| In recent years,the development of network technology has brought great convenience to real life.Different from the past in which people have to book a hotel room in hotel,now people can directly check the location and information of the hotel in the mobile phone software,and then complete the hotel reservation.However,the ensuing of customer experience is also very important.Large-scale network reservation has brought massive text data,these data are tendentious,which is crucial for the hotel and can directly reflect the hotel’s problems.The hotel can also adjust the deployment as soon as possible according to the data,and focus on optimization and publicity,so as to bring more long-term benefits to the business.At the present stage,such emotion recognition technology has been mature,but there is still problem facing that customers usually do not make corresponding comments after living in the hotel,except for bad comments and text comments when the living experience is bad,which brings huge problems to information collection.In addition,RNN neural networks present,when performing operations,slow computation speed,difficulty in accessing information from previous operations,inability to consider any future input from the current state,poor training effect and unclear probability when combined with speech recognition.In addition,there is less data related to sentiment analysis of hotel evaluation texts at home and abroad.To solve the above problem,this paper references the LSTM neural network,which has the advantage in using a special implicit unit that keeps the input for a long time,called the special unit of the memory cell.The special unit is similar to the accumulator and gated neuron,which will have a weight and connect itself at the next time step,copying the true value of its own state and accumulating external signals.But this self-connection is controlled by a multiplicative gate that is learned by another unit and decides when to clear the memory content.Also,the LSTM neural network solves the Vanishing Gradinet problem that arises from gradient backpropagation process with gradual reduction.The main contributions of this paper are summarized as follows.Firstly,this paper constructs a hotel review text dataset,analyzes the hotel review text characteristics,writes a text collection procedure that can efficiently collect hotel review text data from various APPs,and construct a training and testing dataset for the text algorithm model after data cleaning,sorting and labeling.Secondly,this paper proposes to combine LSTM neural network and speech recognition sentiment analysis,using regular expressions and jieba tools to clean the collected data,and then combine the application of words,i.e.Word2 vec,to improve the effect of 0-1 word discrimination,which greatly improves the accuracy of the recognition process.Thirdly,in Chapter 4,an LSTM-Self-Attention model is designed for extracting sentiment features from hotel review data.Hotel review data contains rich emotional features,and traditional machine learning is difficult to extract deep emotional features,while convolutional neural network(CNN)tends to ignore the temporal order of text.LSTM takes into account the temporal order of text,which can well combine the before and after information of data,and then combined with the self-attention mechanism,it can effectively weight the emotional features of hotel reviews.Therefore,this paper proposes the LSTM-Self-Attention model,which can effectively extract hotel sentiment features and achieve positive and negative sentiment binary classification of hotel review text.Finally,this paper uses real-time speech recognition to generate the evaluation text immediately after the evaluation.The experimental results show that the hotel review text data set collected by distributed crawler in this paper,combined with the algorithm model proposed in this paper(LSTM-Self-Attention),compared with other algorithm models,can better extract the emotional features of hotel reviews,have better accuracy,and effectively complete the binary classification of hotel reviews. |
