Abstract:
Gas customer service hotline is a service channel to the majority of gas users who encounter various problems in the use of gas. By means of the channel via telephone, the gas users can communicate and reflect the difficulties encountered, during which the users' words of expressing the problems will reveal their psychological state of mind. Especially, when the user's attitude is unfriendly, the operator need to promptly appease the users' emotions and make their problem satisfactorily settled. Besides, gas customer service hotline can also conduct after-sales visit for the users with unfriendly attitude and confirm whether the problem is resolved, so that these users feel the product after-sales service and continue to ease the use of the product. However, operators are usually busy at certain times such that they may not record the users who need to be conducted after-sales visit. Fortunately, the gas customer service hotline can record the voice of each user communicating with the operator, so that we can convert the speech into texts to analyze the emotions expressed by the users in the texts, and then feedback the emotional state of each user to the operator for conducting after-sales visit. By using TF-IDF based mean Word2vec model and supervised machine learning algorithm, this paper proposes a sentiment analysis method on the chinese texts in the gas customer service hotline. Firstly, Word2vec model is used to train the word vector of each word in the text and TF-IDF algorithm is used to calculate the weight of each word, by which the word vectors are weighted. Secondly, the corresponding dimension of the weighted vector of all the words in the text are accumulated, whose averaged value are taken as the vector of the text, i.e., the characteristics of the text. Finally, these characteristics are trained and predicted by using the supervised machine learning method to achieve the sentiment analysis of the text. The experimental results show that the proposed method can attain high classification accuracy and effectively carry out sentiment analysis.