Generating computer responses to social conversational inputs

Patent metadata

Field	Value
Publication number	US-2016005395-A1
Application number	US-201414323050-A
Country	US
Kind code	A1
Filing date	Jul 3, 2014
Priority date	Jul 3, 2014
Publication date	Jan 7, 2016
Grant date	—

Abstract

Conversational interactions between humans and computer systems can be provided by a computer system that classifies an input by conversation type, and provides human authored responses for conversation types. The input classification can be performed using trained binary classifiers. Training can be performed by labeling inputs as either positive or negative examples of a conversation type. Conversational responses can be authored by the same individuals that label the inputs used in training the classifiers. In some cases, the process of training classifiers can result in a suggestion of a new conversation type, for which human authors can label inputs for a new classifier and write content for responses for that new conversation type.

First claim

What is claimed is:

1 . A process for processing a conversational input, comprising:
receiving input data representing a conversational input into memory; causing feature data derived from the input data to be applied to a plurality of classifiers, each classifier representing a conversation type from among a plurality of conversation types and outputting match data in response to input feature data; and selecting a conversational response according to one or more conversation types from among the plurality of conversation types according to the match data output by the classifiers.

2 . The process of claim 1 , wherein the features include n-grams generated from the conversational input.

3 . The process of claim 2 , wherein the features include data indicative of emotion related to the conversational input.

4 . The process of claim 2 , wherein the features include data indicative of one or more classes for subject matter of the conversational input.

5 . The process of claim 1 , wherein each classifier is a binary classifier such that the match data output in response to the feature data represents a probability that the feature data matches a conversational input labeled as being a positive example of the conversation type associated with the binary classifier.

6 . The process of claim 1 , further comprising deriving the feature data from the conversational input and other information associated with the conversational input.

7 . The process of claim 1 , wherein selecting the conversational response comprises selecting a conversational response from a collection that stores a plurality of conversational responses for the plurality of conversation types.

8 . The process of claim 1 , wherein selecting the conversational response comprises:
selecting one or more conversation types from among the plurality of conversation types according to the match data output by the classifiers; and generating the conversational response according to the selected one or more conversation types.

9 . The process of claim 8 , wherein selecting one or more conversation types includes selecting a conversation type having a best score from among the match data output by the classifiers.

10 . A machine comprising:
computer readable storage; an input through which input data is received, the input data representing a conversational input; a processor configured to cause feature data derived from the input data to be applied to inputs of a plurality of classifiers and to receive data, indicative of a conversational response selected according to match data output by the plurality of classifiers, into the memory; an output providing output data based on the received conversational response.

11 . A process for training a classifier to classify conversational inputs into a conversation type, comprising:
labeling a first number of conversational inputs as positive examples of the conversation type and a second number of conversational inputs as negative examples of the conversation type; building the classifier using the labeled conversational inputs; accessing a corpus of conversational inputs; applying conversational inputs from the corpus to the classifier to obtain an output for each conversational input from the classifier; presenting additional conversational inputs to one or more individuals for labeling based on outputs from the classifier for the conversational inputs; and receiving labels from the one or more individuals for the presented additional conversational inputs; retraining the classifier using the additional labeled conversational inputs.

12 . The process of claim 11 , further comprising:
receiving, from one or more of the one or more individuals, conversational responses for the conversation type associated with the classifier for which the one or more individuals performed the labeling.

13 . The process of claim 11 , further comprising:
identifying additional conversation types from the corpus of conversational inputs.

14 . The process of claim 13 , wherein identifying comprises:
clustering conversational inputs in the corpus; assigning a conversation type to clusters.

15 . The process of claim 13 , wherein identifying comprises:
applying the conversational inputs to a plurality of classifiers; identifying the conversational inputs for which none of the plurality of classifiers indicates a match; assigning one or more new conversation types to the identified conversational inputs.

16 . The process of claim 11 , wherein presenting comprises:
selecting the additional conversational inputs to be presented according to whether the output of the classifier indicates that the additional conversational input to the classifier is neither a positive match nor a negative match.

17 . The process of claim 11 , wherein the conversation type of a classifier is included in a hierarchy of conversation types.

18 . The process of claim 11 , further comprising automatically clustering the corpus of conversational inputs to provide class information for the conversational inputs.

19 . The process of claim 1 , wherein the conversation type of a classifier is included in a hierarchy of conversation types.

20 . The process of claim 19 , wherein a conversation type is selected based on classifiers with a score above a threshold, and a more specific conversation type is selected over a more general conversation type.

Assignees

• Microsoft Corp

Inventors

• Williams Jason
• Zweig Geoffrey
• Lakshmiratan Aparna
• Suarez Carlos Garcia Jurado

Classifications

• G10L2015/223

  Execution procedure of a spoken command · CPC title

• G10L2015/226

  using non-speech characteristics · CPC title

• G10L15/22 Primary

  Procedures used during a speech recognition process, e.g. man-machine dialogue · CPC title

• G06F16/3329 Primary

  Natural language query formulation · CPC title

• G06F40/35

  Discourse or dialogue representation · CPC title