Natural language query processing
Techniques for handling natural language query processing are described. In some examples, a sequence-to-sequence model is used to handle a natural language query. Post-processing of a result of the sequence-to-sequence model utilizes fine-grained information from an entity linker. In some examples, the sequence-to-sequence model and aspects of a natural language query pipeline are used to handle a natural language query.
1. A computer-implemented method comprising:
receiving a natural language query (NLQ);
performing a lexical query to retrieve a set of candidate datasets for the NLQ;
performing named entity recognition (NER) on the NLQ to identify a set of named entities;
performing named entity linking (NEL) to link the set of named entities to corresponding columns or cells in the set of candidate datasets;
based on performing the NEL, selecting a set of top-K candidate datasets from the set of candidate datasets;
generating an intent representation (IR) using a sequence-to-sequence (S2S) machine learning model, the NLQ, and the set of top-K candidate datasets; and
generating a visualization for the intent representation (IR).
2. The computer-implemented method of claim 1 further comprising performing a lexical query to retrieve metadata for the NLQ, wherein the metadata comprises one or more of sorted column names, sorted named entities, sorted named expressions, and sorted named filters.
3. The computer-implemented method of claim 1 , wherein the generated IR comprises the IR and additional data comprising one or more of a metric, a group by listing, a filter, and an indication of a type of visualization to use.
4. A computer-implemented method comprising:
receiving a natural language query (NLQ);
performing a lexical query to retrieve metadata for the NLQ, wherein the metadata is associated with one or more datasets;
generating an intent representation (IR) using a sequence-to-sequence (S2S) machine learning model, wherein input to the S2S machine-learning model includes at least a portion of the retrieved metadata as a schema and the NLQ, and wherein output from the S2S machine-learning model includes the IR;
performing named entity recognition (NER) on the NLQ to identify a set of named entities in the NLQ;
performing named entity linking (NEL) to output fine-grained information, wherein the fine-grained information comprises information for linking one or more entities of the set of named entities to corresponding columns or cells of one or more of the datasets;
post-processing the IR using the fine-grained information by revising at least a subset of data within the IR, re-organizing a structure of the IR, or selecting a final dataset to use from the one or more datasets from an entity linker; and
generating a visualization for the post-processed IR.
5. The computer-implemented method of claim 4 , wherein the S2S machine-learning model is a Transformer-based model.
6. The computer-implemented method of claim 4 , further comprising:
receiving user input from a graphical user interface; and
generating a second IR using the S2S machine learning model based at least in part on the received user input.
7. The computer-implemented method of claim 4 , wherein the schema comprises one or more of sorted column names, sorted named entities, sorted named expressions, and sorted named filters.
8. The computer-implemented method of claim 4 , wherein the generated IR comprises the IR and additional data comprising one or more of a metric, a group by listing, a filter, and an indication of a type of visualization to use.
9. The computer-implemented method of claim 8 , wherein the filter has a particular format.
10. The computer-implemented method of claim 4 , wherein the lexical query is performed by a service of a provider network.
11. The computer-implemented method of claim 4 , further comprising:
processing the natural language query through a query language processing pipeline by:
recognizing one or more entities in the natural language query;
filtering respective candidate linkages for the one or more entities in the natural language query; and
ranking the filtered candidate linkages for the one or more entities in the natural language query.
12. The computer-implemented method of claim 11 , further comprising:
selecting between the intent representation (IR) and the query language processing pipeline.
13. The computer-implemented method of claim 4 , wherein the metadata comprises at least columns that match the one or more entities recognized in the natural language query.
14. The computer-implemented method of claim 13 , wherein the metadata further comprises cell information.
15. A system comprising:
a first one or more electronic devices to implement a database storage service in a multi-tenant provider network; and
a second one or more electronic devices to implement a natural language query processing service in the multi-tenant provider network, the natural language query processing service including instructions that upon execution cause the natural language query processing service to:
receive a natural language query (NLQ);
perform a lexical query to retrieve metadata for the NLQ from the database storage service, wherein the metadata is associated with one or more datasets;
generate an intent representation (IR) using a sequence-to-sequence (S2S) machine learning model, wherein input to the S2S machine-learning model includes at least a portion of the retrieved metadata as a schema and the NLQ, and wherein output from the S2S machine-learning model includes the IR;
perform named entity recognition (NER) on the NLQ to identify a set of named entities in the NLQ;
perform named entity linking (NEL) to output fine-grained information, wherein the fine-grained information comprises information for linking one or more entities of the set of named entities to corresponding columns or cells of the one or more datasets;
post-process the IR using the fine-grained information by revising at least a subset of data within the IR, re-organizing a structure of the IR, or selecting a final dataset to use from the one or more datasets; and
generate a visualization for the post-processed IR.
16. The system of claim 15 , wherein the S2S machine-learning model is a Transformer-based model.
17. The system of claim 15 , wherein the schema comprises one or more of sorted column names, sorted named entities, sorted named expressions, and sorted named filters.
18. The system of claim 15 , wherein the metadata comprises cell information and at least columns that match the one or more entities recognized in the natural language query.