Learning parameters in a feed forward probabilistic graphical model
View Patent ↗According to an aspect, learning parameters in a feed forward probabilistic graphical model includes creating an inference model via a computer processor. The creation of the inference model includes receiving a training set that includes multiple scenarios, each scenario comprised of one or more natural language statements, and each scenario corresponding to a plurality of candidate answers. The creation also includes constructing evidence graphs for each of the multiple scenarios based on the training set, and calculating weights for common features across the evidence graphs that will maximize a probability of the inference model locating correct answers from corresponding candidate answers across all of the multiple scenarios. In response to an inquiry from a user that includes a scenario, the inference model constructs an evidence graph and recursively constructs formulas to express a confidence of each node in the evidence graph in terms of its parents in the evidence graph.
1. A computer program product comprising:
a tangible non-transitory storage medium readable by a processing circuit and storing instructions for execution by the processing circuit to perform a method comprising:
creating an inference model, comprising:
receiving a training set that includes multiple scenarios, each scenario comprising one or more natural language statements, and each scenario corresponding to a plurality of candidate answers;
constructing evidence graphs for each of the multiple scenarios based on the training set, wherein constructing the evidence graphs comprises for each scenario;
extracting factors from the scenario; and
generating intermediate nodes based on the extracted factors,
wherein the factors are root nodes in the evidence graphs and the candidate answers are terminal nodes in the evidence graphs; and
calculating weights for common features across the evidence graphs that will maximize a probability of the inference model locating correct answers from corresponding candidate answers across all of the multiple scenarios, wherein calculating the weights comprises assigning a highest weight to a factor having a largest number of unique terms relative to a selected domain; and
in response to an inquiry from a user, the inquiry comprising a scenario, constructing, using the inference model, an evidence graph and recursively constructing formulas to express a confidence of each node in the evidence graph in terms of corresponding parents in the evidence graph.
2. The computer program product of claim 1 , wherein the constructing evidence graphs further includes generating questions for the factors and the intermediate nodes represent the generated questions.
3. The computer program product of claim 1 , wherein edges coming out of the intermediate nodes are determined using a question answering (QA) system that assigns a confidence value and a feature vector to each edge.
4. The computer program product of claim 1 , wherein the candidate answers are expressed as a parameterized mathematical expression that matches semantics of the inference model.
5. The computer program product of claim 1 , wherein the inference model applies different weights to the factors.
6. A system comprising:
a memory having computer readable computer instructions; and
a processor for executing the computer readable instructions to perform steps including:
creating an inference model, comprising:
receiving a training set that includes multiple scenarios, each scenario comprising one or more natural language statements, and each scenario corresponding to a plurality of candidate answers;
constructing evidence graphs for each of the multiple scenarios based on the training set, wherein constructing the evidence graphs comprises for each scenario;
extracting factors from the scenario; and
generating intermediate nodes based on the extracted factors,
wherein the factors are root nodes in the evidence graphs and the candidate answers are terminal nodes in the evidence graphs; and
calculating weights for common features across the evidence graphs that will maximize a probability of the inference model locating correct answers from corresponding candidate answers across all of the multiple scenarios, wherein calculating the weights comprises assigning a highest weight to a factor having a largest number of unique terms relative to a selected domain; and
in response to an inquiry from a user, the inquiry comprising a scenario, constructing, using the inference model, an evidence graph and recursively constructing formulas to express a confidence of each node in the evidence graph in terms of corresponding parents in the evidence graph.
7. The system of claim 6 , wherein the constructing evidence graphs further includes generating questions for the factors and the intermediate nodes represent the generated questions.
8. The system of claim 6 , wherein edges coming out of the intermediate nodes are determined using a question answering (QA) system that assigns a confidence value and a feature vector to each edge.
9. The system of claim 6 , wherein the candidate answers are expressed as a parameterized mathematical expression that matches semantics of the inference model.
10. The system of claim 6 , wherein the inference model applies different weights to the factors.
11. The system of claim 6 , wherein the scenario corresponds to a medical environment and the candidate answers indicate corresponding diagnoses.