Methods, mediums, and systems for generating causal inference structure between concepts having predictive capabilities
A machine learning (ML) system is provided for integrating common sense into the ML system. In contrast to existing machine learning algorithms that search for statistical correlations between concepts, the ML system is configured to learn the semantic relationships or causality between the concepts. This may be accomplished by training an algorithm or data structure to learn similar vector representations of words present in the same context (e.g., that are present together in the same sentence). The resulting AI/ML structure may be used to guide the generation of a causal graph having predictive capabilities. This causal graph may represent semantic relationships and/or causation between concepts, and hence may be employed to introduce a degree of common sense in the machine learning system.
1. An apparatus, comprising:
a processor; and
a memory comprising instructions that when executed by the processor cause the processor to:
analyze textual input data to determine a causal relationship between a first word in the textual input data and a second word in the textual input data based on semantics of the textual input data, wherein determining the causal relationship comprises mapping the first word to a first concept using an unsupervised artificial neural network by learning a first vector representation for the first word; mapping the second word to a second concept using the unsupervised artificial neural network by learning a second vector representation for the second word; generating a hypothesis comprising the first concept and the second concept, the hypothesis representing a prediction that the first concept is dependent upon the second concept; defining a probability distribution describing a likelihood of observing the first concept given the second concept; and assigning a coefficient to the second concept based on an amount of contribution of the second concept to the first concept, the coefficient for the second concept comprising a vector in a semantic space;
determine a weight for a connection in a causal inference structure, the connection between the first and second words in the textual input data based on the causal relationship determined between the first and second words; and
generate the causal inference structure comprising the first word, the second word, and the connection with the weight between the first and second words.
2. The apparatus of claim 1 , the memory comprising instructions that when executed by the processor cause the processor to identify a semantic relationship between the first and second words in the textual input data to determine the causal relationship between the first and second words.
3. The apparatus of claim 2 , wherein the semantic relationship comprises a frequency with which the first and second words are present in a same sentence of the textual input data.
4. The apparatus of claim 2 , wherein the weight of the connection between the first and second words indicates a strength of the semantic relationship between the first and second words.
5. The apparatus of claim 1 , the memory comprising instructions that when executed by the processor cause the processor to select, using the causal inference structure, words in a message generated in response to user input including an observed concept comprising the first word based, at least in part, on a prediction as to which cause concepts caused the observed concept, wherein at least one of the cause concepts comprises the second word.
6. The apparatus of claim 1 , wherein at least a portion of the textual input data is unlabeled.
7. The apparatus of claim 1 , the memory comprising instructions that when executed by the processor cause the processor to:
identify the first word and the second word in user input; and
output a prediction based on the causal inference structure, the prediction to indicate the second word contributes to a context associated with the first word.
8. The apparatus of claim 1 , wherein the causal inference structure comprises a causal graph.
9. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the one or more processors to:
analyze textual input data to determine a causal relationship between a first word in the textual input data and a second word in the textual input data based on semantics of the textual input data, wherein determining the causal relationship comprises mapping the first word to a first concept using an unsupervised artificial neural network by learning a first vector representation for the first word; mapping the second word to a second concept using the unsupervised artificial neural network by learning a second vector representation for the second word; generating a hypothesis comprising the first concept and the second concept, the hypothesis representing a prediction that the first concept is dependent upon the second concept; defining a probability distribution describing a likelihood of observing the first concept given the second concept; and assigning a coefficient to the second concept based on an amount of contribution of the second concept to the first concept, the coefficient for the second concept comprising a vector in a semantic space;
determine a weight for a connection in a causal inference structure, the connection between the first and second words in the textual input data based on the causal relationship determined between the first and second words; and
generate the causal inference structure comprising the first word, the second word, and the connection with the weight between the first and second words.
10. The non-transitory computer-readable medium of claim 9 , comprising instructions that, when executed by the one or more processors, cause the one or more processors to identify a semantic relationship between the first and second words in the textual input data to determine the causal relationship between the first and second words.
11. The non-transitory computer-readable medium of claim 10 , wherein the semantic relationship comprises a frequency with which the first and second words are present in a same sentence of the textual input data.
12. The non-transitory computer-readable medium of claim 10 , wherein the weight of the connection between the first and second words indicates a strength of the semantic relationship between the first and second words.
13. The non-transitory computer-readable medium of claim 9 , comprising instructions that, when executed by the one or more processors, cause the one or more processors to select, using the causal inference structure, words in a message generated in response to user input including an observed concept comprising the first word based, at least in part, on a prediction as to which cause concepts caused the observed concept, wherein at least one of the cause concepts comprises the second word.
14. The non-transitory computer-readable medium of claim 9 , wherein at least a portion of the textual input data is unlabeled.
15. The non-transitory computer-readable medium of claim 9 , comprising instructions that, when executed by the one or more processors, cause the one or more processors to:
identify the first word and the second word in user input; and
output a prediction based on the causal inference structure, the prediction to indicate the second word contributes to a context associated with the first word.
16. The non-transitory computer-readable medium of claim 9 , wherein the causal inference structure comprises a causal graph.
17. A computer-implemented method, comprising:
analyzing textual input data to determine a causal relationship between a first word in the textual input data and a second word in the textual input data based on semantics of the textual input data, wherein determining the causal relationship comprises mapping the first word to a first concept using an unsupervised artificial neural network by learning a first vector representation for the first word; mapping the second word to a second concept using the unsupervised artificial neural network by learning a second vector representation for the second word; generating a hypothesis comprising the first concept and the second concept, the hypothesis representing a prediction that the first concept is dependent upon the second concept; defining a probability distribution describing a likelihood of observing the first concept given the second concept; and assigning a coefficient to the second concept based on an amount of contribution of the second concept to the first concept, the coefficient for the second concept comprising a vector in a semantic space;
determining a weight for a connection in a causal inference structure, the connection between the first and second words in the textual input data based on the causal relationship determined between the first and second words; and
generating the causal inference structure comprising the first word, the second word, and the connection with the weight between the first and second words.
18. The computer-implemented method of claim 17 , comprising identifying a semantic relationship between the first and second words in the textual input data to determine the causal relationship between the first and second words.
19. The computer-implemented method of claim 17 , comprising selecting, using the causal inference structure, words in a message generated in response to user input including an observed concept comprising the first word based, at least in part, on a prediction as to which cause concepts caused the observed concept, wherein at least one of the cause concepts comprises the second word.
20. The computer-implemented method of claim 17 , comprising:
identifying the first word and the second word in user input; and
outputting a prediction based on the causal inference structure, the prediction indicating the second word contributes to a context associated with the first word.