| dc.description.abstract | There is great expectation regarding the use of natural language as an interface of communication with machines. Among the several applications that implement such an interface, the Semantic Question Answering systems arises, enabling the localization of information in knowledge bases from questions formulated in natural language. It is possible to notice in the work in progress a tendency to implement solutions based on the lexical and morphological information of the questions, ignoring the higher level abstract information of the linguistic processing. This thesis presents an approach that explores in depth the syntactic and structural information of the questions, based on these higher levels of linguistics to understand the meaning of the words and to find answers in semantic knowledge bases. This approach proposes a model that makes use of the linguistic information of the question to determine its type and select the keywords that will be used to search answers in the knowledge base. Unlike other works, the model proposes a solution based on linguistics integrated with two different implementation techniques, aiming to present a solution that exploits the advantages that each technique offers. The approach uses the morphosyntactic and structural informations of the sentence to determine the type of the question and to select its key words. Later, it uses linguistic information to optimize the performance of the algorithm of generation and ranking of candidates for the response based on graph. Finally, if the integrated use of linguistic information with the graph-based technique is not enough for the unequivocal selection of the answer, our approach look for support in the latent semantics of word embedding to validate the answers. The experiments of evaluation of the approach showed a performance above the other competitors, with a score F1 micro of 0.56 and F1 score Macro of 0.593. | en |
