Enriched message embedding for conversation deinterleaving

Patent metadata

Field	Value
Publication number	US-11677705-B2
Application number	US-201916391458-A
Country	US
Kind code	B2
Filing date	Apr 23, 2019
Priority date	Apr 23, 2019
Publication date	Jun 13, 2023
Grant date	Jun 13, 2023

Abstract

An approach is provided that receives a message and applies a deep analytic analysis to the message. The deep analytic analysis results in a set of enriched message embedding (EME) data that is passed to a trained neural network. Based on a set of scores received from the trained neural network, a conversation is identified from a number of available conversations to which the received message belongs. The received first message is then associated with the identified conversation.

First claim

What is claimed is:

1. An information handling system comprising:
one or more processors;
a memory coupled to at least one of the processors; and
a set of computer program instructions stored in the memory and executed by at least one of the processors in order to perform actions comprising:
receiving a first message and a second message;
applying a deep analytic analysis to the received messages to form a first set and a second set of enriched message embedding (EME) data;
processing each conversation included in a plurality of conversations, wherein, when selected, each of the conversations is processed by a trained neural network by:
loading a set of cell states in the neural network with a set of previously saved cell states that correspond to the selected conversation;
passing the first and second sets of EME data to the trained neural network;
receiving a first set of scores from the trained neural network that indicates an affinity of the selected conversation to the received first message and a second set of scores from the trained neural network that indicates an affinity of the selected conversation to the received second message;

identifying, based on comparing a threshold to the first and second sets of scores, one of the conversations to which the received first message belongs;
associating the received first message with the identified conversation;
comparing a best score from the second set of one or more scores to the threshold, wherein the determination that the second message belongs to any of the plurality of conversations is based on the best score failing to reach the threshold;
creating a new conversation that is added to the plurality of conversations upon determination that the second message does not belong to any of the plurality of conversations; and
associating the received second message with the new conversation.

2. The information handling system of claim 1 wherein the actions further comprise:
processing each of the conversations included in the plurality of conversations, wherein, when selected, each of the conversations is processed by the neural network by:
loading a set of cell states in the neural network with a set of previously saved cell states that correspond to the selected one of the conversations;
passing the EME data through the neural network after the loading of the set of cell states; and
receiving a score that indicates an affinity of the selected conversation to the received first message, wherein the score is one of the first set of one or more scores, wherein each of the first set of one or more scores is associated with one of the conversations; and

selecting a best score from the first set of one or more scores, wherein the conversation associated with the best score is identified as the conversation to which the first message belongs.

3. The information handling system of claim 1 wherein the actions further comprise:
initializing the neural network with zero context;
passing the second set of EME data through the initialized neural network, resulting in a new set of cell states being stored in the neural network and output of a new score;
initializing the new conversation messages with the received message; and
initializing a set of new conversation cell states with the new set of cell states, wherein the new conversation messages and the set of new conversation cell states are included in the new conversation.

4. The information handling system of claim 1 wherein the deep analytic analysis further comprises:
performing a topic embedding analysis on the received message, wherein the topic embedding analysis identifies a topic of the received message, a content of the received message, and a vocabulary of the received message;
performing a semantic embedding analysis on the received message, wherein the semantic embedding analysis includes a question analysis, a statement analysis, a positive/negative feedback analysis, and a structural analysis; and
performing an auxiliary information analysis on the received message, wherein the auxiliary information analysis includes a user participation analysis, a user mentioned analysis, a social score analysis, and a topic score analysis.

5. The information handling system of claim 4 wherein the actions further comprise:
generating a first vector from the topic embedding analysis, a second vector from the semantic embedding analysis, and a third vector from the auxiliary information analysis; and
combining the first, second, and third vectors to a form that is suitable as input to the neural network.

6. A computer program product stored in a computer readable storage medium, comprising computer program code that, when executed by an information handling system, performs actions comprising:
receiving a first message and a second message;
applying a deep analytic analysis to the received messages to form a first set and a second set of enriched message embedding (EME) data;
processing each conversation included in a plurality of conversations, wherein, when selected, each of the conversations is processed by a trained neural network by:
loading a set of cell states in the neural network with a set of previously saved cell states that correspond to the selected conversation;
passing the first and second sets of EME data to the trained neural network;
receiving a first set of scores from the trained neural network that indicates an affinity of the selected conversation to the received first message and a second set of scores from the trained neural network that indicates an affinity of the selected conversation to the received second message;

identifying, based on comparing a threshold to the first and second sets of scores, one of the conversations to which the received first message belongs;
associating the received first message with the identified conversation;
comparing a best score from the second set of one or more scores to the threshold, wherein the determination that the second message belongs to any of the plurality of conversations is based on the best score failing to reach the threshold;
creating a new conversation that is added to the plurality of conversations upon determination that the second message does not belong to any of the plurality of conversations; and
associating the received second message with the new conversation.

7. The computer program product of claim 6 wherein the actions further comprise:
processing each of the conversations included in the plurality of conversations, wherein, when selected, each of the conversations is processed by the neural network by:
loading a set of cell states in the neural network with a set of previously saved cell states that correspond to the selected one of the conversations;
passing the EME data through the neural network after the loading of the set of cell states; and
receiving a score that indicates an affinity of the selected conversation to the received first message, wherein the score is one of the first set of one or more scores, wherein each of the first set of one or more scores is associated with one of the conversations; and

selecting a best score from the first set of one or more scores, wherein the conversation associated with the best score is identified as the conversation to which the first message belongs.

8. The computer program product of claim 6 wherein the actions further comprise:
initializing the neural network with zero context;
passing the second set of EME data through the initialized neural network, resulting in a new set of cell states being stored in the neural network and output of a new score;
initializing the new conversation messages with the received message; and
initializing a set of new conve

Assignees

• IBM

Inventors

• Conley Devin A
• Moraes Priscilla S
• Krishnamurthy Lakshminarayanan
• Sar-Shalom Oren

Classifications

• G06N3/08 Primary

  Learning methods · CPC title

• H04L51/216 Primary

  Handling conversation history, e.g. grouping of messages in sessions or threads · CPC title

• H04L51/04

  Real-time or near real-time messaging, e.g. instant messaging [IM] · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title