A method and product for message mapping and combination for intent classification

Abstract

The present disclosure relates to message mapping and combining for intent classification. A method, computer system, and computer program product for automated agent intent detection enhancement are provided. A first message is received from a first user. The first message is generated during a first session between the first user and a first automated agent. A computer generates a second message that includes a same request as the first message but has a different linguistic modality than the first message. The second message and the first message are combined to form a combined message. The combined message is input to the first automated agent such that the first automated agent generates an intent classification for the first message.

Description

Background Technology

[0001] This invention generally relates to assisted automated agents that can interact with humans in online chat, audio sessions, or some other way. The automated agents are programmed to respond to user messages in an automated manner. Summary of the Invention

[0002] According to an exemplary embodiment, a computer-implemented method is provided, the method comprising receiving a first message from a first user. The first message is generated during a first session between the first user and a first automation agent. The computer generates a second message that includes the same request as the first message but has a different language morphology. The second message and the first message are combined to form a combined message. The combined message is input to the first automation agent, causing the first automation agent to generate an intent classification for the first message. Computer systems and computer program products corresponding to the above method are also disclosed herein. Attached Figure Description

[0003] These and other objects, features, and advantages of the present invention will become apparent from the following detailed description of illustrative embodiments thereof, which will be read in conjunction with the accompanying drawings. The various features in the drawings are not to scale, as the illustrations are provided to clearly facilitate understanding of the invention by those skilled in the art in conjunction with the detailed description. In the drawings:

[0004] Figure 1 The illustration depicts a networked computer environment according to at least one embodiment;

[0005] Figure 2A It is an operational flowchart illustrating an upgraded session collection process according to at least one embodiment;

[0006] Figure 2B The illustration shows an example according to at least one embodiment. Figure 2A The upgrade session collection process may involve chatbot sessions related to the upgrade.

[0007] Figure 3A The illustration shows a use according to at least one embodiment. Figure 2A The flowchart shown in the image illustrates the message enhancement process of the repository created during the process.

[0008] Figure 3B The illustration shows an example according to at least one embodiment. Figure 3A message enhancement process or Figure 3C The alternative message enhancement process for chatbot conversations in all aspects;

[0009] Figure 3C The illustration shows a use according to at least one embodiment. Figure 2A The diagram shows the workflow of the alternative message enhancement process for data collected during the upgraded session collection process.

[0010] Figure 4 According to at least one embodiment Figure 1 A block diagram depicting the internal and external components of a computer and server;

[0011] Figure 5 Includes embodiments according to this disclosure Figure 1 The diagram depicts an illustrative cloud computing environment with a networked computer environment; and

[0012] Figure 6 According to embodiments of this disclosure Figure 5 A block diagram illustrating the functional layers of an illustrative cloud computing environment. Detailed Implementation

[0013] This document discloses detailed embodiments of the claimed structures and methods; however, it is to be understood that the disclosed embodiments are merely illustrative of the claimed structures and methods, which may be embodied in various forms. The invention may be embodied in many different forms and should not be construed as limited to the exemplary embodiments set forth herein. Rather, these exemplary embodiments are provided to make this disclosure thorough and complete, and to fully convey the scope of the invention to those skilled in the art. In the description, details of well-known features and techniques may be omitted to avoid unnecessarily obscuring the presented embodiments.

[0014] The exemplary embodiments described below provide a method, computer system, and computer program product for improving intent detection in automated agents, such as chatbots. The use of automated agents to handle various user interactions is increasing in various commercial, governmental, and other settings. Automated agents can receive and respond to queries from humans. Automated agents are configured to provide unmanned dialogue services that humans can participate in.

[0015] This interaction between automated agents and humans can occur via text exchange between humans and automated agents. Specifically, in some examples, humans can type text or type on a keyboard at a computer, such as a cellular phone, to generate text in text boxes on the computer's screen. Automated agents participating in such text conversations can be referred to as chatbots. Human-automated agent interaction can additionally and / or alternatively occur via voice or sign language conversations, where humans speak or gesture, sound is recorded via a microphone or video camera, text is generated from the sound / gesture, and the computer analyzes the text. Human-automated agents can use artificial intelligence and / or machine learning to implement natural language processing in order to analyze input linguistic information such as text. The software of the automated agent can allow the automated agent to interpret input text information and automatically generate appropriate responses to human queries. The automated agent interprets the intent of the message / query sent by the human. The response generated by the automated agent depends on the interpretation given to the incoming message by the automated agent. The software of the automated agent may include decision trees, data storage devices, entity detection for natural language processing, intent detection, natural language processing, artificial intelligence, and / or machine learning to help understand the user's query. Owners / managers can use organizational information categorized into topics to train automated agents, which can then use these topics to assist users.

