Modelagem dos domínios da saúde e da violência na FrameNet Brasil para identificação de casos de violência de gêner

Abstract

Data on gender-based violence in Brazil and around the world are alarming, with one in three women worldwide having suffered physical or sexual violence. Research suggests that gender-based violence is underreported due to several factors, such as fear on the part of the victim or health professional, lack of information on the issue, and lack of knowledge about reporting or registration systems. Incidents of violence are recorded through SINAN – the Notifiable Diseases Information System; however, this system is not integrated with other systems that contain relevant records, such as the e-SUS Primary Health Care System (e-SUS APS) and the SIM – Mortality Information System. This work is part of an effort to link data between these systems, with regard to parameterized fields, and to perform frame-based semantic analysis, with regard to open fields, which contain information on medical care and violence records. Data linkage allows us to understand whether a given victim was treated at the health service before a possible situation of Gender-Based Violence. Furthermore, SIM records allow us to understand whether the victim died. Frame-based semantic analysis seeks to assist the task of computational algorithms by identifying what type of frames were evoked in the related domains, which are health and violence. The data from the corpora present in the work come from the health systems (SUS) of São Caetano do Sul - SP; and Recife - PE. The work contributes to the modeling of 74 frames and 4,457 lexical units for the domains. The frames are related to each other through 113 typed relations. A total of 5,452 qualia relations connect lexical units in the domains. After modeling, the analysis of the prevalence of frames and lexical units in the e-SUS APS and SINAN corpora allows us to identify possible trajectories of violence against women.