Building and validating hierarchical lexicons with a case study on personal values

Abstract

We introduce a crowd-powered approach for the creation of a lexicon for any theme given a set of seed words that cover a variety of concepts within the theme. Terms are initially sorted by automatically clustering their embeddings and subsequently rearranged by crowd workers in order to create a tree structure. This type of organization captures hierarchical relationships between concepts and allows for a tunable level of specificity when using the lexicon to collect measurements from a piece of text. We use a lexicon expansion method to increase the overall coverage of the produced resource. Using our proposed approach, we create a hierarchical lexicon of personal values and evaluate its internal and external consistency. We release this novel resource to the community as a tool for measuring value content within text corpora.