Automated scheme for identifying user intent in real-time

• US 8,924,410 B2
• Filed: 11/18/2011
• Issued: 12/30/2014
• Est. Priority Date: 04/07/2004
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: Bierner, Gann Alexander
• Primary Examiner(s): Richardson, James E
• Application Number:
  US13/299,465
• Publication Number:
  US 20120131033A1
• Time in Patent Office: 1,138 Days
• Field of Search: 707765-768, 707/771, 715/260, 715/816
• US Class Current: 707/768
• CPC Class Codes: G06F 16/3322 using system suggestions G06F16/3325 takes precedence

Abstract

An intent guessing system receives partial user queries as they are entered by the user. The partial user queries are compared with different intents derived from previously logged queries. Guesses are made as to which of the intents are associated with the partial user query. The intent guesses are then provided as responses to the user query. Features are identified for the earlier logged queries and associated with the derived intents. The derived intents and associated features are then used to identify intents for the partial user queries.

Claims

1. A method of processing a current user query, comprising:

maintaining, by a computer system, a model comprising a plurality of entries, each entry of the model comprising a plurality of query features, a query intent, and an indication of a statistical association between the query feature and the query intent, wherein the plurality of query features comprise a query token, a stem for the query token, and a concept represented by the query token;

receiving, at the computer system, an initial portion of the current user query, the initial portion including a particular string of text;

guessing, by the computer system, an intent associated with the initial portion of the current user query while a remainder of the current user query is being received, wherein said guessing is based at least in part on the received initial portion, the query features of the entries of the model, and the indications of statistical association between the query features and the query intents of the model and comprises;

analyzing the initial portion to obtain at least one feature,

comparing the obtained at least one feature of the initial portion to the query features of the entries in the model,

selecting a plurality of entries from the model based on the comparison,

determining the intent for the initial portion of the current user query according to the query intent in the selected plurality of entries and the statistical association between the query feature and the query intent, and

generating a plurality of intent guesses that are different from the current user query and that do not include the particular text string based on the determined intent for the initial portion of the current user query wherein generating the intent guesses comprises identifying from the entries selected from the model a parameterized intent guess that includes at least one concept, identifying words that are associated with the concept but that are not the particular text string, and modifying the parameterized intent guess by replacing the concept with the identified words; and

responding, by the computer system, to the initial portion of the current user query with the plurality of intent guesses including the modified parameterized intent guess.

2. The method of claim 1, wherein maintaining the model comprises training the model prior to receiving the initial portion of the current user query, said training comprising:

logging previous user queries;

extracting a query from the logged previous user queries;

analyzing the extracted query to obtain at least one query feature;

assigning a query intent to the extracted query; and

adding an entry to the model and associating the obtained query feature for the extracted query to the assigned query intent for the extracted query in the added entry of the model.

3. The method of to claim 1, wherein maintaining the model comprises updating the model based on an indication of a user selection of an intent guess, wherein the user selection comprises a selection of one of the plurality of intent guesses, and wherein updating the model comprises updating the indication of the statistical association between the query feature and the query intent for at least one of the entries.

4. The method according to claim 1, further comprising:

identifying, by the computer system, a response associated with each of the plurality of intent guesses;

identifying, by the computer system, an order for the plurality of intent guesses based on the statistical associations between the query feature and the query intent for each or the selected plurality of entries of the model; and

displaying, by the computer system, based on the determined order the response along with the associated intent guesses or the response instead of the associated intent guesses.

5. The method according to claim 4, wherein the responses include links to web pages or additional information associated with the intent guesses.

6. The method according to claim 1, further comprising:

identifying, by the computer system, an order for the plurality of intent guesses based on the statistical associations between the query feature and the query intent for each or the selected plurality of entries of the model; and

displaying, by the computer system, the intent guesses in the identified order.

