Extracting enriched target-oriented common sense from grounded graphs to support next step decision making
Aspects of the invention include systems and methods configured to extract enriched target-oriented common sense from grounded graphs to support efficient next step decision making of an autonomous agent. A non-limiting example computer-implemented method includes extracting common sense from a source. The extracted common sense can include a first knowledge graph. An environment state can be extracted from an observation. The extracted environment state can include a second knowledge graph. The second knowledge graph can include an interactive object and a state of the interactive object. A difference graph including the extracted common sense and the extracted environment state can be generated. A next action is selected based on the difference graph and the next action is taken by an autonomous agent.
1 . A computer-implemented method comprising:
extracting, by an autonomous agent, common sense from a source, the extracted common sense comprising a first knowledge graph;
extracting, by the autonomous agent, an environment state from an observation, the extracted environment state comprising a second knowledge graph, the second knowledge graph comprising an interactive object and a state of the interactive object;
generating, by the autonomous agent, a difference graph comprising the extracted common sense and the extracted environment state, the difference graph encoding the extracted common sense and the extracted environment state in a single connected graph;
selecting, by the autonomous agent and based on the difference graph, a next action; and
taking, by the autonomous agent, the next action.
2 . The computer-implemented method of claim 1 , wherein the first knowledge graph of the extracted common sense is represented as a triplet of {subject, relationship, object}.
3 . The computer-implemented method of claim 1 , wherein extracting the common sense comprises extracting by meaning, wherein extracting by meaning comprises extracting a knowledge graph from the source when both a similarity of a subject in a respective common sense graph of the source and a similarity of an object in the respective common sense graph of the source to an entity exceeds a threshold.
4 . The computer-implemented method of claim 3 , wherein extracting the common sense further comprises narrowing by circumstances, wherein narrowing by circumstances comprises retaining only those knowledge graphs from the extracted common sense that represent valid actions.
5 . The computer-implemented method of claim 3 , wherein extracting the common sense further comprises transforming into a grounded representation, wherein transforming into a grounded representation comprises transforming the subject in the respective common sense graph of the source to the respective entity.
6 . The computer-implemented method of claim 1 , wherein generating the difference graph comprises organizing the extracted common sense and the extracted environment states by interactive objects.
7 . The computer-implemented method of claim 6 , wherein the difference graph comprises a plurality of current state nodes, a plurality of common sense nodes, and a single interactive object node, and wherein the difference graph comprises interactive object-current state edges and interactive object-common sense edges.
8 . The computer-implemented method of claim 1 , further comprising encoding the difference graph, wherein encoding the difference graph comprises converting one or more words in a node of the difference graph into a series of vectors by word embedding.
9 . A system comprising an autonomous agent having a memory, computer readable instructions, and one or more processors for executing the computer readable instructions, the computer readable instructions controlling the one or more processors to perform operations comprising:
extracting common sense from a source, the extracted common sense comprising a first knowledge graph;
extracting an environment state from an observation, the extracted environment state comprising a second knowledge graph, the second knowledge graph comprising an interactive object and a state of the interactive object;
generating a difference graph comprising the extracted common sense and the extracted environment state, the difference graph encoding the extracted common sense and the extracted environment state in a single connected graph;
selecting, based on the difference graph, a next action; and
taking, by the autonomous agent, the next action.
10 . The system of claim 9 , wherein the first knowledge graph of the extracted common sense is represented as a triplet of {subject, relationship, object}.
11 . The system of claim 9 , wherein extracting the common sense comprises extracting by meaning, wherein extracting by meaning comprises extracting a knowledge graph from the source when both a similarity of a subject in a respective common sense graph of the source and a similarity of an object in the respective common sense graph of the source to an entity exceeds a threshold.
12 . The system of claim 11 , wherein extracting the common sense further comprises narrowing by circumstances, wherein narrowing by circumstances comprises retaining only those knowledge graphs from the extracted common sense that represent valid actions.
13 . The system of claim 11 , wherein extracting the common sense further comprises transforming into a grounded representation, wherein transforming into a grounded representation comprises transforming the subject in the respective common sense graph of the source to the respective entity.
14 . The system of claim 9 , wherein generating the difference graph comprises organizing the extracted common sense and the extracted environment states by interactive objects.
15 . The system of claim 14 , wherein the difference graph comprises a plurality of current state nodes, a plurality of common sense nodes, and a single interactive object node, and wherein the difference graph comprises interactive object-current state edges and interactive object-common sense edges.
16 . The system of claim 9 , further comprising encoding the difference graph, wherein encoding the difference graph comprises converting one or more words in a node of the difference graph into a series of vectors by word embedding.
17 . A computer program product comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by one or more processors to cause the one or more processors to perform operations comprising:
extracting, by an autonomous agent, common sense from a source, the extracted common sense comprising a first knowledge graph;
extracting, by the autonomous agent, an environment state from an observation, the extracted environment state comprising a second knowledge graph, the second knowledge graph comprising an interactive object and a state of the interactive object;
generating, by the autonomous agent, a difference graph comprising the extracted common sense and the extracted environment state, the difference graph encoding the extracted common sense and the extracted environment state in a single connected graph;
selecting, by the autonomous agent and based on the difference graph, a next action; and
taking, by the autonomous agent, the next action.
18 . The computer program product of claim 17 , wherein the first knowledge graph of the extracted common sense is represented as a triplet of {subject, relationship, object}.
19 . The computer program product of claim 17 , wherein extracting the common sense comprises extracting by meaning, wherein extracting by meaning comprises extracting a knowledge graph from the source when both a similarity of a subject in a respective common sense graph of the source and a similarity of an object in the respective common sense graph of the source to an entity exceeds a threshold.
20 . The computer program product of claim 19 , wherein extracting the common sense further comprises narrowing by circumstances, wherein narrowing by circumstances comprises retaining only those knowledge graphs from the extracted common sense that represent valid actions.