Training a chatbot for a digital advertisement to simulate common conversations associated with similar digital advertisements

Patent metadata

Field	Value
Publication number	US-10565634-B2
Application number	US-201715666433-A
Country	US
Kind code	B2
Filing date	Aug 1, 2017
Priority date	Aug 1, 2017
Publication date	Feb 18, 2020
Grant date	Feb 18, 2020

Abstract

This disclosure covers systems and methods that generate and train a chatbot to automatically communicate with users concerning subjects related to a digital advertisement received for distribution. In particular, the disclosed systems and methods train a chatbot to simulate common conversation exchanges from messaging threads associated with previous digital advertisements that are similar to a received digital advertisement. By training the chatbot to simulate such conversations, the disclosed systems and methods create a chatbot that can immediately respond to a user's inquiries concerning the received digital advertisement and tailor automated exchanges that further the objectives of an advertiser or merchant associated with the received digital advertisement.

First claim

What is claimed is:

1. A method comprising:
analyzing a received digital advertisement to determine characteristics associated with the received digital advertisement;
analyzing messaging threads associated with similar digital advertisements sharing one or more characteristics with the received digital advertisement, the messaging threads comprising inputs and responses between users of a social networking system and representatives of advertisers associated with the similar digital advertisements;
identifying, from among the messaging threads associated with the similar advertisements, common conversation exchanges comprising pairs of common inputs by the representatives of advertisers and corresponding responses by one or more users of the social networking system to the common inputs;
based on the common conversation exchanges, collecting data related to the received digital advertisement to simulate the common conversation exchanges within automated conversations associated with the received digital advertisement; and
training a chatbot to automatically communicate with users in automated conversation exchanges related to the received digital advertisement by:
identifying word entities and syntax patterns from the pairs of common inputs by the representatives of advertisers and the corresponding responses by the one or more users of the social networking system within the common conversation exchanges; and
constructing inputs and responses simulating the pairs of common inputs by the representatives of advertisers and the corresponding responses by the one or more users of the social networking system, wherein the inputs and responses incorporate words or phrases from the collected data related to the received digital advertisement.

2. The method of claim 1 , wherein collecting data related to the received digital advertisement comprises:
aggregating data related to the received digital advertisement and data related to an advertiser associated with the received digital advertisement;
determining that the aggregated data does not include missing information required to simulate the common conversation exchanges; and
prompting the advertiser to provide the missing information required to simulate the common conversation exchanges.

3. The method of claim 2 , wherein aggregating the data related to the received digital advertisement and the data related to the advertiser associated with the received digital advertisement comprises one or more of:
collecting data from contents of the received digital advertisement;
collecting data from previous digital advertisements of the advertiser;
collecting data from a website of the advertiser;
collecting data from a social networking profile of the advertiser;
collecting data from a product catalogue; or
receiving responses concerning the received digital advertisement from the advertiser through a digital form.

4. The method of claim 1 , wherein analyzing the messaging threads associated with the similar digital advertisements comprises:
comparing the characteristics associated with the received digital advertisement with characteristics associated with previous digital advertisements to identify shared characteristics between the received digital advertisement and the previous digital advertisements;
based on the shared characteristics, generating a similarity score for each of the previous digital advertisements, each similarity score representing a similarity between the received digital advertisement and one of the previous digital advertisements;
identifying the previous digital advertisements having similarity scores above a threshold as the similar digital advertisements; and
accessing the messaging threads associated with the similar digital advertisements.

5. The method of claim 4 , wherein analyzing the messaging threads associated with the similar digital advertisements comprises applying natural language processing to determine an intent of each input and each response within the messaging threads.

6. The method of claim 5 , further comprising:
identifying, from among the messaging threads associated with the similar advertisements, additional common conversation exchanges comprising pairs of common inputs by the one or more users of the social networking system and corresponding responses by the representatives of advertisers to the common inputs;
based on the additional common conversation exchanges, collecting data related to the received digital advertisement to simulate the additional common conversation exchanges within automated conversations associated with the received digital advertisement; and
training the chatbot to automatically communicate with users in automated conversation exchanges related to the received digital advertisement based on the common conversation exchanges and the additional common conversation exchanges.

7. The method of claim 5 , wherein identifying the common conversation exchanges associated with the messaging threads comprises, for each common conversation exchange of the common conversation exchanges:
analyzing a plurality of inputs and a plurality of responses within the messaging threads;
determining that the common inputs within the plurality of inputs have a shared intent and exceed a first commonality threshold representing a threshold measurement of input occurrences within the messaging threads;
determining that common responses by the one or more users of the social networking system to the common inputs within the plurality of responses have a shared intent and exceed a second commonality threshold representing a threshold measurement of response occurrences within the messaging threads; and
identifying common pairs of the common inputs and the common responses as one or more of the common conversation exchanges associated with the messaging threads.

8. The method of claim 1 , further comprising:
identifying, from among the common conversation exchanges, common sequences comprising a repeated pair of a common input and a corresponding response to the common input; and
training the chatbot to order one or more inputs and one or more responses within a messaging thread based on the common sequences.

9. The method of claim 1 , wherein training the chatbot to automatically communicate with users in automated conversation exchanges comprises:
training the chatbot to customize inputs for an automated conversation exchange based on the common inputs and the collected data related to the received digital advertisement by:
identifying one or more word entities and one or more syntax patterns from the common inputs by the representatives of advertisers; and
incorporating one or more words or one or more phrases from the common inputs to customize an input.

10. The method of claim 7 , wherein collecting the data related to the received digital advertisement to simulate the common conversation exchanges comprises:
identifying input-syntax patterns and input-word entities within the common inputs;
collecting input data related to the received digital advertisement that corresponds to the input-syntax patterns and the input-word entities within the common inputs by the representatives of advertisers;
identifying response-syntax patterns and response-word entities within the common responses by the one or more users of the social networking system; and
collecting response data related to the received digital advertisement that corresponds to the response-syntax patterns and the response-word entities within the common responses.

11. The method of claim 10 , wherein training the chatbot to automatically communicate with users in automated conversation exchanges co

Assignees

• Facebook Inc

Inventors

• Talmor Yoram
• Baweja Himanshu

Classifications

• G06F40/166

  Editing, e.g. inserting or deleting · CPC title

• G06Q30/0643

  graphically representing goods, e.g. 3D product representation · CPC title

• G06N3/006

  based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO] · CPC title

• G06Q30/0617 Primary

  Representative agent · CPC title

• G06F40/106

  Display of layout of documents; Previewing · CPC title