Question-Answering using Keyword Entries in the Oil&Gas Domain

To improve the efficiency and accuracy of knowledge access in the oil industry, we propose to build a question-answer system for domain users. The system applies the state-of-the-art natural language processing techniques to analyze the domain knowledge in the form of keywords and their explanations. To this end, we collect over 20,000 keyword entries from reliable resources in the oil domain. The system then uses four different feature extraction techniques, including deep learning models, over these keyword entries to generate features, which are used to compute similarity between any given user question and keyword entries. Eventually, the keyword entries with the highest similarity scores are chosen as the answers and presented to the user. Our empirical evaluation shows that the traditional TF-IDF methods for feature extraction outperforms the other methods in our testing dataset.