Method and system for supporting inductive reasoning queries over multi-modal data from relational databases

1. A method of training with multi-modal data in a relational database, comprising:

generating a first database including a multi-view of the multi-modal data;

retrieving a second set of data from an external source; and

training a first model according the first database and the second set of data;

wherein the first model outputs relationships of the first database with the multi-view and the second set of data.

2. The method of claim 1 , wherein the multi-modal data includes text and image information.

3. The method of claim 1 , wherein the training of the first model comprises:

converting structured data source to unstructured text; and

processing the unstructured text to build an unsupervised neural network model including word embedding.

4. The method of claim 1 , wherein the generating of the first database further comprising generating the multi-view including a dual view of the relational data and identified text.

5. The method of claim 1 , further comprising generating vectors to apply word embedding to a string of tokens generated from the first database.

6. The method of claim 1 , wherein the generating of the vectors includes learning on text transformed and extracted from the first database or the external source.

7. The method of claim 1 , wherein the training the first model includes the word embedding to infer latent semantic information in terms of token associations and co-occurrences and encoding to vectors.

8. A method of cognitive learning, comprising:

receiving modal types of information as input;

processing a relational table with the modal types of data as the input to output an unstructured text corpus in a data preparation stage; and

machine learning a first model from the processing of the relational table and the unstructured text corpus in a training process.

9. The method of cognitive learning of claim 8 , wherein the processing the relational table comprises:

building a second model that also captures relational column names in a pre-processing stage; and

processing each relational row separately and for each column of the relational table to converts data of different data types to the unstructured text corpus in a textification stage.

10. The method of claim 8 , wherein the processing of the relational table with different modal types of information includes text tokens, numeric values and images being replaced with the unstructured text corpus.

11. The method of claim 8 , wherein the machine learning of the first model includes generating a uniform semantic representation of the different model types of data.

12. The method of claim 8 , wherein the machine learning of the first model includes classifying the different model types of information.

13. The method of claim 8 , wherein the machine learning of the first model includes clustering the different model types of information.

14. The method of claim 8 , wherein the machine learning of the first model includes representing token the different model types of information as a string token.

15. A method of cognitive learning, comprising:

receiving relational variables as input;

processing the relational variables to create tokens used for training; and

generating similarity values measuring relationships between input parameters of the input.

16. The method according to claim 15 , further comprising:

extracting name of relational variable to obtain corresponding text tokens for training; and

once the training tokens are extracted, fetching corresponding meaning vectors from a pre-trained model.

17. The method according to claim 16 , further comprising:

generating a word-embedding model from the processing of the relational variables; and

executing nearest neighbor computations using the vectors from a word-embedding model.

18. The method according to claim 15 , wherein the relational variables are formed as a table to output tokenized relations of the input, and

wherein the generating of similarity values measures semantic relationships between input parameters of the input.

19. The method according to claim 15 , further comprising:

generating learned vectors from the external text sources and the tokenized relations; and

sending the learned vectors and pre-computed vectors to a table.

20. The method according to claim 15 , further comprising executing queries based on the similarity values of the semantic relationships.