Unsupervised automated extraction of conversation structure from recorded conversations

Abstract

A method for information processing includes completing, over a corpus of conversations, a conversation structure model including (i) a sequence of conversation parts having a defined order, and (ii) a probabilistic model defining each of the conversation parts. For a given conversation, a segmentation of the conversation is computed based on the computed conversation structure model. Action is taken on the given conversation according to the segmentation.

1 . A method for information processing, the method comprising:
 computing, over a corpus of conversations, a conversation structure model comprising (i) a sequence of conversation parts having a defined order, and (ii) a probabilistic model defining each of the conversation parts;   computing, for a given conversation, a segmentation of the conversation based on the computed conversation structure model; and   acting on the given conversation according to the segmentation.   
     
     
         2 . The method according to  claim 1 , wherein. computing the probabilistic model comprises assigning a probability to an occurrence of each word. The method according to  claim 2 , wherein assigning the probability comprises running a Gibbs sampling process. 
     
     
         4 . The method according to  claim 2 , wherein assigning the probability comprises using a prior probability distribution for one or more of the conversation parts. 
     
     
         5 . The method according to  claim 1 , wherein computing the conversation structure model comprises pre-specifying a fixed number of the conversation parts. 
     
     
         6 . The method according to  claim 1 , wherein computing the conversation structure model comprises selecting a subset of the conversations based on one or more business rules. 
     
     
         7 . The method according to  claim 1 , wherein computing the segmentation of the conversation comprises finding the segmentation that best matches the conversation structure model. 
     
     
         8 . The method according to  claim 1 , and comprising computing a coherence score, which quantifies an extent of fit between the given conversation and the conversation structure model. 
     
     
         9 . The method according to  claim 8 , and comprising, when the coherence score is below a given value, regarding the given conversation as not matching the conversation structure model. 
     
     
         10 . The method according to  claim 8 , wherein estimating the coherence score comprises analyzing a likelihood of the segmentation of the conversation under the conversation structure model. 
     
     
         11 . The method according to  claim 8 , and comprising deciding, based on one or more coherence scores computed between one or more respective conversations in the corpus and the conversation structure model, that the conversation structure model does not capture valid conversations structure. 
     
     
         12 . The method according to  claim 1 , and comprising, subsequent to computing the conversation structure model, merging one or more of the conversation parts into a single conversation part. 
     
     
         13 . The method according to claim wherein the conversations are transcribed from human conversations. 
     
     
         14 . The method according to claim wherein the conversations are recorded conversations, conducted over a telephone, a conference system, or in a meeting. 
     
     
         15 . The method according to  claim 1 , wherein acting on the given conversation comprises presenting a timeline that graphically illustrates the respective order and durations of the conversation parts during the given conversation. 
     
     
         16 . The method according to  claim 1 , wherein acting on the given conversation comprises displaying conversation part duration to computer users. 
     
     
         17 . The method according to  claim 1 , and comprising searching for words within. a conversation or within the corpus based. on a conversation part to which the words are assigned. 
     
     
         18 . The method according to  claim 1 , and comprising correlating the conversation parts of a given participant with participant metadata to identify conversation differences between participants. 
     
     
         19 . A system for information processing, comprising:
 an interface for accessing a corpus of recorded conversations; and   a processor, configured to:   compute, over a corpus of conversations, a conversation structure model comprising (i) a sequence of conversation parts having a defined order, and (ii) a probabilistic model defining each of the conversation parts;   compute, for a given conversation, a segmentation of the conversation based on the computed conversation structure model; and   a on the given conversation according to the segmentation.   
     
     
         20 . The system according to  claim 19 , wherein the processor is configured to compute the probabilistic model by assigning a probability to an occurrence of each word. 
     
     
         21 . The system according to  claim 20 , wherein the processor is configured to assign the probability by running a Gibbs sampling process. 
     
     
         22 . the system according to  claim 20 , wherein the processor is configured to assign the probability by using a prior probability distribution for one or more of the conversation parts. 
     
     
         23 . The system according to  claim 19 , wherein the processor is configured to compute the conversation structure model by pre-specif fixed number of the conversation parts. 
     
     
         24 . The system according to  claim 19 , wherein the processor is configured to compute the conversation structure model by selecting a subset of the conversations based on one or more business rules. 
     
     
         25 . The system according to  claim 19 , wherein the processor is configured to compute the segmentation of the conversation by finding the segmentation that best matches the conversation structure model. 
     
     
         26 . The system according to  claim 19 , wherein the processor is further configured to compute a coherence score, which quantifies an extent. of fit between the given conversation and the conversation structure model. 
     
     
         27 . The system according to  claim 26 , wherein the processor is further configured to, when the coherence score is below a given value, regard the given conversation as not matching the conversation structure model. 
     
     
         28 . The system according to  claim 26 , wherein the processor is configured to estimate the coherence score by analyzing a likelihood of the segmentation of the conversation under the conversation structure model. 
     
     
         29 . The system according to  claim 26 , wherein the processor is further configured to decide, based on one or more coherence scores computed between one or more respective conversations in the corpus and the conversation structure model, that the conversation structure model does not capture a valid conversations structure. 
     
     
         30 . The system according to  claim 19 , wherein the processor is further configured to, subsequent to computing the conversation structure model, merge one or more of the conversation parts into a single conversation part. 
     
     
         31 . The system according to  claim 19 , wherein the conversations are transcribed from human conversations. 
     
     
         32 . The system according to  claim 19 , wherein the conversations are recorded conversations, conducted over a telephone, a conference system, or in a meeting. 
     
     
         33 . The system according to  claim 19 , wherein the processor is further configured to act on the given conversation by presenting a timeline that graphically illustrates the respective order and durations of the conversation parts during the given conversation. 
     
     
         34 . The system according to  claim 19 , wherein the processor is further configured to act on the given conversation by displaying conversation part duration to computer users. 
     
     
         35 . The system according to  claim 19 , wherein the processor is further configured to search for words within a conversation or within the corpus based on a conversation part to which the words are assigned. 
     
     
         36 . The system according to  claim 19 , wherein the processor s further configured to correlate the conversation parts of a given participant with participant metadata to identify conversation differences between participants. 
     
     
         37 . A computer software product, the product comprising a tangible non-transitory computer readable medium in which program instructions are stored, which instructions, when read by a processor, cause the processor to:
 compute, over a corpus of conversations, a conversation structure model comprising (i) a sequence of conversation parts having a defined order, and (Ii) a probabilistic model defining each of the conversation parts;   compute, for a given conversation, a segmentation of the conversation based on the computed conversation. structure model;   act on the given conversation according to the segmentation.   
     
     
         38 . The computer software product according to  claim 37 , wherein the processor is further configured to compute a coherence score, which quantifies an extent of fit between the given conversation and the conversation structure model.