The monitoring of animal health worldwide, especially the early detection of emerging outbreaks, is one of the means of preventing and anticipating the introduction of infectious diseases into a territory. Animal disease surveillance is traditionally based on official data produced by international health authorities like the World Organization for Animal Health. As outbreak notification can be responsible for economic and commercial restrictions, the data quality can be impacted regarding sensitivity and timeliness. On the contrary, unofficial sources such as news reports are immediately available and can timely detect new outbreaks. In order to complete the official data, a platform dedicated to automatic surveillance of electronic media, PADI-web (" Platform for Automated extraction of animal Disease Information from the web "), was created. This tool detects, classifies and extracts information from reports, without any human intervention, regarding five animal diseases of interest: African swine fever, avian influenza, foot-and-mouth disease, Schmallenberg virus and bluetongue. The tool first collects news reports thanks to specific RSS feeds. Then, irrelevant news reports are removed during the classification step. Eventually, an information extraction module allows extracting epidemiological indicators from the text. It relies on text mining methods combining a rule-based approach and an SVM classification. In this Ph.D. project, we will address several challenges: 1. Improving the identification of the information: for instance, dealing with geographical ambiguities 2. Finding what types of indicators we should use to detect animal health events 3. Identifying methods to combine multi-source data (official vs unofficial) and indicators, to produce relevant knowledge for the monitoring.