Fuzzy search of graph database
A method may compiling, using at least one processor, a corpus of training data by obtaining, for each respective concept object in an ontology, a respective concept label and respective annotations describing the respective concept object; generating a vocabulary of terms based on the corpus of training data; training a semantic model using the corpus of training data and the vocabulary of terms, wherein input features for the semantic model are based on context words in proximity to a term in the vocabulary of terms; and storing a set of word embeddings for the vocabulary of terms based on the trained semantic model.
1. A computer-implemented method comprising:
compiling, using at least one processor, a corpus of training data by obtaining, for each respective concept object in an ontology, a respective concept label and respective annotations describing the respective concept object;
generating a vocabulary of terms based on the corpus of training data;
training a semantic model using the corpus of training data and the vocabulary of terms, wherein input features for the semantic model are based on context words in proximity to a term in the vocabulary of terms,
wherein the semantic model is a neural network and wherein an input layer of the neural network and an output layer of the neural network are of a size equivalent to a number terms in the vocabulary of terms; and
wherein training the semantic model includes using word pairs from the corpus of training data, a first word of a word pair being an input and a second word of the word pair being an expected output; and
storing a set of word embeddings for the vocabulary of terms based on the trained semantic model.
2. The computer-implemented method of claim 1 , wherein the neural network includes a hidden layer of neurons with no activation function.
3. The computer-implemented method of claim 2 , wherein each respective row of a weight matrix between the input layer and the single hidden layer of neurons is a respective word embedding of the set of word embeddings for a respective term in the vocabulary of terms.
4. The computer-implemented method of claim 1 , wherein the respective annotations include a definition of the respective concept object.
5. The computer-implemented method of claim 1 , wherein the respective annotations include a URL with a definition of the respective concept object and wherein compiling the corpus of training data includes scraping the definition from the URL.
6. The computer-implemented method of claim 1 , wherein compiling the corpus of training data includes, for the respective concept object, retrieving a set of column name identifiers from a set of relational databases that have been identified as a data sources for the respective concept object.
7. The computer-implemented method of claim 1 , wherein compiling the corpus of training data includes, for the respective concept object, retrieving data associated with a concept object related to the respective concept object.
8. The computer-implemented method of claim 7 , wherein the concept object related to the respective concept object is a superclass concept of the respective concept object.
9. A computer-implemented method comprising:
receiving, at a processing computing system, a search query from a user;
querying a data store for a search result using the search query as an input, the data store including a plurality of concept objects for an ontology;
in response to the search result indicating a concept object for the input is not in the data store:
determining a set of result concept objects for the search query based on a closeness of the search query to the plurality of concept objects in the data store, wherein closeness is determined according to word embeddings from a semantic model trained using data based on the ontology, wherein the semantic model is a neural network and wherein an input laver of the neural network and an output layer of the neural network are of a size equivalent to a number terms in a vocabulary of terms;
presenting the set of result concept objects to the user; and
receiving a selection of a concept object of the set of result concept objects from the user; and
presenting data retrieved from the data store for the selected concept object.
10. A non-transitory computer-readable medium comprising instructions, which when executed by at least one processor, configure the at least one processor to perform operations comprising:
compiling, using the at least one processor, a corpus of training data by obtaining, for each respective concept object in an ontology, a respective concept label and respective annotations describing the respective concept object;
generating a vocabulary of terms based on the corpus of training data;
training a semantic model using the corpus of training data and the vocabulary of terms, wherein input features for the semantic model are based on context words in proximity to a term in the vocabulary of terms,
wherein the semantic model is a neural network and wherein an input layer of the neural network and an output layer of the neural network are of a size equivalent to a number terms in the vocabulary of terms; and
wherein training the semantic model includes using word pairs from the corpus of training data, a first word of a word pair being an input and a second word of the word pair being an expected output; and
storing a set of word embeddings for the vocabulary of terms based on the trained semantic model.
11. The computer-readable medium of claim 10 , wherein the neural network includes a hidden layer of neurons with no activation function.
12. The computer-readable medium of claim 11 , wherein each respective row of a weight matrix between the input layer and the single hidden layer of neurons is a respective word embedding of the set of word embeddings for a respective term in the vocabulary of terms.
13. The computer-readable medium of claim 10 , wherein the respective annotations include a definition of the respective concept object.
14. A system comprising:
at least one processor; and
a storage device comprising instructions, which when executed by the at least one processor, configure the system to:
compile a corpus of training data by obtaining, for each respective concept object in an ontology, a respective concept label and respective annotations describing the respective concept object;
generate a vocabulary of terms based on the corpus of training data;
train a semantic model using the corpus of training data and the vocabulary of terms, wherein input features for the semantic model are based on context words in proximity to a term in the vocabulary of terms,
wherein the semantic model is a neural network and wherein an input layer of the neural network and an output layer of the neural network are of a size equivalent to a number terms in the vocabulary of terms; and
wherein to train the semantic model, word pairs are used from the corpus of training data, a first word of a word pair being an input and a second word of the word pair being an expected output; and
store a set of word embeddings for the vocabulary of terms based on the trained semantic model.
15. The system of claim 14 , wherein the neural network includes a hidden layer of neurons with no activation function.
16. The system of claim 15 , wherein each respective row of a weight matrix between the input layer and the hidden layer of neurons is a respective word embedding of the set of word embeddings for a respective term in the vocabulary of terms.
17. The system of claim 14 , wherein the respective annotations include a definition of the respective concept object.