[0016] Within the context of a conversation, humans can exhibit random behavior and produce random sentences or statements. A single purpose of a query can be presented using a vast array of human-generated word formats and expressions, negotiated with an automated agent. The various formats and / or expressions used can vary depending on who is speaking and the party they are conversing with. Two statements / sentences / questions can have different organizational structures and use different verbs, nouns, and / or adjectives, but still have similar intent. Two queries can have different organizational structures and use different verbs, nouns, and / or adjectives, but still provide the same or similar requests.

[0017] Automated agents may be challenged to interpret queries with unique wording and generate appropriate responses that will answer human queries. This embodiment generates improved input to be provided to the automated agent, thereby enhancing its ability to interpret user queries and respond appropriately to humans. These enhancements will help lead to increased human satisfaction with their interactions with the automated agent and will help save resources for human agents that would otherwise be needed as backups when the automated agent fails to meet human queries seeking answers or information from the organization.

[0018] This embodiment technically reflects an observation that individuals seeking customer service via technologies such as web chat or voice conversations communicate more precisely and clearly with automated agents than typically with human agents. Messages provided by humans to automated agents may be more ambiguous and / or shorter than corresponding messages provided by humans to human agents. It has been observed that when someone is notified that they are now participating in a conversation conducted by a human agent, that person typically provides more information and context for their query, such as multiple sentences, compared to when that person types, speaks, or gestures a message to the automated agent. Humans may make this modification based on the assumption that the automated agent is unlikely to understand complex statements or requests. In this embodiment, a message destined for the automated agent is received, a prediction is made of how or possibly how a message with the same request will be sent to the human agent, and then a combination of the original message and the predicted message is input, for example, fed into the automated agent. The predicted message may have a different linguistic form than the original message, such as different word choice. Requests destined for humans may display a longer linguistic form. Requests destined for humans may display a more detailed linguistic form with more words. For a first phrase (e.g., a sentence or paragraph) to have different linguistic morphology, the first phrase must have at least one word that differs from the second phrase. The two phrases may include some of the same words, but will have at least one different word. Human-to-human request terminology is mapped to human-to-automation agent terminology. The combination of the two messages constitutes an enhancement of the first message. This embodiment makes it possible for the automation agent to better understand and respond to the enhanced message, which is richer in information than the original message alone.

[0019] This embodiment can be implemented as a supplement to existing automated agents and their architectures without requiring any retraining and / or reconstruction of the automated agent itself or its underlying data. Therefore, this embodiment offers implementation / installation benefits. Specifically, the embodiment can be implemented and installed in a more flexible manner without interfering with the operation of the automated agent. This embodiment can be implemented without building new automated agents and / or virtual assistants. There is no need to prompt human participants in the human-machine conversation with additional information or additional paraphrases of the original message. Therefore, this embodiment can be implemented to enhance robot intent detection and performance while preserving the robot architecture and robot training data for the deployed robot. This avoidance of robot architecture reconstruction is particularly valuable for large-scale models. Due to the more informative augmented messages, automated agents (e.g., robots) are able to improve the accuracy of their correct intent predictions. This embodiment can be embodied as a supplement to existing automated agents and / or virtual assistants. This embodiment includes extracting and utilizing user-to-robot requests, as well as extracting and utilizing user-to-human agent requests during the data collection phase.

[0020] Therefore, this embodiment can utilize various automation techniques, such as lookup tables, machine learning, text comparison, word-based semantic similarity comparison, and / or other artificial intelligence, to enhance messages outside the robot's architecture. Thus, this embodiment can enhance intent detection performed by the automated agent and improve the automated agent's ability to appropriately respond to users seeking assistance. These benefits can be obtained using modules existing outside the robot's architecture.

