In this context, coding means “categorizing” or labeling the tweets into predetermined categories. Each conversation may contain several individual posts by Twitter users (tweets), including the original tweet and responses usually from For simplicity, “tweet” and “conversation” are interchangeable. In this research, the unit of analysis is a Twitter conversation. We were able to download 365,250 individual tweets which gave us a total of 169,023 conversations dating from January 1st, 2015 through to September 2nd, 2020. In order to speed up the review process, we used the paid Twitter developer API for bulk downloads of all the tweets shared by Airbnb users. All tweets used in this study are publicly available on the Twitter platform at the time of publishing and mentioned one or more of the following Twitter accounts: etc. Note: the methodology shared below was initially described by Binns and Kempf (2020) in their report for ASIS which is titled, “ Using Social Media to Gather Security Intelligence” and is quoted or paraphrased here. After several iterations and by employing different natural language processing techniques, we achieved an F1-score ranging from 0.72 to 0.86 across all categories. Using this “gold standard” dataset we were able to then train our custom-built machine learning system to review the remaining 160,000+ conversations. This allowed us to measure intercoder reliability and we eventually reached 100% reliability by reviewing and correcting all the tweets they did not agree on. We had two coders human-review and categorize a total of 5,278 tweets without seeing how the other coder was categorizing each conversation. Of the 169,023 conversations, 127,183 were from guests reporting their problems. We used a combination of human coding of individual Airbnb Twitter complaints, machine learning, and natural language processing techniques to accelerate the categorization and understanding of such a large dataset of 365,250 individual tweets or 169,023 Twitter conversations. In this academic research, we set out to understand one simple question, “which Airbnb guest problems are most likely to be complained about via Twitter?”. Please note: If you’d like to be the first to know when we update this study or when we publish new research, subscribe HERE ➜ This 2021 Airbnb research is a follow-up to my 2017 study, “Is Airbnb Safe?” which was inspired by my own Airbnb nightmares in Paris with my wife and baby.īelow is a summary of our key findings and further down the page we list the cities with the most Airbnb complaints. The 2.5-year project was funded by ASIS International (the world’s largest membership organization for security management professionals) and via a grant from John Jay College. I conceived the original study method in early 2019 and shortly thereafter began a collaboration with researchers from John Jay College of Criminal Justice and the University of Colorado School of Public Affairs in Colorado Springs. We reviewed all mentions and replies of etc. We analyzed 127,183 Airbnb guest complaints to find out which problems are most likely to be complained about via Twitter.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |