Homestay Customer Opinion Mining Based On Character-level Convolutional Neural Networks

Online review contains a wealth of customer opinion information.Customer opinion mining can help the related industries.It is also an excellent channel for companies to understand consumers' needs,improve related products or services,and promote product sales.The survey way is mainly questionnaire survey,which takes a lot of time,and the requirements of the researchers are very high,and the sample volume involved is very limited.Considering that user-generated content has become public big data in recent years,consumers generally publish a large number of comments after purchase.From the distribution and quantity of data,such online comments are more reliable than traditional research methods.The traditional method has not been enough to accurately reflect the opinions and needs of customers in terms of quantity and quality.Therefore,researching unstructured text mining technology and improving service quality by means of opinion mining technology is the key to quickly accumulate competitive advantage and has certain practical value.We did it based on the comment text analysis,with the help of automation technology and deep learning,we did some studies about customer opinion mining with the review data of Ctrip Homestay.At the end,a variety of text classification algorithms are compared and the model is described.The advantages of specific research contributions are reflected in the following three aspects.Firstly,we designed a comment data collection method based on the combination of Requests POST and Scrapy,according to the special structure of Ctrip's website and the crawler strategy of Ctrip.com,based on the data collection method of Requests POST and Scrapy in Python,which increases the efficiency of data collection.Break through the limitations of anti-reptiles,and by writing a data cleaning script,the collected numbers can be cleaned and then stored in the database as a source of data for this article.Secondly,we designed a construction method of Homestay topic dictionary based on Latent Dirichlet allocation,customer topic extraction is the basis of fine-grained analysis.The commentary of the Homestay Inn is easy to see the phenomenon of multiple homestays.We proposes a corpus cleaning strategy for long sentences and short sentences.The comment preprocessing method with punctuation marks can make the different evaluation subjects scattered in a comment pass punctuation,and then use a variety of vectorization methods,cluster the comment topic with the implicit Dirichlet method.Finally,we established a Homestay theme attribute dictionary.Compared with the standard documents of the hotel and the evaluation indicators of the hotel in Homestay,this article enriches and embodies various indicators.It is verified by experiments that the application of this method in the implicit topic mining.Thirdly,we proposed a customer opinion sentiment mining method based on Character-level Convolutional Neural Networks(C-CNN-SA),using automatic annotation and convolutional neural networks to conduct fine-grained sentiment analysis on customer opinions under different topics,using customer scoring and commenting emotional mapping,using data automatic annotation and weak supervision pre-training to automatically expand the dataset without using In the case of word segmentation.Experiments have confirmed that the classification accuracy and F-value of this method are higher than others in the two-category linguistic emotion classification.In the character-level classification model,experiments have confirmed that convolutional neural networks have the best effect in short text sentiment classification,and character-level convolutional neural networks have obvious advantages in training speed and effect.