[0021] refer to Figure 1 The document describes a networked computer environment 100 according to one embodiment. The networked computer environment 100 may include a first user computer 102a1, a second user computer 102a2, a first server 112, a second server 122, and a human agent computer 132. The first user computer 102a1, the second user computer 102a2, and the first server 112 may be hosted and enabled to run dialog programs 110a1, 110a2, and 110b, respectively, wherein dialog programs 110a1 and 110a2 on the first and second user computers 102a1 and 102a2 occur on the client side of the dialog, and dialog program 110b on the first server 112 occurs on the automated agent side of the dialog. The first server 112 and the second server 122 may each be a computer. The human agent computer 132 may also be hosted and enabled to run dialog program 110c, specifically for the human agent side of the dialog. Various computers in the networked computer environment 100, such as the first user computer 102a1, the second user computer 102a2, the first server 112, the second server 122, and the human agent computer 132, can communicate with each other via the communication network 116.

[0022] The first user computer 102a1, the first server 112, and the human agent computer 132 can participate in conversations that can be collected and analyzed as part of an upgraded session collection process 200. Figure 2A The upgraded session collection process 200 is described in the preceding text and will be further described later. Other user computers may also participate in sessions or conversations that can be collected and analyzed as part of the upgraded session collection process 200, but for simplicity, Figure 2A The session collection process 200 for the upgrade is shown in the document and the steps described herein are described as occurring relative to the first user computer 102a1.

[0023] The second server 122 can be used to execute... Figure 3AThe message enhancement process 300, depicted and described later, can be used to enhance messages subsequently provided by a human using a second user computer 102a2 and participating in an unmanned dialogue with an automated agent based on a first server 112 having a dialogue program 110b. The second server 122 can receive and / or intercept messages provided by this other human and enhance them so that when the enhanced message is received by the automated agent at the first server 112, the automated agent can better interpret and appropriately respond to the message (because the message has been enhanced). Although Figure 1 Two servers are depicted, namely a first server 112 and a second server 122. However, in some embodiments, the message enhancement program 120, the natural language processor 128, and the upgraded session store 125 may reside in the same server where the automated agent implemented by the conversational program 110b—the automated agent side—may reside, for example, in the first server 112.

[0024] Networked computer environment 100 may include many computers and many servers, although Figure 1 The diagram shows three computers and two servers. The communication network 116, which allows communication between the first user computer 102a1, the second user computer 102a2, the first server 112, the second server 122, and the human agent computer 132, can include various types of communication networks, such as the Internet, wide area network (WAN), local area network (LAN), telecommunications network, wireless network, public switched telephone network (PTSN), and / or satellite network.

[0025] It should be understood that Figure 1 This is merely an illustration of one implementation and does not imply any limitation on the environment in which different embodiments may be implemented. Many modifications can be made to the depicted environment based on design and implementation requirements.

[0026] Communication network 116 may include connections such as wired or wireless communication links or fiber optic cables. Although Figure 1 The networked computing environment 100 shown depicts two servers, but the communication network 116 itself may include additional servers, such as one or more network edge servers and one or more edge / gateway servers, which may be enabled to run or assist in the operation of the upgraded session collection process 200, message enhancement process 300, and alternative message enhancement process 370. In some embodiments, the communication network 116 may be or include a high-speed network, such as a 4G or 5G network. Implementing the upgraded session collection process 200, message enhancement process 300, and alternative message enhancement process 370 in a 5G network would enable at least some embodiments to be implemented at the edge to improve network performance.

[0027] The first user computer 102a1 may include a first processor 104a1, a first data storage device 106a1, and a first software program 108a1. The first software program 108a1 may be stored on the first data storage device 106a1. The first processor 104a1 may access the first data storage device 106a1 to launch the first software program 108a1. The dialogue program 110a1 may be an example of the first software program 108a1 stored on the first data storage device 106a1 and executable via the first processor 104a1. The dialogue program 110a1 may facilitate communication with an automated agent and may include the generation of text boxes / chat boxes, voice communication platforms, and / or video communication platforms. The dialogue program 110a1 may also implement speech-to-text transcription for audio conversations and gesture-to-text transcription for sign language conversations, so that text is transmitted over the communication network 116 instead of audio / video files. The dialogue program 110a1 may implement natural language processing and / or other artificial intelligence to achieve such transcription, so as to convert spoken language and / or sign language into text and vice versa. Conversely, the dialogue program 110a1 can also perform text-to-speech transcription and text-to-gesture transcription to allow any audio and / or sign language messages received from a human agent to be converted into text for human understanding / reading at the first user computer 102a1.

