Capturing rich response relationships with small-data neural networks
View Patent ↗The present disclosure relates to a response analysis system that employs a small-data training dataset to train a neural network that accurately performs domain-agnostic opinion mining. For example, in one or more embodiments, the response analysis system trains a response classification neural network using part of speech information (e.g., syntactic information) to learn and apply response classification labels for opinion text responses. In particular, the response analysis system employs part of speech information patterns without regard to word patterns to determine whether words in a text response correspond to an opinion, the target of the opinion, or neither. In addition, the trained response classification neural network has a significantly reduced learned parameter space, which decreases processing, memory requirements, and overall complexity.
1. A system comprising:
at least one processor; and
at least one non-transitory computer-readable storage medium storing instructions that, when executed by the at least one processor, cause the system to:
determine a part of speech for each word within an input sentence;
determine a response classification label for each word within the input sentence based on the determined parts of speech for each word;
identify a target portion of the input sentence and an opinion portion of the input sentence based on the response classification label for each word within the input sentence; and
provide, to a client device associated with a user, the target portion and the opinion portion of the input sentence.
2. The system of claim 1 , wherein the input sentence is part of a free-form text survey response.
3. The system of claim 1 , wherein the response classification label comprises one of a target classification label, an opinion classification label, or a neither classification label.
4. The system of claim 3 , wherein:
the target portion comprises at least one word within the input sentence corresponding to the target classification label; and
the opinion portion comprises at least one word within the input sentence corresponding to the opinion classification label.
5. The system of claim 1 , further comprising instructions that, when executed by the at least one processor, cause the system to provide the input sentence to a response classification neural network, wherein the response classification neural network determines the response classification label for each word within the input sentence based in part on the part of speech for each word.
6. The system of claim 1 , further comprising instructions that, when executed by the at least one processor, cause the system to parse the input sentence to identify the part of speech for each word within the input sentence.
7. The system of claim 1 , wherein the instructions, when executed by the at least one processor, cause the system to determine the response classification label for each word within the input sentence based on the determined parts of speech for each word and further based on one or more adjacent words.
8. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computing device to:
determine a part of speech for each word within an input sentence;
determine a response classification label for each word within the input sentence based on the determined parts of speech for each word;
identify a target portion of the input sentence and an opinion portion of the input sentence based on the response classification label for each word within the input sentence; and
provide, to a client device associated with a user, the target portion and the opinion portion of the input sentence.
9. The non-transitory computer-readable medium of claim 8 , wherein the input sentence is part of a free-form text survey response.
10. The non-transitory computer-readable medium of claim 8 , wherein the response classification label comprises one of a target classification label, an opinion classification label, or a neither classification label.
11. The non-transitory computer-readable medium of claim 10 , wherein:
the target portion comprises at least one word within the input sentence corresponding to the target classification label; and
the opinion portion comprises at least one word within the input sentence corresponding to the opinion classification label.
12. The non-transitory computer-readable medium of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the computing device to provide the input sentence to a response classification neural network, wherein the response classification neural network determines the response classification label for each word within the input sentence based in part on the part of speech for each word.
13. The non-transitory computer-readable medium of claim 8 , further comprising instructions that, when executed by the at least one processor, cause the computing device to parse the input sentence to identify the part of speech for each word within the input sentence.
14. A method comprising:
determining, by at least one processor, a part of speech for each word within an input sentence;
determining, by at least one processor, a response classification label for each word within the input sentence based on the determined parts of speech for each word;
identifying, by at least one processor, a target portion of the input sentence and an opinion portion of the input sentence based on the response classification label for each word within the input sentence; and
providing, to a client device associated with a user, the target portion and the opinion portion of the input sentence.
15. The method of claim 14 , wherein the input sentence is part of a free-form text survey response.
16. The method of claim 14 , wherein the response classification label comprises one of a target classification label, an opinion classification label, or a neither classification label.
17. The method of claim 16 , wherein:
the target portion comprises at least one word within the input sentence corresponding to the target classification label; and
the opinion portion comprises at least one word within the input sentence corresponding to the opinion classification label.
18. The method of claim 14 , further comprising providing the input sentence to a response classification neural network, wherein the response classification neural network determines the response classification label for each word within the input sentence based in part on the part of speech for each word.
19. The method of claim 14 , further comprising parsing the input sentence to identify the part of speech for each word within the input sentence.
20. The method of claim 14 , further comprising associating a part of speech label with each word within the input sentence based on the part of speech identified for each word within the input sentence.