Vera Lúcia Strube de Lima

A systematic review of named entity recognition in biomedical texts

Biomedical Named Entities (NEs) are phrases or combinations of phrases that denote specific objects or groups of objects in the biomedical literature. Research on Named Entity Recognition (NER) is one of the most disseminated activities in the automatic processing of biomedical scientific articles. We analyzed articles relevant to NER in biomedical texts, in the period from 2007 to 2009, through a systematic review. The results identify the main methods in the recognition of Biomedical NEs, features and methodologies for a NER system implementation. Aside from the tendencies identified, some gaps are detected that may constitute opportunities for new studies in the area.

Evaluation of a thesaurus-based query expansion technique

This paper concerns the use and evaluation of a thesaurusbased query expansion method in information retrieval. The query expansion process assigns weights to different types of relations obtained fom vocabulary structures, providing an efficient way to measure distance measure between different terms. This method, tested for Portuguese, improved the overall information retrieval performance on small corpora and over the Internet.

Binary lexical relations for text representation in information retrieval

Text representation is crucial for many natural language processing applications. This paper presents an approach to extraction of binary lexical relations (BLR) from Portuguese texts for representing phrasal cohesion mechanisms. We demonstrate how this automatic strategy may be incorporated to information retrieval systems. Our approach is compared to those using bigrams and noun phrases for text retrieval. BLR strategy is shown to improve on the best performance in an experimental information retrieval system.

Open information extraction based on lexical semantics

BackgroundOpen Information Extraction (Open IE) aims to obtain not predefined, domain-independent relations from text. This article introduces the Open IE research field, thoroughly discussing the main ideas and systems in the area as well as its main challenges and open issues. The paper describes an open extractor elaborated from the belief that it is not necessary to have an enormous list of patterns or several types of linguistic labels to better perform Open IE. The extractor is based on generic patterns that identify relations not previously specified, including rules corresponding to Cimiano and Wenderoth proposal to learn Qualia structure.MethodsNamed LSOE (Lexical-Syntactic pattern-based Open Extractor) and designed to validate such strategy, this extractor is presented and its performance is compared with two Open IE systems.ResultsThe results demonstrate that LSOE extracts relations that are not learned by other extractors and achieves compatible precision.ConclusionsThe work reported here contributes with a new Open IE approach based on pattern matching, demonstrating the feasibility of an extractor based on simple lexical-syntactic patterns.

Open Information Extraction Based on Lexical-Syntactic Patterns

Open Information Extraction (Open IE) is an unsupervised strategy to draw out relations from text without predefining these relations, regardless the domain. This paper describes a novel Open IE approach that performs unsupervised extraction of triples by applying a few lexical-syntactic patterns to POS-tagged texts. In order to validate this strategy we developed a prototype and compared its performance with two Open IE systems. The proposed approach achieved promising results, overcoming those from the state-of-the-art systems. The paper concludes with an analysis of errors and directions for future work.

A semi-automatic method for domain ontology extraction from Portuguese language Wikipedia's categories

The increasing need for ontologies and the difficulties of manual construction give place to initiatives proposing methods for automatic and semi-automatic ontology learning. In this work we present a semi-automatic method for domain ontologies extraction from Wikipedias categories. In order to validate the method, we have conducted a case study in which we implemented a prototype generating a Tourism ontology. The results are evaluated against a manually built Golden Standard reporting 79.51% Precision and 91.95% Recall, comparable to those found in the literature for other languages.

Tools for nominalization: an alternative for lexical normalization

The recognition of morphological variation and conceptual proximity of the words is crucial for tasks where the lexical normalization is used, such as term generation and matching in an information retrieval environment. We present tools that automatically perform nominalization for lexical normalization in Portuguese. Comparing the effects of three alternative strategies (stemming, lemmatizing, and our proposal: nominalization), we demonstrate through an experimental evaluation that nominalization, as lexical normalization, contributes to the performance improvement in a probabilistic information retrieval approach for Portuguese.

Query expansion based on thesaurus relations: evaluation over internet

In this work we present a heuristic for query expansion and its evaluation with information retrieval over the Internet. We obtained the precision and recall measures for the top-50 documents from 13 different queries. For those we had good results on estimated recall and F-measure values indicating that query expansion is a reasonable technique when few documents are retrieved.

Possessive Pronominal Anaphor Resolution in Portuguese Written Texts

This paper describes a proposal for Portuguese possessive pronominal anaphor (PPA) resolution, a problem little considered so far. Particularly, we address the problem of Portuguese 3rd person intrasentential PPAs seu/sua/seus/suas (his/her/their/its, for human and non-human subjects in English), which constitute 30% of pronominal occurrences in our corpus (Brazilian laws about environment protection). Considering some differences between PPAs and other kinds of anaphors, such as personal or demonstrative pronouns, we define three knowledge sources (KSs) for PPA resolution: surface patterns (taking in account factors such as syntactic parallelism), possessive relationship rules and sentence centering. These knowledge sources are organized in a blackboard architecture for PPA resolution, which provides both knowledge and procedure distribution among autonomous entities (reflexive agents), each of them specialized in a particular aspect of the problem solving. The proposal has been implemented and its results are discussed at the end of the work.

Experiments on extracting semantic relations from syntactic relations

This work presents the results of the application of a technique for automatic extraction of semantic relations among words from a corpus. The technique used is the one proposed by Grefenstette in [1]. We brought contributions to the syntactic context notion in [1], aiming to improve the identification of semantically related words. Then, we carried on three different experiments using a Portuguese language corpus: the first one compares the original Grefenstettes technique with the technique modified with our contributions, the second experiment investigates which syntactic relation is more relevant when identifying semantic relations, and the last experiment investigates the influence of the parser errors on the quality of the extracted semantic relations. Results and their analyses are detailed in this article.