[0028] The above description provided for the first user computer 102a1 and its components and connections is also equivalent to the second user computer 102a2 and its components and / or software, namely the processor 104a2, data storage device 106a2, software program 108a2, and dialog program-client 110a2. Examples of the use of the second user computer 102a2 will be provided for… Figures 3A-3C The embodiments shown are described herein.

[0029] The first server 112 can store and run a dialogue program 110b for the automated agent side. The first server 112 may include a processor, a session store 115, and a natural language processor 118. Sessions occurring as part of the execution of dialogue programs 110a and 110b can be stored in the data storage device of the session store 115 in the first server 112. Dialogue program 110b can invoke the natural language processor 118 to read and interpret text received in messages from humans at the first or second user computers 102a1 and 102a2, which is received as part of an unmanned dialogue (from the perspective of the automated agent). Such dialogues involving the automated agent can occur between a human at the first user computer 102a1 and dialogue program 110b at the first server 112, or between the same or another human at the second user computer 102a2 and dialogue program 110b at the first server 112. Dialogue program 110b can also perform speech-to-text transcription for audio sessions and gesture-to-text transcription for sign language sessions.

[0030] The second server 122 can store and run a message enhancement program 120 that can enhance one or more messages generated at a client / user computer (such as the second user computer 102a2) and sent to an automation agent (dialogue program 110b - the automation agent side) at the first server 112. The message enhancement program 120 can intercept / receive these messages via a communication network 116 to enhance them. The second server 122 may include an upgraded session store 125, which includes data storage devices and can host a subset of sessions stored in the session store 115. In some cases, the upgraded session store 125 may also include a machine learning model trained via inputs from identified upgraded sessions. Figure 2A The upgraded session collection process 200 depicted can be executed on sessions from the session repository 115 in the first server 112 to collect sessions with matching requests, such that these sessions with matching requests can be stored in the upgraded session repository 125. The second server 122 may include a second processor 124 that can be used to execute the message enhancement program 120. In some embodiments, the second server 122 may include a natural language processor 128 to facilitate message enhancement performed by the message enhancement program 120. The message enhancement program 120 may invoke the natural language processor 128 to read and interpret text received in a received intercepted message to enhance that message. The message enhancement program 120 may also implement speech-to-text transcription for audio sessions and gesture-to-text transcription for sign language sessions.

[0031] The reference to intercepting messages can be articulated within the concept that an organization controlling an automated agent can consciously consent to the invocation of message enhancement program 120 to improve its automated agent's ability to satisfy human queries. Therefore, messages received or to be received by the conversational program 110b at the first server 112 can be redirected, rerouted, transmitted, and / or forwarded to message enhancement program 120 at the second server 122, such that incoming messages can be enhanced before being input into an automated agent (e.g., a chatbot).

[0032] The human agent computer 132 can store and run dialogue programs 110c for the human agent side. The human agent computer 132 can participate as... Figure 2A The upgraded session is part of the upgraded session collection process 200 described herein and subsequently. In cases where the automated agent (dialogue program 110b—the automated agent side) at the first server 112 fails to successfully understand or effectively respond to a human query, the human agent can use the human agent computer 132, and specifically the dialogue program 110c, for example, at the first user computer 102a1, to respond to the human query. The human agent computer 132 and the human agent operating thereon are unnecessary and, at least in most cases, not used for execution. Figure 3A The message enhancement process described in section 300 Figure 3B Chatbot conversation 350 as depicted in the text and Figure 3C The alternative message enhancement process described in 370.

[0033] The human agent computer 132 may include a third processor 134 and a data storage device 136. Software programs 138 (such as a conversational program 110c) may be stored on the data storage device 136. The third processor 134 may access the data storage device 136 and thereby launch the software program 138. The conversational program 110c may be an example of a software program 138 stored on the data storage device 136 and executable via the third processor 134. The conversational program 110c may facilitate communication with a human agent and may include the generation of text boxes / chat boxes, voice communication platforms, and / or video communication platforms. The conversational program 110c may also implement speech-to-text transcription for audio conversations and gesture-to-text transcription for sign language conversations, enabling the transmission of text instead of audio / video files via the communication network 116. Conversely, the conversational program may also implement text-to-audio conversion and / or text-to-gesture (with video) conversion to allow intervention by human agents who rely on audio and / or sign language (video) for communication.

