Generating a query statement based on unstructured input

• US 10,025,819 B2
• Filed: 11/13/2014
• Issued: 07/17/2018
• Est. Priority Date: 11/13/2014
• Status: Active Grant

• Current Assignee: Adobe Inc.
• Original Assignee: Adobe Systems Incorporated (Adobe Inc.)
• Inventors: Chang, Walter, Vlassis, Nikos, Garcia, Francisco
• Primary Examiner(s): Mackes, Kris E
• Application Number:
  US14/540,602
• Publication Number:
  US 20160140123A1
• Time in Patent Office: 1,342 Days
• CPC Class Codes: G06F 16/24522 Translation of natural language queries to structured queries

Abstract

Techniques for generating a query statement to query a dataset may be provided. For example, the query statement can be generated from natural language input, such as a natural language utterance. To do so, the input can be analyzed to detect a sentence, identify words in the sentence, and tag the words with the corresponding word types (e.g., nouns, verbs, adjectives, etc.). Expressions using the tags can be generated. Data about the expressions can be inputted to a classifier. Based on a detected pattern associated with the expressions, the classifier can predict a structure of the query statement, such as what expressions correspond to what clauses of the query statement. Based on this prediction, words associated with the expressions can be added to the clauses to generate the query statement and accordingly query the dataset.

Claims

1. A computer-implemented method comprising:

identifying words from a natural language request to search a dataset;

associating each of the words with an identifier that identifies a part of speech type of the corresponding word;

generating groupings of identifiers of the words, each of the groupings comprising a respective set of the identifiers associated with a set of the words, and each of the groupings corresponding to a particular part of speech type, the identifiers comprising a noun tag;

providing the words and the groupings to a machine learning classifier to generate a structured query to the dataset, the machine learning classifier configured to;

select a structure of the structured query for a noun grouping based on the groupings by at least predicting that a word associated with the noun tag in the noun grouping belongs to a select clause or a where clause of the structure, and

generate the structured query by using, in the structure, the set of words associated with the set of the identifiers of the noun grouping, wherein using the set of words comprises adding the word to the select clause or the where clause; and

querying the dataset based on the structured query provided from the machine learning classifier.

2. The computer-implemented method of claim 1, wherein selecting the structure of the structured query comprises comparing a pattern formed by the identifiers in the groupings to data based on previous observations for patterns of multiple structured queries by using a similarity or distance function.

3. The computer-implemented method of claim 1, further comprising:

identifying terms related to the one or more of the words based on at least a lexical ontology structure storing relationships between the terms and the one or more of the words, wherein the lexical ontology structure is generated based on a plurality of datasets;

adding the terms to an element of the structured query;

receiving query results based on using the structured query to query the dataset;

receiving query results based on using the structured query to query the dataset; and

ranking the query results based on whether a query result was generated based on a term from the lexical ontology structure.

4. The computer-implemented method of claim 1, wherein:

associating each of the words with an identifier comprises identifying that the word is a noun and tagging the word with the noun tag, and

generating the groupings comprises generating the noun grouping comprising the noun tag.

5. The computer-implemented method of claim 1, wherein the structure of the structured query comprises the select clause, the where clause, a from clause, or an order by clause, and wherein selecting the structure of the query comprises:

selecting a noun from the words or the noun grouping from the groupings;

if the noun is selected, selecting, from the groupings, the noun grouping corresponding to the noun to predict whether the noun grouping corresponds to the select clause;

generating a vector comprising features, wherein the features comprise identifiers of the noun grouping and a subset of the groupings;

providing an input based on the vector to the machine learning classifier;

identifying a probability that the noun grouping corresponds to the select clause, the where clause, the from clause, or the order by clause, wherein the probability is based on an output of the machine learning classifier in response to the input; and

adding, based on the probability, the noun to the select clause, the where clause, the from clause, or the order by clause.

6. The computer-implemented method of claim 1, wherein selecting the structure of the query statement class comprises:

selecting a grouping from the groupings, the grouping comprising one or more identifiers corresponding to one or more words;

generating a binary representation based on the one or more word identifiers of the grouping;

inputting the binary representation and the value to a random forest classifier; and

predicting that the one or more words belong to an element of the structure based on an output of the random forest classifier.

7. The computer-implemented method of claim 1, wherein selecting the structure of the structured query comprises:

selecting the word from the words;

selecting, from the groupings, the noun grouping associated with the word;

selecting, from the groupings, a set of groupings based on proximity to the noun grouping;

identifying a pattern based on the noun grouping and the set of the groupings; and

predicting the structure based on the pattern.

