Exploring Supervised Techniques for Automated Recognition of Intention Classes from Portuguese Free Texts on Agriculture

Technical and scientific knowledge is vast and complex, particularly in interdisciplinary fields such as sustainable agriculture, which is available in several interrelated, geographically dispersed and interdisciplinary online textual information sources. In this context, it is essential to support people with computational mechanisms that allow them to retrieve and interpret information in an appropriate way, as communication in these software systems is typically asynchronous and textual. User’s intention recognition and analysis in textual documents results in benefits for better information retrieval. However, intentions are expressed implicitly in texts in natural language and the specificities of the domain and cultural aspects of language make it difficult to process and analyze the text by computer systems. This requires the study of methods for the automatic recognition of intention classes in text. In this article, we conduct extensive experimental analyses on techniques based on language models and machine learning to detect instances of intention classes in texts about sustainable agriculture written in Portuguese. In our methodology, we perform a morphological analysis of the sentences and evaluate four Word Embeddings techniques (Word2Vec, Wang2Vec, FastText and Glove) combined with four machine learning techniques (Support Vector Machine, Artificial Neural Network, Random Forest and Transfer Learning). The results obtained by applying the techniques proposed in a database with textual information on sustainable agriculture indicate promising possibilities in the recognition of intentions in free texts in Portuguese language on sustainable agriculture.