[0034] For reference Figure 4As discussed, the first server 112 and the second server 122 may each include an internal component 402a and an external component 404a, respectively. The first user computer 102a1, the second user computer 102a2, and the human agent computer 132 may also each include, for example... Figure 4 The internal component 402b and external component 404b are depicted. The first and second servers 112, 122 can also operate in cloud computing service models, such as Software as a Service (SaaS), Platform as a Service (PaaS), or Infrastructure as a Service (IaaS). The first and second servers 112, 122 can also reside in cloud computing deployment models, such as private clouds, community clouds, public clouds, or hybrid clouds. The first user computer 102a1, the second user computer 102a2, and the human agent computer 132 can each be, for example, a mobile device, telephone, personal digital assistant, netbook, laptop computer, tablet computer, desktop computer, or any type of computing device capable of running programs, accessing networks, and communicating with other servers and / or with another server (e.g., the first server 112 and / or the second server 122) that is remotely located relative to the former. The first user computer 102a1, the second user computer 102a2, and the human agent computer 132 can each include a display screen, speakers, microphone, camera, and keyboard or other input devices to enable better communication for humans participating in customer query sessions as clients or as human agents. According to various implementations of this embodiment, the message enhancement program 120 can interact with the upgraded session store 125 and the session store 115 that can be embedded in various storage devices, such as, but not limited to, various computers / mobile devices in the network, the first server 112 and / or the second server 122, or another cloud storage service.

[0035] As described herein, the use of content storage on edge servers can reduce the network traffic required to execute customer queries between human clients and automated agents. This reduction in network traffic can facilitate efficient processing for executing the methods according to this embodiment. While the client participates in the session at the first user computer 102a1 and / or the second user computer 102a2, the client, the automated agent at the first server 112, and the messaging enhancement program 120 at the second server 122 can leverage their network infrastructure to obtain appropriate connectivity to the environment, such as 5G connectivity. This embodiment can utilize existing and future 5G infrastructure and its increased bandwidth and latency, as well as the scaling of applications requiring large amounts of real-time data. The first and second servers 112, 122 can trigger a flow of data and commands to be processed by a distributed program available at one or more network edge servers located at the network edge and / or at a corresponding edge / gateway server located at the network gateway.

[0036] The computer system with message enhancement program 120 operates as a dedicated computer system that can assist in performing message enhancement processes 300 to facilitate automated agent understanding and replace message enhancement process 370. Specifically, message enhancement program 120 transforms the computer system into a dedicated computer system compared to currently available general-purpose computer systems on which message enhancement program 120 is not installed.

[0037] Replacement Figure 1 As shown, deployed in a separate server, namely the second server 122, the messaging enhancement program 120 can alternatively be implemented as a software plug-in component on one, some, or all of the first user computer 102a1, the second user computer 102a2, and the first server 112, where humans and automated agents participate in a session. If the messaging enhancement program 120 is configured within the first server 112 and / or various user computers, such as those using the conversational program 110a from the client's or employee's side to process the first and second user computers 102a1, 102a2 of their employer organization, a reduction in network traffic can be achieved.

[0038] Now for reference Figure 2A The flowchart illustrates an upgraded session collection process 200, which can be performed using message enhancement program 120 according to at least one embodiment. Message enhancement program 120 can perform the upgraded session collection process 200 by participating in dialogue programs 110a1, 110a2, 110b, and 110c. This upgraded session collection process 200 can help generate a message enhancement library including modified messages. The message enhancement library may include upgraded dialogues with matching requests for human-to-automation agent dialogue portions and human-to-human agent dialogue portions. The message enhancement library, requests, and / or matching dialogue portions may be stored in an upgraded session store 125 in a second server 122, i.e., a server in which message enhancement program 120 is stored. Figure 3A As part of the message enhancement process 300 described herein, restated or paraphrased messages can be used to help enhance messages. Paraphrased messages can also be used to train machine learning models that can be used to help enhance messages, as... Figure 3C This is part of the alternative message enhancement process 370 described in the text. The upgraded session collection process 200 helps identify good human-to-automation agent and human-to-human agent session pairs by using factors such as whether the two parts come from the same session and / or from the same user.

