Semantic Annotation and Shallow Parsing for the Creation of Abstract Graphs from Political Debates
Many governments have taken initiatives to open up and share their data such as parliamentary debates. This type of corpus is generally very large, covers several topics and is not always organized in a linear manner. In this research, we propose a method to automatically extract abstract representations based on graphs that represent the topics discussed during political debates and the relationships between these topics. To this end, we use semantic annotators based on Linked Data for extracting topics. In this way, we can represent the discussed topics with concepts whose semantics has already been defined on the Linked Data cloud in a structured way, unlike existing methods which generally rely on simple keywords. Also, we extract relations between the concepts based on the information available on the Linked Data Cloud and provide high level relations between these entities from the corpus of debates. These relations are extracted with morpho-syntactic patterns defined manually and disambiguated using VerbNet. With the concepts and relationships extracted, we construct an abstract graph representing the debates. This graph is successively reduced based on several parameters to keep only the most important entities and relationships. The generated graph, in addition to enabling semantic search, could be reused by other systems for the generation of abstractive summarization or question answering.
For more information, click here
Semantic Annotation for the Analysis of Political Debates: A Graph-based Approach
International Conference on the Advances in Computational Analysis of Political Text, Dubrovnik