8. The computer-implemented method of claim 1, wherein the structured query comprises a structured query language (SQL) query, wherein the structure of the SQL query comprises the select clause or the where clause, wherein the words comprise a noun, and further comprising:

identifying a term related to the noun;

detecting whether the noun is to be added to the select clause or the where clause;

adding the term to the select clause based on detecting that the noun is to be added to the select clause; and

providing an option on a user interface to add the term to the where clause based on detecting that the noun is to be added to the where clause.

9. The computer-implemented method of claim 1, wherein the natural language request is generated based on a natural language utterance associated with a search for information.

10. The computer-implemented method of claim 1, further comprising:

determining that the natural language request is associated with the query to the dataset;

detecting whether the dataset comprises a relational database or a noSQL database;

structuring the structured query to include a SQL statement if the dataset comprises the relational database; and

structuring the structured query statement to include a NoSQL statement if the dataset comprises the NoSQL database.

11. A system comprising:

a processor; and

a memory communicatively coupled to the processor and bearing instructions that, upon execution by the processor, cause the system to at least perform operations comprising;

receiving natural language text comprising words;

generating expressions from the natural language text based on the words, each of the expressions comprising at least an identifier of a part of speech type corresponding to a word from the words, and each of the expressions corresponding to a particular part of speech type, the expression comprising a noun expression that includes a noun tag identifying a noun as one of the words; and

predicting, by using a machine learning classifier, at least a clause of a query statement for the noun expression based on the expressions, and wherein the machine learning classifier is configured to;

select a structure of the query statement by at least predicting that a word associated with the noun tag in the noun expression belongs to a select clause or a where clause of the structure, and

generate the query statement based on adding the word to the select clause or the where clause.

12. The system of claim 11, wherein the natural language text is generated based on unstructured input received from a personal assistant application hosted on a computing device, wherein the unstructured input comprises one or more of:

a natural language utterance or a natural language text input.

13. The system of claim 11, wherein the natural language text is generated based on unstructured input to search a dataset, wherein the dataset comprises annotations of content of the dataset, wherein the unstructured input is independent of the annotations, and wherein the query statement is configured to query the dataset based on the annotations without limiting the unstructured input.

14. The system of claim 11, wherein predicting the clause of the query statement comprises:

selecting, from the expressions, the noun expression comprising the noun tag, the noun tag associated with a potential class for the clause;

selecting a subset of the expressions based on proximity of the subset to the noun expression;

detecting a pattern based on the noun expression and the subset; and

predicting that the clause has a particular class based on the pattern.

15. The system of claim 11, predicting the clause of the query statement comprises:

providing data associated with the expressions to the machine learning classifier; and

detecting the clause based on an output of the machine learning classifier, the output comprising probabilities that the expression is associated with the clause.

16. The system of claim 11, wherein predicting the clause of the query statement is based on providing data associated with the expressions to the machine learning classifier, wherein the machine learning classifier is trained based on vectors comprising identifiers of part of speech types.

17. The system of claim 11, wherein:

the natural language text is associated with a search for attributes from a web site analytics database, wherein the attributes comprise one or more of;

a who attribute, a what attribute, a when attribute, a where attribute, or a how many attribute associated with visitors of a web site,

generating the expressions comprises selecting the noun expression associated with a word, and

predicting the clause comprises predicting that the noun expression is associated with an attribute from the attributes and setting the clause based on the attribute.

18. A non-transitory computer-readable storage medium storing instructions that, when executed on a computing device, configure the computing device to perform operations comprising:

receiving natural language text;

classifying one or more words from the natural language text as a noun;

associating the one or more words with a noun tag based on the classifying;

generating an expression comprising the noun tag; and

predicting, by using a machine learning classifier, a clause of a query statement for the expression based on a pattern associated with the expression and other expressions, the other expressions generated based on tags associated with classified words or terms from the natural language text, wherein the machine learning classifier is configured to;

select a structure of the query statement by at least predicting that the one or more words associated with the noun tag in the expression belongs to a select clause or a where clause of the structure, and

generate the query statement by adding the one or more words associated with the noun tag of the expression to the select clause or the where clause.

19. The non-transitory computer-readable storage medium of claim 18, wherein the natural language text is received to train the machine learning classifier, wherein the machine learning classifier is used to predict the clause of the query statement and is trained based on the expression and the other expressions.

20. The non-transitory computer-readable storage medium of claim 18, wherein the instructions, when executed on the computing device, further configure the computing device to perform operations comprising:

identifying first terms related to words from the natural language text based on a dictionary storing synonyms of the words;

identifying second terms related to the words based on an ontology tree storing relationships of the words with other words;

adding the first terms and the second terms to the query statement;

receiving query results based on submitting the query statement to a dataset; and

ranking the query results based on whether a query result was generated based on the words, the first terms, or the second terms.