[0039] In step 202 of the upgraded session collection process 200, a set of sessions between the user and the automated agent is obtained. These sessions can be obtained separately via... Figure 1The specific dialogue programs 110a1, 110b, and 110c shown occur on the first user computer 102a1, the first server 112, and the human agent computer 132. Each message generated and sent by dialogue programs 110a1, 110b, and 110c can be stored in a local data storage device and / or a cloud storage device, such as in a session store 115 on the first server 112. As the effectiveness of various embodiments increases with more data, the sessions obtained in step 202 can include those from many different humans who use corresponding computers to communicate with the automated agent, such as with the dialogue program—the automated agent side 110b. For simplicity, the upgraded session collection process 200 will be described using a specific example relating to the user at the first user computer 102a1. As an additional supplement to the session set, other dialogues between the automated agent and humans, and between human agents and humans, can be recorded elsewhere and added as a supplement to the session store 115 to complement the existing data. Figure 1 The computer-generated sessions shown are examples. As a default setting for the session collection process 200 used to facilitate upgrades, the automated agent can record all messages received, generated, and / or sent by the automated agent as part of a session with one or more humans. This will be described later. Figure 2B An example is shown that may include an upgraded session 250 from the initial session set obtained in step 202.

[0040] In at least some embodiments, a human can initiate a conversation with an automated agent by visiting an organization's website and activating a feedback link within the website. The website can then launch a conversational platform that enables dialogue between the human and the automated agent via text messages, audio messages, and / or sign language messages. Therefore, conversational programs 110a1, 110a2, 110b, and 110c can be incorporated into the website's software. The website can, for example, generate one or more graphical user interfaces using text boxes to facilitate human engagement of the automated agent in the conversation. Figure 2B An example of upgraded session 250 is shown, in which a human first speaks with the chatbot and then with a human agent.

[0041] The saved sessions can be stored in separate groups of message exchanges. Sessions can be stored such that each specific session runs from its beginning to its end, until the user exits the session, for example, by leaving the chat. The entire session may include escalation portions that occur due to the automated agent failing to meet and / or understand human needs, thus requiring the session to be upgraded to a human agent. Figure 2BThe escalating dialogue 250 shown includes an initial portion 258 between the human and the chatbot, and an escalating portion 260 between the same human and a human agent. If the automated agent of the chatbot 110b at the first server 112 cannot understand and / or satisfy the query from the human at the first user computer 102a1, the human working for and / or representing the organization can operate the chatbot 110c at the human agent computer 132 to communicate with the human at the first user computer 102a1. The portion of the session following the introduction of the human agent may be referred to as the escalating portion 260 and may be part of the set of sessions obtained in step 202.

[0042] The default setting for dialogue programs 110a1, 110a2, 110b, and 110c may be to record each session for the purpose of enhancing the performance of the automated agent. Additionally and / or alternatively, for the purpose of training and enhancing the automated agent and its ability to understand messages from humans, dialogue programs 110a1, 110a2, 110b, and 110c may request human consent to allow session recording via a graphical user interface.

[0043] The obtained session content can be stored in memory that is part of and / or accessible by the dialog programs 110a1, 110a2, 110b, and 110c. For example, the session can be saved in... Figure 1 In the session storage 115 shown, Figure 4 The RAM 408 shown, the memory of the server connected to the communication network 116, and / or other memory of the dialog programs 110a1, 110a2, 110b, 110c that can be accessed via the communication network 116 and / or via a wired connection.

[0044] In step 204 of the upgraded session collection process 200, a subset of sessions is identified. Each session in this subset was not resolved by the automated agent, and therefore the session is escalated to a human agent. This identification of the session subset in step 204 can occur by searching the set of sessions obtained in step 202. Each session in the set of sessions obtained in step 202 can be tagged with the participant's name and the outcome of the session. For example, each conversation can be tagged to indicate the computer identity of the human chatting with the automated agent, such as an IP address. Figure 1In the embodiment shown, the tag may indicate the IP address of the first user computer 102a1 or the second user computer 102a2. When the set contains sessions from various automated agents, the identity of the specific automated agent involved in a particular session may be part of the tag in the stored session data. This addition of such a tag to a session can also occur automatically by programmatic analysis of which parties are participating in the session. If the session requires an upgrade to a human agent, the participants and the tag may include a third identity, namely the identity of the human agent. This identity may be the IP address of the human agent's computer, for example, Figure 1 The IP address of the human agent computer 132 is shown in the figure.

