View article

We propose an algorithmic approach for ranking of differing textual descriptions (accounts) of the same event or story according to their likeliness to best describe the source. Application domains include the ranking of eyewitness reports, historical accounts, and news reports. For this, we developed the concept of “semantic centrality” as a measure of how central a text is among a collection of texts, in terms of its semantic overlap or similarity with all other texts. This measure is based on natural language processing theory, as well as graph theory.

Using three different collections of humanly generated texts (gathered through a recall task,“Chinese Whispers” task, and real-world news reports), we evaluated the proposed method for algorithmic ranking of textual accounts by their trustworthiness to describe source events. In one experiment algorithmic ranking is compared to human ranking. Results indicate that semantic centrality as a measure for trustworthiness of textual accounts is promising and deserves further research attention.