E-commerce Logistics Evaluation Sentiment Analysis

With the development of e-commerce,online shopping has gradually become our main consumption method,which has spawned a large number of logistics industries,and the quality of logistics services directly affects customers' satisfaction with online shopping.Therefore,mining and analyzing the logistics review data can not only help the merchants to better understand the logistics situation,select the appropriate cooperative manufacturers,but also provide a reference for the logistics industry to improve services.Therefore,sentiment analysis of logistics reviews has important research significance and practical value.This article conducts sentiment analysis on consumer reviews in the field of logistics in the evaluation of e-commerce,and investigates consumer attitudes to logistics services.The main tasks are as follows:1.Extract keywords with high frequency in logistics to construct a keyword library,filter out words with high frequency and emotional tendency from the logistics review data set,and select positive and negative words as seed words from the basic dictionary.The SO-PMI algorithm calculates the difference in point mutual information between the words to be judged and the seed words,and according to the size of the result,judges whether to add it to the keyword library for expansion.2.Improve the TF-IDF weighting according to the keywords matched by the regular expression,improve the accuracy of the emotion classification in logistics,and use the improved TF-IDF to weight the word vectors and add all the word vectors.Then get the sentence vector of the entire text.3.This article collects logistics corpus,performs deduplication,cleaning,word segmentation,introduces stop words,adds weighted word vectors to the LSTM model,and builds a logistics evaluation model based on LSTM.Experiments show that this model has a better classification effect on text,and overcomes the shortcomings of certain limitations based on sentiment dictionary and machine learning as the data volume increases.4.Through experimental comparison,under the same experimental conditions,the improved TF-IDF classification effect is better than the standard model;The weighted Word2 vec accuracy is also significantly improved compared to the original model;The combined model of LSTM and weighted word vector makes up for thedeficiency of a single model,and the classification results are ideal.