[0045] The identification in step 204 may include filtering out any conversations from the start to the end of the conversation that do not include a human agent. For step 204, each conversation with a human agent tag may be selected to be placed in a subset. Therefore, in some embodiments, the identification in step 204 may include a text search via tag reading and a text comparison via a comparator, which may be part of message enhancement program 120 and / or conversation programs 110a1, 110a2, 110b, 110c.

[0046] In step 206 of the upgraded session collection process 200, user requests are identified for each human-to-automation agent segment within a subset. This subset refers to the set of upgraded sessions identified in step 204 that include sessions upgraded to human agents. However, step 206 involves the initial portion of the session before it is upgraded to a human agent. For example, for Figure 2B The upgraded dialogue 250 shown in the figure, step 206 relates to the start portion 258. Step 206 helps to identify the purpose of the message originally sent by a human (e.g., a customer) to the automation agent, such as from the dialogue program 110a1 of the first user computer 102a1 to the dialogue program 110b (automation agent) of the first server 112.

[0047] The conversational portion, such as messages, may include some irrelevant elements as well as some substantial user utterances. This user request identification in step 206 may include filtering out some or all irrelevant elements that could be part of the conversation's human-to-automation agent portion. Irrelevant elements may include trivial or superficial conversational elements and may include all greetings. The automation agent can easily understand and appropriately respond to words such as "hello," "hi," "hey," "greetings," "thumbs up," "morning," "afternoon," "my," "name," "yes," "live," "chat," and "thank you," which can be examples of irrelevant elements in the message. The identification in step 206 is part of a larger filtering process used to find conversations where the automation agent needs assistance in understanding and responding. The larger filtering may include steps 208, 210, and 212 of the upgraded conversation collection process 200. This filtering helps identify conversations that may challenge the automation agent in understanding and effectively responding to human-generated messages. The identification in step 206 may include identifying the first user utterances in the conversation portion that are not irrelevant elements. A text comparator, as part of the message enhancement process 120, can analyze the text of each word in the start portion 258 to look for word matches in a list of irrelevant elements. If a start portion 258 exists that consists entirely of irrelevant elements, that session portion can be discarded due to its unsuitability for use in the session store 125 for upgrades and for use in the message enhancement process 300.

[0048] In at least some cases, the first user utterance in the message that is not an irrelevant element can itself be considered a user request within the session segment. The identification in step 206 may include identifying multiple user requests within a single session segment, so that none of the user requests are filtered out as irrelevant elements.

[0049] This identification in step 206 can be performed by message enhancement program 120 at the second server 122.

[0050] In step 208 of the upgraded session collection process 200, a user request is identified for the corresponding human-to-human agent portion of the upgraded session. This human-to-human agent session portion may, at least in some embodiments, originate from the same session that began in the human-to-automation agent session analyzed in step 206. If the automation agent does not understand or cannot resolve the human's query, the session is subsequently upgraded to a human agent, such as a human using human agent computer 132. Therefore, step 208 relates to the later portion of the upgraded session, such as user messages generated after the session has been upgraded to a human agent. Figure 2BThe upgraded session 250 shown includes an upgraded section 260, which illustrates the session between the human user and the human agent and occurs after the start section 258. Step 208 helps identify the purpose of a message sent by a human (e.g., a customer) to the human agent, such as from a conversational program 110a1 on the first user computer 102a1 to a conversational program 110c on the human agent computer 132.

