Determining stages of intent using text processing
A messaging server provides conversational text subsets to a machine-learned model that analyzes the text subsets to identify intents expressed therein. The messaging server determines intent groups associated with the text subsets based on the expressed intents. An intent group describes a category representing a subject area in which a text subset may express intent, and also describes a stage of the category representing a strength of the expressed intent. The messaging server applies decay factors to the intent groups. The decay factors include decay rates that describe how long the types of intents represented by the intent groups are maintained. The messaging server has access to suggestions having associated targeting criteria including intent groups to which the suggestions are targeted. The message server uses the targeting criteria to select suggestions targeted to users associated with particular text subsets, and delivers the selected suggestions to the users.
1. A method comprising:
receiving a conversational text subset;
providing the conversational text subset to a machine-learned intent model executing on a computer system, the intent model analyzing the conversational text subset to identify intents expressed within the conversational text subset;
determining an intent group for a category associated with conversational text subset based on the intents expressed within the conversational text subset, the intent group describing a combination of the category representing a subject area in which the conversational text subset expressed an intent and a stage representing a strength of the expressed intent;
associating a set of phrases with the intent group;
expanding the set of phrases associated with the intent group for the category by using the set of phrases as seed phrases to identify other phrases in a corpus that express a same intent in the category and a same strength of the expressed intent as the seed phrases to produce an expanded vocabulary for the intent group including the phrases and the identified other phrases associated with the pair of the category and the stage representing the strength of the expressed intent, the corpus comprising conversational text representing a conversation between two or more participants generated within a preceding time interval; and
building the machine-learned intent model using the expanded vocabulary for the intent group.
2. The method of claim 1 , further comprising:
assigning a user of a messaging server associated with the conversational text subset to the intent group;
identifying a length of time for which the user has been assigned to the intent group;
applying a decay factor to the intent group, the decay factor including a decay rate that describes how long the user maintains a type of intent represented by the intent group; and
selectively unassigning the user from the intent group responsive to the length of time for which the user has been assigned to the intent group and the decay factor applied to the intent group.
3. The method of claim 2 , wherein there are a plurality of intent groups and different intent groups of the plurality have different associated decay factors with different decay rates, and wherein applying the decay factor to the intent group comprises apply a decay factor associated with the intent group to the intent group.
4. The method of claim 1 , further comprising:
selecting a suggestion for delivery to a user of a messaging server associated with the conversational text subset, the suggestion selected responsive at least in part to the intent group associated with the conversational text subset; and
delivering the selected suggestion to the user via the messaging server.
5. The method of claim 4 , wherein selecting the suggestion for delivery to the user comprises:
determining targeting criteria associated with the suggestion, the targeting criteria comprising an intent group to which the suggestion is targeted;
determining that the user is associated with the conversational text subset associated with the intent group to which the suggestion is targeted; and
selecting the suggestion for delivery to the user responsive to determining that the user is associated with the conversational text subset associated with the intent group to which the suggestion is targeted.
6. The method of claim 1 , wherein determining the intent group comprises:
determining that the conversational text subset is associated with a plurality of different intent groups based on the intents expressed within the conversational text subset.
7. A non-transitory computer-readable medium storing computer program instructions executable by a processor to perform operations comprising:
receiving a conversational text subset;
providing the conversational text subset to a machine-learned intent model executing on a computer system, the intent model analyzing the conversational text subset to identify intents expressed within the conversational text subset;
determining an intent group for a category associated with conversational text subset based on the intents expressed within the conversational text subset, the intent group describing a combination of the category representing a subject area in which the conversational text subset expressed an intent and a stage representing a strength of the expressed intent;
associating a set of phrases with the intent group;
expanding the set of phrases associated with the intent group for the category by using the set of phrases as seed phrases to identify other phrases in a corpus that express a same intent in the category and a same strength of the expressed intent as the seed phrases to produce an expanded vocabulary for the intent group including the phrases and the identified other phrases associated with the pair of the category and the stage representing the strength of the expressed intent, the corpus comprising conversational text representing a conversation between two or more participants generated within a preceding time interval; and
building the machine-learned intent model using the expanded vocabulary for the intent group.
