Extracting Abstraction Dimensions by Identifying Syntax Patterns from Texts

Extracting abstraction dimensions from a large set of texts is a way to efficiently manage, explore, and understand the texts from multiple dimensions. Subject, action, object, and adverbial are categories formed by abstracting various specific representations of syntax elements, which can be used to identify relations between them when constructing and understanding sentences. Abstraction relations between syntax elements within a set of texts represent its abstraction dimensions. This paper proposes a low-cost approach to automatically extracting abstraction dimensions from texts by constructing class trees on syntax elements with high quality and expressiveness. A high-quality class tree guarantees that all syntax elements and their subclass relations within texts can be represented. The independency of the class trees ensures that there is no redundant representation between the trees. The expressiveness ensures that most sentences can be accessed from every tree and the rest of sentences can be accessed from at least one tree so that the trees can be searched with queries in natural language representation. The average precision, recall, and F1-score of the constructed class trees are all greater than 80%. The applications of the proposed approach to implementing queries in natural language demonstrate that different types of question patterns for querying subject or object have high coverages on texts, and searching multiple trees with the question patterns can quickly locate answer sentences with reduced search space. This research provides a feasible approach to automatically constructing a space of texts with multiple abstraction dimensions to support advanced applications by extracting abstraction dimensions from a large set of texts.