https://hdl.handle.net/20.500.12008/48532 Thu, 17 Sep 2026 21:51:35 GMT 2026-09-17T21:51:35Z https://hdl.handle.net/20.500.12008/56544 Título: Automatic knowledge-graph creation from historical documents: the Chilean dictatorship as a case study Autor: Díaz, Camila; Dunstan, Jocelyn; Etcheverry, Lorena; Fonck, Antonia; Grez, Alejandro; Mery, Domingo; Reutter, Juan; Rojas, Hugo Resumen: We present our results regarding the automatic construction of a knowledge graph from historical documents related to the Chilean dictatorship period (1973-1990). Our approach consists on using LLMs to automatically recognize entities and relations between these entities, and also to perform resolution between these sets of values. In order to prevent hallucination, the interaction with the LLM is grounded in a simple ontology with 4 types of entities and 7 types of relations. To evaluate our architecture, we use a gold standard graph constructed using a small subset of the documents, and compare this to the graph obtained from our approach when processing the same set of documents. Results show that the automatic construction manages to recognize a good portion of all the entities in the gold standard, and that those not recognized are mostly explained by the level of granularity in which the information is structured in the graph, and not because the automatic approach misses an important entity in the graph. Looking forward, we expect this report will encourage work on other similar projects focused on enhancing research in humanities and social science, but we remark that better evaluation metrics are needed in order to accurately fine-tune these types of architectures. Mon, 01 Jan 2024 00:00:00 GMT https://hdl.handle.net/20.500.12008/56544 2024-01-01T00:00:00Z https://hdl.handle.net/20.500.12008/56413 Título: Post-OCR correction using large language models with constrained decoding Autor: Sastre, Ignacio; Etcheverry, Lorena; Rey, Guillermo; Moncecchi, Guillermo; Rosá, Aiala Resumen: This article addresses the problem of correcting noisy Optical Character Recognition (OCR) outputs from digitized historical documents, specifically those from the Berrutti Archive related to Uruguay’s civic-military dictatorship. These documents—produced with typewriters, diverse layouts, and overlaid annotations—pose significant hallenges for standard OCR tools, resulting in highly errorprone text. We present a novel post-OCR correction method that leverages fine-tuned open-source Large Language Models (LLMs) combined with a constrained decoding strategy. This strategy incorporates character-level similarity between the OCR input and the generated output at decoding time, steering the model toward corrections that closely preserve the original text structure. We evaluate our method on a gold-standard dataset of over 2000 annotated lines and show that it outperforms prompting and standard fine-tuning approaches, reducing both character error rate (CER) and word error rate (WER). The corrected outputs provide more accurate input for downstream tasks, such as named entity recognition, relation and event extraction, and knowledge graph construction, thereby supporting the broader goal of extracting knowledge from historically significant and sensitive archives. Wed, 01 Jan 2025 00:00:00 GMT https://hdl.handle.net/20.500.12008/56413 2025-01-01T00:00:00Z https://hdl.handle.net/20.500.12008/56289 Título: Skeleton of locally most separable 3-terminal graphs Autor: Würth, Simon; Romero, Pablo Resumen: A 3-terminal graph is a graph with 3 distinguished vertices, called terminals. Let Tn,m be the set of all connected 3-terminal graphs on n vertices and m edges. Let G be in Tn,m. A spanning subgraph H of G is 3-separable when H has precisely 3 components, each containing one terminal. For each p in [0, 1], the separability of G at p, denoted SG(p), is the probability that the resulting subgraph of G is 3-separable after each edge in G is removed independently with probability 1−p. The graph G is a locally most separable 3-terminal graph (LMS3TG) if for each H in Tn,m there exists δ > 0 such that SG(p) ≥ SH(p) whenever p ∈ (1 − δ, 1). Let Cn,m be the set of all connected simple graphs on n vertices and m edges. For each G in Cn,m, the skeleton G′ of G is the graph arising from G by the contraction of each of its bridges. In this note, we show that the skeleton of each LMS3TG in Tn,m is an almost-complete graph. Descripción: Documento elaborado en el marco de la pasantía de Investigación de PEDECIBA - Informática.; Orientador: Pablo Romero. Thu, 01 Jan 2026 00:00:00 GMT https://hdl.handle.net/20.500.12008/56289 2026-01-01T00:00:00Z https://hdl.handle.net/20.500.12008/56235 Título: Grafos uniformemente más confiables y grafos fuertes Autor: Cabrera, Agustín Resumen: Sea Cn,m la clase de grafos conexos y simples con n vértices y m aristas. El co-rango de Cn,m y de cada uno de sus grafos es igual a m − n + 1. Sea G en Cn,m. Para cada ρ en [0, 1], se define la confiabilidad de G en ρ como la probabilidad de que el subgrafo obtenido de remover cada arista de G independientemente con probabilidad ρ sea conexo. Decimos que G es uniformemente más confiable si para cada H en Cn,m y cada ρ en [0, 1] se cumple que RG(ρ) ≥ RH(ρ). Decimos que un corte de G es un subconjunto de aristas U de G que cumple que G − U no es conexo. El número de cortes de G con k elementos se denota µk(G). Decimos que G es fuerte si para cada H en Cn,m y cada k en {0, . . . , m} se cumple que µk(G) ≤ µk(H). Es cierto que todo grafo fuerte es uniformemente más confiable. Boesch [J. Graph Theory 10 (1986), 339–352] conjeturó que todo grafo uniforme más confiable es fuerte. Se sabe que en cada una de las clases Cn,m cuyo co-rango es 4 o menor existe al menos un grafo uniformemente más confiable, y además cada grafo uniformemente más confiable es fuerte. Si bien existen infinitas clases Cn,m cuyo co-rango es igual a 5, se demostró recientemente que hay tan solo una cantidad finita de dichas clases que tienen un grafo que es uniformemente más confiable. Un proyecto ambicioso consiste en buscar una clase Cn,m que posea un grafo uniformemente más confiable que no sea fuerte. Dicha clase, en caso de existir, refutaría a la conjetura de Boesch. En este proyecto se realiza un estudio computacional relativo a la existencia o inexistencia de grafos uniformemente más confiables y de grafos fuertes dentro de clases Cn,m de co-rango 5. Descripción: Módulo de Taller de Ingeniería en Computación.; Orientador: Pablo Romero. Thu, 01 Jan 2026 00:00:00 GMT https://hdl.handle.net/20.500.12008/56235 2026-01-01T00:00:00Z