Research On Sentiment Analysis Of Telecom Operators' Customer Service Comments

As a new platform for rapid communication between operators and users,Weibo's large-scale user base generates a huge amount of information every day,and summarizes,analyzes and summarizes the sentiment orientation behind these information,which can help operators achieve publicity.Important functions such as services and marketing.Traditional sentiment analysis related research includes two types of dictionary-based and machine-based learning methods,but the establishment,maintenance and update of sentimental dictionary requires a lot of manpower,while the machine learning-based method relies on artificial feature engineering and hidden sentimental information in the data sequence.Since the rise of deep learning,its strong expressive ability and the advantages of no need for manual feature selection and construction have achieved outstanding performance in sentiment analysis.In this paper,the sentiment analysis of two aspects based on machine learning and deep learning are carried out on the operator weibo commentary text data.The specific research work is as follows:(1)Climbing the operator's Weibo comment text to establish 15000 data in the corpus of this research,and carry out text preprocessing and word vector training to obtain data input of machine learning model and deep learning model;(2)In the machine-based learning sentiment research,the traditional TF-IDF and the neural network model-based Word2 vec are used to extract the features,so as to construct the experimental data as the input of the machine learning model SVM,Na?ve Bayes,Logistic Regression.And conduct sentimental analysis experimental design and results;(3)For the limitations of the machine learning model,the deep learning model is introduced into the sentiment analysis.Since the BiLSTM can obtain the dimensional controllable feature vector containing the global semantic information,the influence of the sequence information on the output is reduced,and the introduction of the Attention mechanism can pay more attention to the probability distribution value of the important word pair output.Redundancy problem,so this paper improves a BiLSTM model that integrates the Attention mechanism to solve the sentiment classification problem.Through the experimental design and result analysis on different types of data sets,the effectiveness of the deep learning model in text sentiment analysis is obtained,and the improved model achieves the best classification results.The paper has 48 figures,15 tables and 61 references.