| With the advent of the post-epidemic era,the online short-term rental market is bound to rebound along with the recovery of tourism,and targeted operations by gaining insights into renters’ preferences and needs can help boost listing sales.This uses review data from the Airbnb platform from 2020 to the present and listings released in late 2022 to explore and analyze the factors influencing short term rental sales in Amsterdam and Chicago based on review text analysis and machine learning algorithms.Firstly,text analysis was performed on tenants’ online reviews.The text of each review is divided into emotional polarity,and 100 high-frequency words in positive reviews and 30 high-frequency words in negative reviews are obtained respectively after removing deactivated words,and the high-frequency words are categorized according to six aspects: housing attributes,landlord attributes,platform attributes,hardware facilities,transportation,and management services,so as to explore tenants’ concerns when renting and the reasons that lead to negative comments by tenants,and to provide a theoretical basis for later feature derivation and rationalization suggestions to provide a theoretical basis.Regression analysis was then performed.After data pre-processing of the listing data table,about 40 hardware facilities and living and entertainment items were extracted from the two text class fields of listing description and facilities based on the results of text analysis,and feature derivation was performed.Lasso model and stepwise regression model were established respectively to initially screen out the features that have significant effects on sales;on this basis,random forest(RF),GBRT and XGBoost were introduced to construct integrated learning models respectively,using The performance of the models was judged by three indicators,MAE,and RMSE,and it was found that the integrated learning model was more effective than the single regression model,in which the Lasso-RF model had higher values in both the training and test sets.The Lasso-RF model has higher values than other models,reaching 0.98 and 0.89 in the training and test sets,respectively,and the model also has better stability.The Lasso-RF model was used to obtain the importance ranking of the factors influencing sales in Amsterdam and Chicago,respectively,and to interpret the attribution and analyze the similarities and differences.Finally,targeted recommendations are made based on the analysis results.For the city of Amsterdam,the platform can operate in a targeted manner by incentivizing tenants to generate rating and evaluation behaviors and introducing coupons;for the city of Chicago,the platform should invest major efforts in assisting landlords to improve the infrastructure of their houses. |
