Understanding a query intention for medical artificial intelligence systems using semi-supervised deep learning

1. A method, comprising:

receiving, by a device, a query including a sequence of words;

converting, by the device, the sequence of words into a sequence of feature vectors;

generating, by the device and using a bi-directional long-short term memory (LSTM) followed by an attention mechanism and a softmax layer, a semantic meaning of the sequence of feature vectors; and

regularizing, by the device and using a Gaussian mixed model (GMM), the LSTM based on the generated semantic meaning of the sequence of feature vectors,

wherein the regularizing the LSTM comprises classifying, by the GMM, labeled parts of the feature vectors based on minimizing a softmax cross entropy loss between predicted probabilities and true labels, and minimizing a negative log likelihood of the GMM.

2. The method of claim 1 , wherein the converting includes at least one of a concatenating technique, a word embedding technique, and a position embedding technique.

3. The method of claim 1 , wherein each feature vector of the sequence of feature vectors is a concatenation of a word feature and a position feature.

4. The method of claim 3 , wherein the word feature corresponds to location of a word in a dictionary, and wherein the position feature corresponds to a position of the word in the sequence of words.

5. The method of claim 1 , wherein the generating comprises generating the semantic meaning based on a relationship between the words of the sequence of words.

6. The method of claim 1 , further comprising training the GMM by using an attention mechanism layer.

7. A device, comprising:

at least one memory configured to store program code;

at least one processor configured to read the program code and operate as instructed by the program code, the program code including:

receiving code configured to cause the at least one processor to receive a query including a sequence of words;

converting code configured to cause the at least one processor to convert the sequence of words into a sequence of feature vectors;

generating code configured to cause the at least one processor to generate, using a bi-directional long-short term memory (LSTM) followed by an attention mechanism and a softmax layer, a semantic meaning of the sequence of feature vectors; and

regularizing code configured to cause the at least one processor to regularize, using a Gaussian mixed model (GMM), the LSTM based on the generated semantic meaning of the sequence of feature vectors,

wherein the regularizing code is configured to cause the at least one processor to regularize the LSTM by classifying, by the GMM, labeled parts of the feature vectors based on minimizing a softmax cross entropy loss between predicted probabilities and true labels, and minimizing a negative log likelihood of the GMM.

8. The device of claim 7 , wherein the converting code is further configured to cause the at least one processor to convert the sequence of words using at least one of a concatenating technique, a word embedding technique, and a position embedding technique.

9. The device of claim 7 , wherein each feature vector of the sequence of feature vectors is a concatenation of a word feature and a position feature.

10. The device of claim 9 , wherein the word feature corresponds to location of a word in a dictionary, and wherein the position feature corresponds to a position of the word in the sequence of words.

11. The device of claim 7 , wherein the generating code is further configured to cause the at least one processor to generate the semantic meaning based on a relationship between the words of the sequence of words.

12. The device of claim 7 , further comprising:

training code configured to cause the at least one processor to train the GMM using an attention mechanism layer.

13. A non-transitory computer-readable medium storing instructions, the instructions comprising: one or more instructions that, when executed by one or more processors of a device, cause the one or more processors to:

receive a query including a sequence of words;

convert the sequence of words into a sequence of feature vectors;

generate, using a bi-directional long-short term memory (LSTM) followed by an attention mechanism and a softmax layer, a semantic meaning of the sequence of feature vectors; and

regularize, using a Gaussian mixed model (GMM), the LSTM based on the generated semantic meaning of the sequence of feature vectors,

wherein the one or more processors of the device are caused to regularize the LSTM by classifying, by the GMM, labeled parts of the feature vectors based on minimizing a softmax cross entropy loss between predicted probabilities and true labels, and minimizing a negative log likelihood of the GMM.

14. The non-transitory computer-readable medium of claim 13 , wherein the converting includes at least one of a concatenating technique, a word embedding technique, and a position embedding technique.

15. The non-transitory computer-readable medium of claim 13 , wherein each feature vector of the sequence of feature vectors is a concatenation of a word feature and a position feature.

16. The non-transitory computer-readable medium of claim 15 , wherein the word feature corresponds to location of a word in a dictionary, and wherein the position feature corresponds to a position of the word in the sequence of words.

17. The non-transitory computer-readable medium of claim 13 , wherein the one or more instructions cause the one or more processors to generate the semantic meaning based on a relationship between the words of the sequence of words.

18. The non-transitory computer-readable medium of claim 13 , wherein the one or more instructions further cause the one or more processors to train the GMM by using an attention mechanism layer.