A Bayesian Abduction Model For Sensemaking

This research develops a Bayesian Abduction Model for Sensemaking Support (BAMSS) for information fusion in sensemaking tasks. Two methods are investigated. The first is the classical Bayesian information fusion with belief updating (using Bayesian clustering algorithm) and abductive inference. The second method uses a Genetic Algorithm (BAMSS-GA) to search for the k-best most probable explanation (MPE) in the network. Using various data from recent Iraq and Afghanistan conflicts, experimental simulations were conducted to compare the methods using posterior probability values which can be used to give insightful information for prospective sensemaking. The inference results demonstrate the utility of BAMSS as a computational model for sensemaking. The major results obtained are: (1) The inference results from BAMSS-GA gave average posterior probabilities that were 103 better than those produced by BAMSS; (2) BAMSS-GA gave more consistent posterior probabilities as measured by variances; and (3) BAMSS was able to give an MPE while BAMSS-GA was able to identify the optimal values for kMPEs. In the experiments, out of 20 MPEs generated by BAMSS, BAMSS-GA was able to identify 7 plausible network solutions resulting in less amount of information needed for sensemaking and reducing the inference search space by 7/20 (35%). The results reveal that GA can be used successfully in Bayesian information fusion as a search technique to identify those significant posterior probabilities useful for sensemaking. BAMSS-GA was also more robust in overcoming the problem of bounded search that is a constraint to Bayesian clustering and inference state space in BAMSS