7. The method of claim 1, wherein the intent guesses do not include a same set of words as the particular text string.

8. The method of claim 7, wherein the intent guesses do not include any same words as the particular text string.

9. A method of processing a current user query, comprising:

training a model, said training comprising;

logging previous user queries;

extracting a query from the logged previous user queries;

analyzing the extracted query to obtain a plurality of features, wherein the plurality of features comprise a query token, a stem for the query token, and a concept represented by the query token;

assigning an intent to the extracted query; and

adding an entry to the model and mapping the obtained features for the extracted query to the assigned intent for the extracted query in the added entry of the model;

receiving an initial portion of the current user query, wherein the initial portion includes a particular string of text;

guessing an intent associated with the initial portion of the current user query while a remainder of the current user query is being received, wherein said guessing is based at least in part on the received initial portion, the query features of the entries of the model, and the indications of statistical association between the query features and the query intents of the model and further comprises;

analyzing the initial portion to obtain at least one feature,

comparing, using a processing device, the obtained at least one feature of the initial portion to entries in the model,

selecting, using the processing device, at least one entry from the model based on the comparison

determining the intent for the initial portion of the current user query according to an intent included in the selected entry, and

generating the intent guess based on the determined intent for the initial portion of the current user query wherein the intent guess is different from the current user query and that does not include the particular text string and wherein generating the intent guesses comprises identifying from the entries selected from the model a parameterized intent guess that includes at least one concept, identifying words that are associated with the concept but that are not the particular text string, and modifying the parameterized intent guess by replacing the concept with the identified words; and

responding to the initial portion of the current user query with the intent guess including the modified parameterized intent guess.

10. The method according to claim 9, further comprising:

identifying a response associated with the intent guess; and

either displaying the response along with the associated intent guess or displaying the response instead of the associated intent guess.

11. The method according to claim 9, further comprising:

identifying a plurality of intent guesses associated with the current user query;

identifying an order for the intent guesses of the plurality according to how well the intent guesses match the current user query; and

displaying the intent guesses in the identified order.

12. A system comprising:

a processor; and

a memory coupled with and readable by the processor and storing therein a set of instructions which, when executed by the processor, causes the processor to process a current user query by;

maintaining a model comprising a plurality of entries, each entry of the model comprising a plurality of query features, a query intent, and an indication of a statistical association between the query feature and the query intent, wherein the plurality of query features comprise a query token, a stem for the query token, and a concept represented by the query token,

receiving an initial portion of the current user query, the initial portion including a particular string of text,

guessing an intent associated with the initial portion of the current user query while a remainder of the current user query is being received, wherein said guessing is based at least in part on the received initial portion, the query features of the entries of the model, and the indications of statistical association between the query features and the query intents of the model and comprises;

analyzing the initial portion to obtain at least one feature,

comparing the obtained at least one feature of the initial portion to the query features of the entries in the model,

selecting a plurality of entries from the model based on the comparison,

determining the intent for the initial portion of the current user query according to the query intent in the selected plurality of entries and the statistical association between the query feature and the query intent, and

generating a plurality of intent guesses that are different from the current user query and that do not include the particular text string based on the determined intent for the initial portion of the current user query wherein generating the intent guesses comprises identifying from the entries selected from the model a parameterized intent guess that includes at least one concept, identifying words that are associated with the concept but that are not the particular text string, and modifying the parameterized intent guess by replacing the concept with the identified words, and

responding to the initial portion of the current user query with the plurality of intent guesses including the modified parameterized intent guess.

13. A non-transitory computer readable memory comprising a set of instructions stored therein which, when executed by a processor, causes the processor to process a current user query by:

maintaining a model comprising a plurality of entries, each entry of the model comprising a plurality of query features, a query intent, and an indication of a statistical association between the query feature and the query intent, wherein the plurality of query features comprise a query token, a stem for the query token, and a concept represented by the query token;

receiving an initial portion of the current user query, the initial portion including a particular string of text;

guessing an intent associated with the initial portion of the current user query while a remainder of the current user query is being received, wherein said guessing is based at least in part on the received initial portion, the query features of the entries of the model, and the indications of statistical association between the query features and the query intents of the model and comprises;

analyzing the initial portion to obtain at least one feature,

comparing the obtained at least one feature of the initial portion to the query features of the entries in the model,

selecting a plurality of entries from the model based on the comparison,

determining the intent for the initial portion of the current user query according to the query intent in the selected plurality of entries and the statistical association between the query feature and the query intent, and

generating a plurality of intent guesses that are different from the current user query and that do not include the particular text string based on the determined intent for the initial portion of the current user query wherein generating the intent guesses comprises identifying from the entries selected from the model a parameterized intent guess that includes at least one concept, identifying words that are associated with the concept but that are not the particular text string, and modifying the parameterized intent guess by replacing the concept with the identified words; and

responding to the initial portion of the current user query with the plurality of intent guesses including the modified parameterized intent guess.