# US 12,183,332 B2 · Unsupervised automated extraction of conversation structure from recorded conversations

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 12,183,332 B2
- Document ID: US12183332B2
- Status: Granted
- Kind: B2
- Application number: 18/310,558
- Filed: 2023-05-02
- Published: 2023-08-24
- Granted: 2024-12-31
- Earliest claimed: 2019-07-24
- Assignees: GONG.io Ltd.
- Applicants: Gong.Io Ltd
- Inventors: Inbal Horev; Eilon Reshef; Omri Allouche; Yoav Madorsky; Hanan Shteingart
- Web page: https://patentgrep.com/patent/US12183332B2

## CPC

- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G10L 15/197: Probabilistic grammars, e.g. word n-grams
- G06F 40/289: Phrasal analysis, e.g. finite state techniques or chunking
- G06F 40/35: Discourse or dialogue representation
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/00: Machine learning
- G10L 15/04: Segmentation; Word boundary detection
- G10L 15/18: using natural language modelling
- G10L 15/22: Procedures used during a speech recognition process, e.g. man-machine dialogue
- G06F 40/30: Semantic analysis
- G06N 5/041: {Abduction}
- G10L 15/1815: {Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning}
- G10L 15/1822: {Parsing for meaning understanding}
- G10L 15/183: using context dependencies, e.g. language models
- G10L 15/26: Speech to text systems (G10L15/08 takes precedence)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A method for information processing includes computing, 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.

## Claims

### Claim 1

A method for processing a conversation, the method comprising:
creating a conversation structure model from a corpus of conversations, the conversation structure model including a fixed number of conversation parts arranged in an ordered sequence having a defined order, such that the conversation parts are constrained to appear in the defined order; and
applying a given conversation to the conversation structure model by enforcing the conversation structure model on the given conversation to output the given conversation in accordance with the conversation structure model, the enforcing comprising:
analyzing word distributions in the given conversation to best fit each of the conversation parts and the ordered sequence of the conversation parts of the conversation structure model; and
based on the analysis, outputting the given conversation in segments corresponding to the fixed number of conversation parts of the conversation structure model, with the segments ordered in a sequence corresponding to the ordered sequence of the conversation parts in the conversation structure model,
wherein the conversation structure model is constrained such that each of the conversation parts appears no more than once in the given conversation.

### Claim 2 (depends on 1)

The method of claim 1, wherein the fixed number of conversation parts includes two or more.

### Claim 3 (depends on 2)

The method of claim 2, the method additionally comprising: analyzing each of the segments of the given conversation against the corresponding conversation part of the conversation structure model to determine a coherence for the given conversation.

### Claim 4

A computer-implemented method for processing a conversation, the computer-implemented method comprising:
accessing, by one or more processors, a conversation structure model trained from a corpus of conversations, the conversation structure model including a fixed number of conversation parts arranged in an ordered sequence having a defined order, such that the conversation parts are constrained to appear in the defined order;
deploying the trained model by the one or more processors;
inputting a given conversation comprising words into the conversation structure model; and
enforcing the conversation structure model on the given conversation to output the given conversation as segmented conversation formed segments, the segmented conversation formed segments being created based on analyzing the words from the given conversation, the number of the segmented conversation formed segments corresponding to the number of conversation parts of the conversation structure model, with the segmented conversation formed segments ordered in a sequence corresponding to the ordered sequence of the conversation parts in the conversation structure model,
wherein the conversation structure model is constrained such that each of the conversation parts appears no more than once in the given conversation.

### Claim 5 (depends on 4)

The computer-implemented method of claim 4, wherein the analyzing the words from the given conversation comprises analyzing the words based on word distributions.

### Claim 6

A computer-implemented method for processing a conversation, the computer-implemented method comprising:
accessing, by one or more processors, a conversation structure model having been trained based on: 1) collected word occurrence probabilities from conversations in a corpus of conversations to form conversation parts, and 2) collected sequences of conversation parts from the conversations from the corpus of conversations, the conversation structure model being an optimal conversation including a predetermined number of conversation parts, each of the conversation parts arranged in an ordered sequence having a defined order, such that the conversation parts are constrained to appear in the defined order;
deploying the conversation structure model by the one or more processors;
inputting a given conversation comprising words into the conversation structure model; and
enforcing the conversation structure model on the given conversation to output the given conversation as segmented conversation formed segments, the segmented conversation formed segments being created based on analyzing word distributions of the words from the given conversation, the number of the segmented conversation formed segments corresponding to the number of conversation parts of the conversation structure model, with the segmented conversation formed segments ordered in a sequence corresponding to the ordered sequence of the conversation parts in the conversation structure model,
wherein the conversation structure model is constrained such that each of the conversation parts appears no more than once in the given conversation.

