24 May 2024
09:00 Doctoral defense Room 85 of IC2
Theme
Evolution of Entities in Temporal Knowledge Graphs
Student
Anderson Rossanez
Advisor / Teacher
Julio Cesar dos Reis - Co-supervisor: Ricardo da Silva Torres
Brief summary
Considering corpora of temporal texts, such as articles that document knowledge produced in scientific research, an adequate representation of temporal knowledge, as well as the analysis of its evolution, will potentially benefit researchers, the scientific community and society in general.
Analyzing the evolution of knowledge is a challenging research problem, requiring the identification and organization of entities representing concepts found in documents. Entities are generally referenced in long and complex sentences, presenting implicit relationships, abbreviations and coreferences. Identifying entities requires prior knowledge in the text domain, making it difficult to perform the task automatically. Encoding the relevance of entities into temporal knowledge representations requires additional computational processing techniques. Analyzing the resulting immense amount of temporal data poses difficulties, requiring appropriate visualization techniques.
This thesis aims to design, develop and evaluate new methods for representing widespread knowledge in unstructured texts in natural language, via semi-automatic generation of knowledge graphs. The research aims to assist in the analysis of the evolution of knowledge, specific to the variation in the relevance of entities. We develop the concept of temporal knowledge graphs in modeling the evolution of knowledge, in unstructured text corpora. In our approach, we apply complex network measurements to temporal knowledge graphs to encode the relevance of their entities. Additionally, we researched new visual representations for this evolution, promoting analyzes based on the evolution of knowledge.
Our contributions include (1) a method for generating temporal knowledge graphs from corpora of unstructured texts using natural language processing techniques; (2) introduction of centrality measures based on knowledge graphs, encoding the relevance of entities, supported by (3) a study that characterizes properties of complex networks in knowledge graphs, which guarantees the applicability of complex network measures in analyzes of knowledge graphs; visualization methods for (4) representing and navigating through the structure of temporal knowledge graphs, and (5) representing variations in relevance of entities over time.
This investigation also contributes to the implementation of software tools that instantiate the original methods developed. These tools were used to evaluate the methods, using real text documents, obtained from scientific sources, in the fields of Computer Science and Neurology.
Examination Board
Headlines:
| Julio Cesar dos Reis | IC / UNICAMP |
| Andre Santanche | IC / UNICAMP |
| Solange Oliveira Rezende | ICMC / USP |
| Damires Yluska de Souza Fernandes | IFPB |
| Fernando Vieira Paulovich | TU/e/Netherlands |
Substitutes:
| Jacques Wainer | IC / UNICAMP |
| Renata Wassermann | IME / USP |
| Daniel dos Santos Kaster | CCE/UEL |