Techniques for linking data to provide improved searching capabilities

A data management service, comprising:

• one or more processors;
• and
• one or more memories storing computer-executable instructions that, when executed with the one or more processors, cause the data management service to:
• obtain a machine-learning model that is configured to identify a similarity between a first label and a second label provided as input, the machine-learning model being previously trained based at least in part on a supervised learning algorithm and a training data set including pairs of labels that have been previously identified as being valid or invalid pairs, the first label corresponding to technical metadata and the second label corresponding to a domain term;
• generate a (LSH) forest based at least in part on one or more respective n-gram tokens generated from respective domain labels of a plurality of domain labels;
• receive, at a user interface, user input identifying an input label;
• identify a plurality of candidate labels from the LSH forest based at least in part on querying the LSH forest using one or more n-gram tokens generated from the input label, the plurality of candidate labels comprising a subset of domain labels from the plurality of domain labels;
• generate a respective pairing for the input label and each candidate label of the plurality of candidate labels;
• generate a respective feature set for each respective pairing of the input label and each candidate label;
• provide each respective feature set to the machine-learning model as input;
• receive output from the machine-learning model, the output identifying one or more of the candidate labels as one or more recommended labels, the one or more recommended labels being identified by the output as being similar to the input label;
• and
• provide, at the user interface, a recommendation corresponding to a recommendation label of the one or more recommended labels identified by the machine-learning model.

A non-transitory computer-readable storage medium comprising executable instructions that, when executed with one or more processors of a computing device, cause the computing device to:

• obtain a machine-learning model that is configured to identify a similarity between a first label and a second label provided as input, the machine-learning model being previously trained based at least in part on a supervised learning algorithm and a training data set including pairs of labels that have been previously identified as being valid or invalid pairs, the first label corresponding to technical metadata and the second label corresponding to a domain term;
• generate a (LSH) forest based at least in part on one or more respective n-gram tokens generated from respective domain labels of a plurality of domain labels;
• receive, at a user interface, user input identifying an input label;
• identify a plurality of candidate labels from the LSH forest based at least in part on querying the LSH forest using one or more n-gram tokens generated from the input label, the plurality of candidate labels comprising a subset of domain labels from the plurality of domain labels;
• generate a respective pairing for the input label and each candidate label of the plurality of candidate labels;
• generate a respective feature set for each respective pairing of the input label and each candidate label;
• provide each respective feature set to the machine-learning model as input;
• receive output from the machine-learning model, the output identifying one or more of the candidate labels as one or more recommended labels, the one or more recommended labels being identified by the output as being similar to the input label;
• and
• provide, to the user interface, a recommendation corresponding to a recommended label of the one or more recommended labels identified by the machine-learning model.