Fuzzy MLNs and QSTAGs for Activity Recognition and Modelling with RUSH

Toward the Automated Construction of Probabilistic Knowledge Graphs for the Maritime Domain

“…As many types of maritime incidents are recorded in a textual format (e.g., post-incident reports) and other associated activities such as pirates' communications and planning are entirely or partially conducted through words, being able to extract and construct a knowledge base from such textual data would not only provide an effective method for query answering, but also facilitate further analysis and forecasting (e.g., identifying dangerous zones, spatial-temporal patterns, predicting trends and potentially discovering plans for attacks). More significantly, the probabilistic knowledge graph constructed from text data can be combined with those constructed from other sources of data (e.g., rich contextual/domain knowledge and movement data generated by sensors as in [19]) in order to provide comprehensive situational awareness and assist with the detection of illegal maritime activities.…”

“…DeepDive can effectively address the problems of extraction, cleaning, and integration jointly, and treat them as a single probabilistic inference problem that takes all available information into account. Specifically, DeepDive was formulated on top of a large-scale inference engine based on Markov Logic [31], currently used as part of the high-level fusion and reasoning capability within RUSH for real-world situational awareness [19], [31]. This in effect allows Maritime DeepDive to (i) model the complexity and uncertainty of the extracted knowledge from natural language, (ii) combine learning from both expert knowledge and available data (via distant supervision), and (iii) facilitate future combination with probabilistic knowledge and events generated from hard data within RUSH [19] in a consistent and coherent manner.…”

“…Specifically, DeepDive was formulated on top of a large-scale inference engine based on Markov Logic [31], currently used as part of the high-level fusion and reasoning capability within RUSH for real-world situational awareness [19], [31]. This in effect allows Maritime DeepDive to (i) model the complexity and uncertainty of the extracted knowledge from natural language, (ii) combine learning from both expert knowledge and available data (via distant supervision), and (iii) facilitate future combination with probabilistic knowledge and events generated from hard data within RUSH [19] in a consistent and coherent manner. To build the maritime knowledge graph in Maritime Deep-Dive, we carry out a few data preprocessing steps.…”

“…To this end, Reasoning under Uncertainty with Soft and Hard data (RUSH) [18], [19] was proposed as a framework for reasoning and learning from diverse sources for real-world situational awareness. With respect to soft data in particular, RUSH is motivated by harnessing human knowledge implicitly embedded in abundant natural language data often existing in many real-world scenarios (e.g., news articles and intelligence reports).…”

Paraphrasing Techniques for Maritime QA system

“…However, there is often no time for personnel to translate input to a machine and interpret complex machine outputs in military operations. In our previous § Equal contribution work, we proposed Maritime DeepDive [34], an automated construction of knowledge graphs from unstructured natural language data sources as part of the RUSH project [22] for situational awareness. We are interested in endowing our situational awareness system with a capability to assist human decision makers by allowing them to interact with the system using human natural language.…”