Interactive coding of responses to open-ended questions in Russian

Abstract

We propose an interactive technique to categorize the responses to open-ended questions. The open-ended question requires a response which is a natural language phrase. A typical analysis of the phrases starts with their ’coding’, that is, identifying themes of the responses and tagging the responses with the themes they represent. The proposed coding technique is based on interactive cluster analysis. We study theoretically and empirically the hierarchical (agglomerative, divisive) and partitional clustering algorithms to pick the best one for short Russian responses. We address the problem of the short phrase sparseness with thesaurus smoothing. We introduce an iterative process where users can provide some feedback to a clustering result. A domain-oriented system of statements is developed for users’ feedback. The system is proved to be able to provide any clusters the user desires. The technique is implemented as a web service for responses in Russian.