Story retrieval and comparison using concept patterns

• Main Authors: Krakauer, Caryn E. (Author), Winston, Patrick Henry (Author)
• Format: Article
• Language: English
• Published: © The Association for Computational Linguistics, 2022-04-06T17:38:11Z.
• Subjects: Article
• Online Access: Get fulltext

Traditional story comparison uses key words to determine similarity. However, the use of key words misses much of what makes two stories alike. The method we have developed use high level concept patterns, which are comprised of multiple events, and compares them across stories. Comparison based on concept patterns can note that two stories are similar because both contain, for example, revenge and betrayal concept patterns, even though the words revenge and betrayal do not appear in either story, and one may be about kings and kingdoms while the other is about presidents and countries. Using a small corpus of 15 conflict stories, we have shown that similarity measurement using concept patterns does, in fact, differ substantially from similarity measurement using key words. The Goldilocks principle states that features should be of intermediate size; they should be not too big, and they should not too small. Our work can be viewed as adhering to the Goldilocks principle because concept patterns are features of intermediate size, hence not so large as an entire story, because no story will be exactly like another story, and not so small as individual words, because individual words tend to be common in all stories taken from the same domain. While our goal is to develop a human competence model, we note application potential in retrieval, prediction, explanation, and grouping.
This material is based on work supported by the U.S. Office of Naval Research, Grant No. N00014-09-1-0597. Any opinions, findings, conclusions or recommendations therein are those of the author(s) and do not necessarily reflect the views of the Office of Naval Research.