Sistema automático de consulta ao boletim de serviços do IFG apoiado em processamento de linguagem natural

Abstract

The amount of texts generated daily within any company or institution is enormous, and this scenario would not be different in the Federal Institute of Goiás. There are a large number of documents that are generated due to the bureaucratic institutional dynamics, which depend on their actions through resolutions, ordinances, and others. These documents are made publicly available to society, but some of these are of specific interest to the IFG's internal community because they are actions that directly involve institutional daily life. In this respect, students and staff have at their disposal a service called Service Bulletin. To have access to them, it is necessary to access a specific website. On this site, your organization is grouped by year/month into a single file, causing users to have to check each content that makes up the newsletter manually one by one to find the information they need, demanding time and effort, making it a dull and unsuccessful job. To overcome this problem, this work proposes the development of a search solution based on the state of the art in processes of automatic treatment of textual documents, using, for this purpose, techniques of Natural Language Processing and Information Retrieval. After the bibliographic survey of the respective themes, some programming libraries were listed that could make possible the construction of the solution and, from the experimental use of these, it was possible to evaluate their effectiveness. In the course of the work will be described and discussed such tools and how they contributed to what was intended. As a methodology to implement an effective final solution we considered the use of the inverted index technique and, alternatively, the use of existing document storage and search tool, Elasticsearch. As expected, the result was quite satisfactory and we believe that after the last tests, the solution may be available to the end user in a short period.