Accurate relationship extraction with word embeddings using minimal training data

1. A processor-implemented method for word embedding relationship extraction, the method comprising:

generating a plurality of smoothed transition vectors from a plurality of training data, wherein the plurality of training data is arranged into a table where each line represents a set of multi part analogies:

receiving a plurality of text and a query;

converting the plurality of received text to a word embedding space; converting the received query to a set of coordinates from the word embedding space and a set of the plurality of determined smoothed transition vectors;

determining a plurality of candidate answers based on adding the set of the smoothed transition vectors to the set of coordinates in the word embedding space, wherein the determining the plurality of candidate answers further comprises: adjusting the plurality of smoothed transition vectors by triangulation; determining a new set of coordinates by applying the adjusted plurality of smoothed transition vectors to the set of coordinates; and converting the determined new set of coordinates to the plurality of candidate answers;

determining an answer to the received query based on applying a filter, wherein the filter is selected from a group consisting of a type filtering, a conflicting type filtering, and an equivalence filtering; and

displaying the determined answer.

2. The method of claim 1 , wherein determining the plurality of smoothed transition vectors from plurality of training data further comprises:

converting the plurality of training data to a plurality of transition vectors using a word embedding technique; and

smoothing the plurality of transition vectors with the plurality of training data.

3. The method of claim 1 , wherein converting the plurality of received text to a word embedding space further comprises:

converting the plurality of received text to a word embedding space by a vectorization of a plurality of words and a plurality of phrases of the plurality of received text to a set of coordinates, wherein each set of coordinates corresponds to the plurality of words and the plurality of phrases.

4. The method of claim 1 , wherein applying a type filtering to the plurality of candidate answers further comprises:

determining a multivariate normal distribution (MVN) value of each of the plurality of candidate answers; and

removing each candidate answer within the plurality of candidate answers based on the determined MVN value associated with a candidate answer not being within a first threshold value.

5. The method of claim 1 , wherein applying a conflicting type filtering to the plurality of candidate answers further comprises:

determining a multivariate normal distribution (MVN) value of each of the plurality of candidate answers; and

removing each candidate answer within the plurality of candidate answers based on a difference between the determined MVN value associated with the candidate answer and an MVN value of a transition vector not being within a second threshold value.

6. The method of claim 1 , wherein applying an equivalence filtering to the plurality of candidate answers further comprises:

determining a multivariate normal distribution (MVN) value by calculating a difference between each candidate answer within the plurality of candidate answers and a transition vector; and

removing each candidate answer within the plurality of candidate answers based on the MVN value of the calculated difference between each candidate answer and the transition vector not being within a third threshold value.

7. A computer system for word embedding relationship extraction, the computer system comprising:

one or more processors, one or more computer-readable memories, one or more computer-readable tangible storage medium, and program instructions stored on at least one of the one or more tangible storage medium for execution by at least one of the one or more processors via at least one of the one or more memories, wherein the computer system is capable of performing a method comprising:

generating a plurality of smoothed transition vectors from a plurality of training data, wherein the plurality of training data is arranged into a table where each line represents a set of

multi part analogies:

receiving a plurality of text and a query;

converting the plurality of received text to a word embedding space; converting the received query to a set of coordinates from the word embedding space and a set of the plurality of determined smoothed transition vectors;

determining a plurality of candidate answers based on adding the set of the smoothed transition vectors to the set of coordinates in the word embedding space, wherein determining the plurality of candidate answers further comprises: adjusting the plurality of smoothed transition vectors by triangulation; determining a new set of coordinates by applying the adjusted plurality of smoothed transition vectors to the set of coordinates; and converting the determined new set of coordinates to the plurality of candidate answers;

determining an answer to the received query based on applying a filter, wherein the filter is selected from a group consisting of a type filtering, a conflicting type filtering, and an equivalence filtering; and

displaying the determined answer.

8. The computer system of claim 7 , wherein determining the plurality of smoothed transition vectors from plurality of training data further comprises:

converting the plurality of training data to a plurality of transition vectors using a word embedding technique; and

smoothing the plurality of transition vectors with the plurality of training data.

