The search for improving health safety contributes to the prevention of harm to people and reduces costs for health systems. In this context, the focus of this project is on the application of data mining, natural language processing, and information retrieval techniques on three distinct fronts but with a common thread - health safety. The first work front deals with safety in the context of compiling evidence on the effectiveness of health treatments. The second work front deals with safety in the context of information retrieved by search engines. The third work front deals with medications, as they represent the most common clinical intervention in the health area. According to the World Health Organization (WHO), unsafe medication practices and medication-related errors are the main causes of preventable harm to health systems worldwide. The project's contribution is in the proposal of a machine learning-based system to automatically detect prescriptions with potential errors. Among the scientific contributions is the proposal of new algorithms and methodologies to solve the aforementioned problems.

Funded by CNPq

The growing availability of data in the medical field coupled with the maturation of data mining techniques go on fostering collaboration between Medicine and Computing. This collaboration is motivated by the potential positive impact that the discovery of knowledge about medical data can generate for the improvement of people's health. However, there are some challenges that need to be addressed so that the application of mining algorithms can yield useful results that have practical value for the medical area. Medical data is typically heterogeneous, with both structured and unstructured data (text, images and vital signs). Furthermore, the volume and the need for confidentiality may represent practical challenges. The focus of this project is on the application of data mining and natural language processing to analyze electronic health records.

Funded by CNPq

E&P data management specialists agree that the most valuable information for an organization is represented in the form of unstructured data. Textual documents, emails, images, and diagrams are typical examples of such information. Estimates say that employees spend almost ten hours a week searching and gathering information. Aiming at reducing the effort in searching for relevant information, the overall goal of our project is to provide a solution for multimodal information retrieval (MIR). MIR is the process of organizing and enabling search for different types of data, i.e. modalities, such as text, image, audio, video, or 3D models. Here, our focus is on unstructured texts and figures.

Funded by CENPES/Petrobras

Opinion Mining is the computational study of opinions, sentiments and emotions expressed in texts. Its groal is, from a collection of documents, to identify, classify and aggregate the sentiment about a target automatically. This area has been gaining attention with the increase in the number of users who publish their opinions and experiences on the Web. The focus of this project is on Multilingual Mining Opinion (MOML), which aims to address opinions expressed in several languages, allowing to analyze a greater number of reviews, as well as obtaining a greater diversity and density of people's views. Our motivation comes from observing three limitation of MOML. The first limitation refers to the fact that the habit of writing comments evaluating products and / or services is much more frequent in some regions of the world than in others. In order to remedy this deficiency, we propose to develop methods that use customer reviews written in foreign languages ​​to provide product summaries on Brazilian websites. The second limitation observed concerns the scarcity of resources for MOML involving languages ​​other than English, especially considering the classification of emotions expressed in the text (and not just the polarity). In this context, the objective is to develop methods for the classification of emotions in multilingual texts. The third limitation refers to the treatment of contradictions found in the evaluation texts in order to identify sentences that express contradictory feelings and how this impacts the Opinion Mining methods.

Funded by CNPq

Multilingual information is available in many different sources and formats . This fact has motivated research aimed at finding mappings between data represented in different languages in the areas of Information Retrieval, Natural Language Processing, and more recently, Databases . This project focuses on researching and proposing new methods for matching of multilingual data in different scenarios. Our goals are:(i) collecting parallel corpora on the web , (ii) finding correspondences in multilingual Wikipedia, and (iii) detecting multilingual plagiarism. The results of this project will represent contributions to the fields of Information Retrieval and Natural Language Processing by providing parallel corpora that are very important resources for the advancement of these areas. The wide availability of multilingual data also facilitates plagiarism. Multilingual plagiarism detection will also be targeted in this project. In this topic, our main contribution will be the inclusion of analysis of citations and references that is essential for the confirmation of plagiarism.
Funded by CNPq (Edital Universal)

With the dissemination of the Web, millions of people gain access to information from several areas of knowledge. The number of digital information available has grown immensely. However, despite its many benefits, the Web is one of the easiest means to enable plagiarism. Plagiarism is one of the most serious forms of academic misconduct. It is defined as “the use of another person's written work without acknowledging the source”. Recent research has shown that this type of misconduct is increasingly frequent in the academic world. This fact has been motivating techniques to automate plagiarism detection. This project focuses on cross-language plagiarism detection, in which the contents of a document are translated without making any reference to its source. Our aim is to develop an efficient method to enable the detection of cross-language plagiarism.
Funded by CNPq (Edital Universal)

The aim of this project is to contribute to the development of  Cross-Language Information Retrieval (CLIR) involving Portuguese. The motivation is the growing need to explore documents in foreign  languages, experienced by more and more people. CLIR research has been developing quickly since the late 90's. Despite recent advances there are still many aspects left unexplored, specially regarding the use of Portuguese. In this project we hope to develop a system that accepts queries in Portuguese and searches for documents in English. In addition, the following aspects will be investigated: (i) development of stemming algorithms for Portuguese; (ii) proposal of new techniques for mapping concepts between languages through the analysis of parallel and comparable corpora; (iii) study of the process of relevance feedback in a CLIR environment; and (iv) development of techniques for the identification of multi-work expressions.
Funded by CNPq (Edital Universal)

In the classical view, the areas of Information Retrieval (IR) and Databases (DB) have little in common. DB normally deal with structured data while IR deals with unstructured documents, typically in the form of free text. Considering that data stored by most organisations is both structured and unstructured, and that users frequently have to query data in both formats, there is a growing need to integrate the two areas. The aim of this project is to apply IR concepts into DB systems to facilitate the process of solving imprecise queries. The challenge it to modify the processing of queries to include the notions of similarity and relevance.
Funded by CAPES-PRODOC