting in online reviews of mobile phones.The main purpose of this article is to conduct sentiment analysis on mobile online reviews,divide online reviews into different categories according to their emotional tendencies,and analyze the specific concerns of users in different types of reviews,to improve product quality and users for mobile phone manufacturers and e-commerce platforms Experience provides the basis for decision-making.The main work of this article is summarized as follows:1.Acquisition and preprocessing of Jingdong Mall mobile online comment information data.Using python crawler technology to obtain about 260,000 pieces of online comment information of Jingdong Mall mobile phones,after preprocessing data such as deduplication,mechanical compression,and short sentence deletion of the original data,we obtained about 240,000 pieces of online comment information of mobile phones.Then stratified sampling method was used to extract 10,000 pieces of comment data for labeling.After word segmentation,feature item selection,dimensionality reduction and weight calculation,etc.,a sparse matrix for establishing a classification model was obtained.2.Establish a simple Bayesian model of sentiment classification for mobile online reviews.Randomly select 70% of the labeled samples as the training set to train the learning model of naive Bayesian emotion classification,and determine the value of the parameter alpha through the learning curve of the naive Bayesian Laplace smoothing coefficient.When alpha is 0.05,the classification The effect is the best,and the accuracy rate of the Naive Bayes model to predict the emotion category is88.6%.3.Build an SVM model for sentiment classification of mobile online reviews.By trying different kernel functions,gamma parameters and penalty coefficient C,training the classification learning model of SVM,and using the grid search method to optimize the SVM model,the accuracy of the SVM model prediction after gridsearch optimization is 89.8%.4.LDA theme analysis of online reviews of mobile phones in different emotion categories.All mobile phone online reviews are classified,and the LDA theme analysis is carried out according to positive reviews and negative reviews,and the user's focus on mobile phone features is mainly reflected in the photo effect,running speed,screen and resolution,battery,standby time,appearance,cost performance,etc.At the same time,it is concluded that the user's focus on the e-commerce platform and the products and services of the merchants is mainly reflected in the aspects of logistics,gifts,price protection,after-sales,etc.In this paper,the Naive Bayes model and the SVM model are applied to the sentiment classification of mobile online reviews,and the model is continuously optimized by selecting reasonable parameters to improve the accuracy of prediction,which has a strong theoretical value.According to the research conclusion and the actual situation,this article gives reasonable suggestions to the mobile phone manufacturers from four aspects of brand innovation,battery technology improvement,system research and development,and after-sales maintenance,and proposes to the e-commerce platform from the three aspects of warehousing logistics,price protection mechanism,packaging and gifts.Reasonable suggestions have strong practical value. |
PDF Full Text Request