Systems and methods for learning semantic patterns from textual data
View Patent ↗In some embodiments, a system is provided comprising at least one processor programmed to process an input text to identify a plurality of semantic patterns that match the input text, wherein, for at least one semantic pattern of the plurality of semantic patterns: the at least one semantic pattern comprises a plurality of semantic entities identified from the at least one input text, and the plurality of semantic entities occur in a common context within the at least one input text. The at least one processor may be further programmed to use statistical information derived from training data to associate a respective weight with each semantic pattern of the plurality of semantic patterns.
1. A system comprising at least one processor programmed to:
identify at least one semantic pattern from at least one training text, the at least one semantic pattern comprising a combination of semantic entities occurring in a common context within a portion of the at least one training text; and
generate a weight associated with each of the at least one semantic pattern, wherein:
the weight of a semantic pattern, of the at least one semantic pattern, is indicative of a level of confidence that the semantic pattern accurately represents a meaning of the portion of the at least one training text, and
generating the weight associated with the semantic pattern comprises computing at least one measure of mutual information for the combination of semantic entities of the semantic pattern; and
generate a collection of semantic patterns for subsequent use in analyzing text to determine meaning of the text, the collection of semantic patterns comprising the at least one semantic pattern and the weight associated with the at least one semantic pattern.
2. The system of claim 1 , wherein the at least one measure of mutual information comprises a measure of pointwise mutual information.
3. The system of claim 1 , wherein the at least one measure of mutual information comprises a measure of weighted mutual information.
4. The system of claim 1 , wherein:
the at least one semantic pattern comprises a valency frame;
the combination of semantic entities comprises a plurality of valency frame components of the valency frame; and
the common context in which the plurality of valency frame components occur comprises a valency structure controlled by a controlling valency frame component.
5. The system of claim 4 , wherein:
the plurality of valency frame components further comprise at least one dependent valency frame component depending from the controlling valency frame component.
6. The system of claim 5 , wherein the controlling valency frame component comprises a verb and the at least dependent valency frame component comprises an argument of the verb.
7. The system of claim 6 , wherein the argument of the verb comprises a preposition.
8. The system of claim 1 , wherein:
the collection comprises a plurality of semantic patterns, each semantic pattern being associated with a respective weight; and
the at least one processor is programmed to sort the plurality of semantic patterns into an ordered list of semantic patterns according to the respective weights.
9. The system of claim 1 , wherein each semantic entity of the combination of semantic entities comprises a respective word tuple and an annotation associated with the respective word tuple, and wherein the at least one processor is programmed to:
process the at least one training text to identify the word tuples of the combination of semantic entities; and
construct each semantic entity at least in part by associating the respective word tuple with the annotation associated with the respective word tuple.
10. The system of claim 9 , wherein the annotation associated with the word tuple of the at least one semantic pattern comprises an indication of a part of speech of the word tuple within the at least one training text.
11. A computer-implemented method comprising acts of:
identifying at least one semantic pattern from at least one training text at least one semantic pattern comprising a combination of semantic entities occurring in a common context within a portion of the at least one training text;
generating a weight associated with each of the at least one semantic pattern, wherein:
the weight of a semantic pattern, of the at least one semantic pattern, is indicative of a level of confidence that the semantic pattern accurately represents a meaning of the portion of the at least one training text, and
generating the weight associated with the semantic pattern comprises computing at least one measure of mutual information for the combination of semantic entities of the semantic pattern; and
generating a collection of semantic patterns for subsequent use in analyzing text to determine meaning of the text, the collection of semantic patterns comprising the at least one semantic pattern and the weight associated with the at least one semantic pattern.
12. The method of claim 11 , wherein the at least one measure of mutual information comprises a measure of pointwise mutual information.
13. The method of claim 11 , wherein the at least one measure of mutual information comprises a measure of weighted mutual information.
14. The method of claim 11 , wherein:
the at least one semantic pattern comprises a valency frame;
the combination of semantic entities comprises a plurality of valency frame components of the valency frame; and
the common context in which the plurality of valency frame components occur comprises a valency structure controlled by a controlling valency frame component.
15. The method of claim 14 , wherein:
the plurality of valency frame components further comprise at least one dependent valency frame component depending from the controlling valency frame component.
16. The method of claim 15 , wherein the controlling valency frame component comprises a verb and the at least dependent valency frame component comprises an argument of the verb.
17. The method of claim 16 , wherein the argument of the verb comprises a preposition.
18. The method of claim 11 , wherein:
the collection comprises a plurality of semantic patterns, each semantic pattern being associated with a respective weight; and
the method comprises an act of sorting the plurality of semantic patterns into an ordered list of semantic patterns according to the respective weights.
19. The method of claim 11 , wherein each semantic entity of the combination of semantic entities comprises a respective word tuple and an annotation associated with the respective word tuple, and wherein the method comprises acts of:
processing the at least one training text to identify the word tuples of the combination of semantic entities; and
constructing each semantic entity at least in part by associating the respective word tuple with the annotation associated with the respective word tuple.
20. The method of claim 19 , wherein the annotation associated with the word tuple of the at least one semantic pattern comprises an indication of a part of speech of the word tuple within the at least one training text.
21. At least one non-transitory computer-readable storage medium having stored thereon instructions which, when executed, cause at least one processor to perform a method comprising acts of:
identifying at least one semantic pattern from at least one training text, the at least one semantic pattern comprising a combination of semantic entities occurring in a common context within a portion of the at least one training text;
generating a weight associated with each of the at least one semantic pattern, wherein:
the weight of a semantic pattern, of the at least one semantic pattern, is indicative of a level of confidence that the semantic pattern accurately represents a meaning of the portion of the at least one training text, and
generating the weight associated with the semantic pattern comprises computing at least one measure of mutual information for the combination of semantic entities of the semantic pattern; and
generating a collection of semantic patterns for subsequent use in analyzing text to determine meaning of the text, the collection of semantic patterns comprising the at least one semantic pattern and the weight associated with the at least one semantic pattern.