[Back]

B. Panayotov:
"A Visual Exploration Tool forTemporal Analysis of Customer Reviews";
Supervisor: E. Gröller; Institute of Visual Computing & Human-Centered Technology, 2020; final examination: 2020-05-04.

AbstractThis thesis explores textual review data and how it changes over time. The thesisis motivated by the constantly generated textual reviews. Review sites like Yelp andTripAdvisor are generating hundreds of thousands of reviews monthly. Analysing thisamount of data is impossible by simply reading every individual review. We look forways to answer questions that business analysts, business owners, and investors ask aboutcustomer review data. This thesis asks questions such as: Why do review scores andtopics change over time? What are the major topics people discuss? What are the typicalreasons why review scores suddenly increase or decrease? What are topics that invokepermanent or transient changes in a large collection of review scores?We created a tool called Review Watcher, which provides novel approaches to examineand analyse review changes over time. The tool aims to provide simple, easily accessibleinformation regarding temporal changes in a collection of restaurant reviews. The tooluses real data provided by Yelp. It employs graphical ways to indicate changes in reviewscores over different periods of time. The tool analyses the review scores over time, andit tries to explain changes in these scores based on the textual content of the reviews.The tool utilises automated text processing algorithms to highlight important and oftenused words in text corpora.We used a qualitative evaluation to determine how well the tool manages to answer theresearch questions. We completed a user study with experts in the field of economics.They shared the insights they gathered using Review Watcher and compared them totheir experiences working with other tools for customer satisfaction and review analysis.As a result of our research, we show that Review Watcher manages to provide betterinsight into what are major topics in a collection of textual reviews. In the thesis, we showthat Review Watcher is better suited to highlighting review changes occurring over timeand giving insights to why the changes occurred, compared to existing tools for reviewexploration. The tool is also proving capable of handling millions of textual reviews oftens of thousands of restaurants with acceptable loading times for the user. The userstudy also reveals some of the tool´s limitations and potential for future work, for examplein introducing improved categorisation functions and geographical information aboutrestaurants.

http://dx.doi.org/10.34726/hss.2020.70281

https://publik.tuwien.ac.at/files/publik_291771.pdf