### Claim 7 (depends on 6)

The computer-implemented method of claim 6, wherein the number of conversation parts includes a plurality of conversation parts.

### Claim 8

A system for conversation processing, the system comprising:
a memory of a computer, the memory configured for storing a conversation structure model:
the conversation structure model having been built based on a corpus of conversations, and
the conversation structure model including a fixed number of conversation parts arranged in an ordered sequence having a defined order, such that the conversation parts are constrained to appear in the defined order; and
one or more processors configured to:
deploy the conversation structure model stored in the memory;
obtain a given conversation; and
apply the given conversation to the conversation structure model including enforcing the conversation structure of the conversation structure model on the given conversation to output the given conversation in accordance with the conversation structure model, the enforcing comprising:
analyzing the word distributions in the given conversation to best fit each of the conversation parts and the ordered sequence of the conversation parts of the conversation structure model; and
based on the analysis, outputting the given conversation in segments corresponding to the fixed number of conversation parts of the conversation structure model, with the segments ordered in a sequence corresponding to the ordered sequence of the conversation parts in the conversation structure model,
wherein the conversation structure model is constrained such that each of the conversation parts appears no more than once in the given conversation.

### Claim 9 (depends on 8)

The system of claim 8, wherein the fixed number of conversation parts includes two or more.

### Claim 10 (depends on 8)

The system of claim 8, wherein the enforcing additionally comprises: analyzing each of the segments of given the conversation the corresponding against conversation part of the conversation structure model to determine a coherence for the given conversation.

### Claim 11 (depends on 8)

The system of claim 8, wherein the fixed number of conversation parts arranged in an ordered sequence is obtained from an extraction of the conversation parts of the corpus.

### Claim 12 (depends on 8)

The system of claim 8, wherein the conversation structure model includes a trained conversation structure model.

### Claim 13

A system for conversation processing, the system comprising:
a memory of a computer, the memory configured for storing a conversation structure model;
the conversation structure model having been built from a corpus of conversations, the conversation structure model including a fixed number of conversation parts arranged in an ordered sequence having a defined order, such that the conversation parts are constrained to appear in the defined order; and
one or more processors configured to:
deploy the conversation structure model;
input a given conversation comprising words into the conversation structure model; and
enforce the conversation structure model on the given conversation, to output the given conversation as segmented conversation formed segments, the segmented conversation formed segments being created based on analyzing the words from the given conversation, the number of the segmented conversation formed segments corresponding to the fixed number of conversation parts of the conversation structure model, with the segmented conversation formed segments ordered in a sequence corresponding to the ordered sequence of the conversation parts in the conversation structure model,
wherein the conversation structure model is constrained such that each of the conversation parts appears no more than once in the given conversation.

### Claim 14 (depends on 13)

The system of claim 13, wherein the one or more processors configured to analyze the words from the given conversation performs the analysis based on word distributions.

### Claim 15 (depends on 13)

The system of claim 13, wherein the fixed number of conversation parts arranged in an ordered sequence is obtained from an unsupervised extraction of the conversation parts of the corpus.

### Claim 16 (depends on 13)

The system of claim 13, wherein the conversation structure model includes a trained conversation structure model.

### Claim 17

A system for processing a conversation, the system comprising:
a memory of a computer, the memory configured for storing a conversation structure model;
the conversation structure model having been built based on: 1) collected word occurrence probabilities from conversations in a corpus of conversations to form conversation parts, and 2) collected sequences of conversation parts from the conversations from the corpus of conversations, the conversation structure model being an optimal conversation including a predetermined number of conversation parts, each of the conversation parts arranged in an ordered sequence having a defined order, such that the conversation parts are constrained to appear in the defined order; and
one or more processors configured to:
deploy the conversation structure model;
input a given conversation comprising words into the conversation structure model; and
enforce the conversation structure model on the given conversation to output the given conversation as segmented conversation formed segments, the segmented conversation formed segments being created based on analyzing word distributions of the words from the given conversation, the number of the segmented conversation formed segments corresponding to the number of conversation parts of the conversation structure model, with the segmented conversation formed segments ordered in a sequence corresponding to the ordered sequence of the conversation parts in the conversation structure model,
wherein the conversation structure model is constrained such that each of the conversation parts appears no more than once in the given conversation.

### Claim 18 (depends on 17)

The system of claim 17, wherein the number of conversation parts includes a plurality of conversation parts.

### Claim 19 (depends on 18)

The system of claim 18, wherein the number of conversation parts includes a fixed number of conversation parts.

### Claim 20 (depends on 18)

The system of claim 18, wherein each of the plurality of conversation parts is obtained from an unsupervised extraction of the conversation parts of the corpus and is arranged in the ordered sequence.

### Claim 21 (depends on 17)

The system of claim 17, wherein the conversation structure model trained includes a conversation structure model.