9. The computer system of claim 7 , wherein converting the plurality of received text to a word embedding space further comprises:

converting the plurality of received text to a word embedding space by a vectorization of a plurality of words and a plurality of phrases of the plurality of received text to a set of coordinates, wherein each set of coordinates corresponds to the plurality of words and the plurality of phrases.

10. The computer system of claim 7 , wherein applying a type filtering to the plurality of candidate answers further comprises:

determining a multivariate normal distribution (MVN) value of each of the plurality of candidate answers; and

removing each candidate answer within the plurality of candidate answers based on the determined MVN value associated with a candidate answer not being within a first threshold value.

11. The computer system of claim 7 , wherein applying a conflicting type filtering to the plurality of candidate answers further comprises:

determining a multivariate normal distribution (MVN) value of each of the plurality of candidate answers; and

removing each candidate answer within the plurality of candidate answers based on a difference between the determined MVN value associated with the candidate answer and an MVN value of a transition vector not being within a second threshold value.

12. The computer system of claim 7 , wherein applying an equivalence filtering to the plurality of candidate answers further comprises:

determining a multivariate normal distribution (MVN) value by calculating a difference between each candidate answer within the plurality of candidate answers and a transition vector; and

removing each candidate answer within the plurality of candidate answers based on the MVN value of the calculated difference between each candidate answer and the transition vector not being within a third threshold value.

13. A computer program product for word embedding relationship extraction, the computer program product comprising:

one or more computer-readable tangible storage medium and program instructions stored on at least one of the one or more tangible storage medium, the program instructions executable by a processor, the program instructions comprising:

program instructions to generate a plurality of smoothed transition vectors from a plurality of training data, wherein the plurality of training data is arranged into a table where each line represents a set of multi part analogies:

program instructions to receive a plurality of text and a query;

program instructions to convert the plurality of received text to a word embedding space; program instructions to convert the received query to a set of coordinates from the word embedding space and a set of the plurality of determined smoothed transition vectors;

program instructions to determine a plurality of candidate answers based on adding the set of the smoothed transition vectors to the set of coordinates in the word embedding space, wherein program instructions to determine the plurality of candidate answers further comprises: program instructions to adjust the plurality of smoothed transition vectors by triangulation; program instructions to determine a new set of coordinates by applying the adjusted plurality of smoothed transition vectors to the set of coordinates; and program instructions to convert the determined new set of coordinates to the plurality of candidate answers;

program instructions to determine an answer to the received query based on applying a filter, wherein the filter is selected from a group consisting of a type filtering, a conflicting type filtering, and an equivalence filtering; and

program instructions to display the determined answer.

14. The computer program product of claim 13 , wherein program instructions to determine the plurality of smoothed transition vectors from plurality of training data further comprises:

program instructions to convert the plurality of training data to a plurality of transition vectors using a word embedding technique; and

program instructions to smooth the plurality of transition vectors with the plurality of training data.

15. The computer program product of claim 13 , wherein program instructions to convert the plurality of received text to a word embedding space further comprises:

program instructions to convert the plurality of received text to a word embedding space by a vectorization of a plurality of words and a plurality of phrases of the plurality of received text to a set of coordinates, wherein each set of coordinates corresponds to the plurality of words and the plurality of phrases.

16. The computer program product of claim 13 , wherein program instructions to apply a type filtering to the plurality of candidate answers further comprises:

program instructions to determine a multivariate normal distribution (MVN) value of each of the plurality of candidate answers; and

program instructions to remove each candidate answer within the plurality of candidate answers based on the determined MVN value associated with a candidate answer not being within a first threshold value.

17. The computer program product of claim 13 , wherein program instructions to apply a conflicting type filtering to the plurality of candidate answers further comprises:

program instructions to determine a multivariate normal distribution (MVN) value of each of the plurality of candidate answers; and

program instructions to remove each candidate answer within the plurality of candidate answers based on a difference between the determined MVN value associated with the candidate answer and an MVN value of a transition vector not being within a second threshold value.