A Two-Iteration Clustering Method to Reveal Unique and Hidden Characteristics of Items Based on Text Reviews
Alon Dayan, Osnat Mokryn, Tsvi Kuflik · 2015
This paper presents a new method for extracting unique features of items based on their textual reviews. The method is built of two similar iterations of applying a weighting scheme and then clustering the resultant set of vectors. In the first iteration, restaurants of similar food genres are grouped together into clusters. The second iteration reduces the importance of common terms in each such cluster, and highlights those that are unique to each specific restaurant. Clustering the restaurants again, now according to their unique features, reveals very interesting connections between the restaurants.