8. The computer-readable medium of claim 7 , the operations further comprising:
assigning a user of a messaging server associated with the conversational text subset to the intent group;
identifying a length of time for which the user has been assigned to the intent group;
applying a decay factor to the intent group, the decay factor including a decay rate that describes how long the user maintains a type of intent represented by the intent group; and
selectively unassigning the user from the intent group responsive to the length of time for which the user has been assigned to the intent group and the decay factor applied to the intent group.
9. The computer-readable medium of claim 8 , wherein there are a plurality of intent groups and different intent groups of the plurality have different associated decay factors with different decay rates, and wherein applying the decay factor to the intent group comprises apply a decay factor associated with the intent group to the intent group.
10. The computer-readable medium of claim 7 , further comprising:
selecting a suggestion for delivery to a user of a messaging server associated with the conversational text subset, the suggestion selected responsive at least in part to the intent group associated with the conversational text subset; and
delivering the selected suggestion to the user via the messaging server.
11. The computer-readable medium of claim 10 , wherein selecting the suggestion for delivery to the user comprises:
determining targeting criteria associated with the suggestion, the targeting criteria comprising an intent group to which the suggestion is targeted;
determining that the user is associated with the conversational text subset associated with the intent group to which the suggestion is targeted; and
selecting the suggestion for delivery to the user responsive to determining that the user is associated with the conversational text subset associated with the intent group to which the suggestion is targeted.
12. The computer-readable medium of claim 7 , wherein determining the intent group comprises:
determining that the conversational text subset is associated with a plurality of different intent groups based on the intents expressed within the conversational text subset.
13. A computer system comprising:
a computer processor adapted to execute computer program instructions; and
a non-transitory computer-readable storage medium storing computer program instructions executable by the processor to perform operations comprising:
receiving a conversational text subset;
providing the conversational text subset to a machine-learned intent model executing on a computer system, the intent model analyzing the conversational text subset to identify intents expressed within the conversational text subset;
determining an intent group for a category associated with conversational text subset based on the intents expressed within the conversational text subset, the intent group describing a combination of the category representing a subject area in which the conversational text subset expressed an intent and a stage representing a strength of the expressed intent;
associating a set of phrases with the intent group;
expanding the set of phrases associated with the intent group for the category by using the set of phrases as seed phrases to identify other phrases in a corpus that express a same intent in the category and a same strength of the expressed intent as the seed phrases to produce an expanded vocabulary for the intent group including the phrases and the identified other phrases associated with the pair of the category and the stage representing the strength of the expressed intent, the corpus comprising conversational text representing a conversation between two or more participants generated within a preceding time interval; and
building the machine-learned intent model using the expanded vocabulary for the intent group.
14. The computer system of claim 13 , the operations further comprising:
assigning a user of a messaging server associated with the conversational text subset to the intent group;
identifying a length of time for which the user has been assigned to the intent group;
applying a decay factor to the intent group, the decay factor including a decay rate that describes how long the user maintains a type of intent represented by the intent group; and
selectively unassigning the user from the intent group responsive to the length of time for which the user has been assigned to the intent group and the decay factor applied to the intent group.
15. The computer system of claim 13 , the operations further comprising:
selecting a suggestion for delivery to a user of a messaging server associated with the conversational text subset, the suggestion selected responsive at least in part to the intent group associated with the conversational text subset; and
delivering the selected suggestion to the user via the messaging server.
16. The computer system of claim 15 , wherein selecting the suggestion for delivery to the user comprises:
determining targeting criteria associated with the suggestion, the targeting criteria comprising an intent group to which the suggestion is targeted;
determining that the user is associated with the conversational text subset associated with the intent group to which the suggestion is targeted; and
selecting the suggestion for delivery to the user responsive to determining that the user is associated with the conversational text subset associated with the intent group to which the suggestion is targeted.
17. The computer system of claim 13 , wherein determining the intent group comprises:
determining that the conversational text subset is associated with a plurality of different intent groups based on the intents expressed within the conversational text subset.