[0051] The conversation portion, such as messages, can include some irrelevant elements and some substantial user utterances for this human-to-human agent portion. This user request identification in step 208 can include filtering out some or all of the irrelevant elements that may be part of the human-to-human agent portion of the conversation. Irrelevant elements can be defined, analyzed, and filtered out in the same manner as in the analysis start portion 258, which is part of step 206. The automated agent can easily understand and appropriately respond to words such as “hello,” “hi,” “hey,” “greeting,” “thumbs up,” “morning,” “afternoon,” “my,” “name,” “yes,” “live,” “chat,” and “thank you,” which can be examples of irrelevant elements in the message. This identification in step 208 is part of a larger filtering process to remove conversations that the automated agent does not need assistance to understand and respond to, thereby leaving those conversations that include escalating conversations. This larger filtering process can also include steps 206, 210, and 212 of the escalating conversation collection process 200. This filtering helps to identify conversations that may challenge the automated agent to understand and effectively respond to human-generated messages. The identification in step 208 may include identifying the first user utterance that is not an irrelevant element in the upgraded section 260. If the session section consists entirely of irrelevant elements, the session section and / or the session may be discarded because it is not suitable for use in the upgraded session store 125.

[0052] In at least some cases, a first user utterance that is not an irrelevant element in the human-to-human agent information can itself be considered a user request for that session segment, for example, for the upgrade segment 260. The identification in step 208 may include identifying multiple user requests within that single session segment, such as the upgrade segment 260, so that each user request is not filtered out as an irrelevant element.

[0053] This identification in step 208 can be performed by message enhancement program 120 at the second server 122.

[0054] For steps 206 and 208, in some embodiments, the identified non-irrelevant utterances may themselves be requests. In other embodiments, the identified non-irrelevant utterances may be input into a lookup table, decision tree, and / or machine learning model to obtain an output identifying user requests. In some embodiments, the set of non-irrelevant utterances identified in the analyzed message constitutes a user request. In other embodiments, a single message may include multiple user requests, each corresponding to a plurality of utterances. Figure 2B In the illustrated embodiment, the user request in the initial message 254 could be "vacation" assistance. The user request in the paraphrased message 256 could be "reactivate account". In some embodiments, the meaning of the acronym "LOA" may not be incorporated into the user request identified for the paraphrased message 256, because the machine learning model or automated agent may not be familiar with the acronym during the analysis.

[0055] In step 210 of the upgraded session collection process 200, a first identified user request is compared with a second identified user request. The first identified user request may refer to the user request identified in step 206 for the human-to-automated agent session portion. The second identified user request may refer to the user request identified in step 208 for the human-to-human agent session portion. This comparison in step 210 can be performed by message enhancement program 120. This comparison may include text comparison and semantic similarity comparison. Text comparison can be performed via a comparator of message enhancement program 120. For semantic similarity comparison, the requests can also be converted into word-based vectors via natural language processing, and the vectors can be compared. Vectors that are sufficiently close to each other (e.g., closer than a predetermined threshold) can be considered semantically matched.

[0056] In step 212 of the upgraded session collection process 200, a determination is made regarding whether the similarity is greater than a predetermined threshold. If the determination is affirmative and the similarity is greater than the predetermined threshold, the upgraded session collection process 200 proceeds to step 214. If the determination is negative and the similarity is less than the predetermined threshold, the upgraded session collection process 200 proceeds to step 216. In some embodiments, the predetermined threshold may be selected by the organization's administrator and may include weighting factors that emphasize accuracy and / or allow for processing a wider variety of message inputs. A similarity score for the analysis in step 212 may be generated during the comparison in step 210.

[0057] In step 214 of the upgraded session collection process 200, sessions are added to a repository for future intent detection enhancements. Because a session is considered suitable for message enhancements to help future human-to-automation agent sessions, it can be added to an upgraded session repository 125, which can constitute a repository for future intent detection enhancements. Sessions can be stored in the upgraded session repository 125 in such a way that paraphrased message 256 and / or its user request is linked to the initial message 254 and / or its user request. In some embodiments, the upgraded session repository 125 can be used as a lookup table.

[0058] In step 216 of the upgraded session collection process 200, sessions are not added to the storage for future intent detection enhancement. This step 216 occurs for session portions that do not have a user request similarity greater than a predetermined threshold determined in step 212. These sessions may remain in session storage 115 or may be deleted from session storage 115. Because these sessions are considered unhelpful for message enhancement in future human-to-automation agent sessions, they are not added to the upgraded session storage 125 that constitutes the storage for future intent detection enhancement. Moreover, in other embodiments, these sessions may not be used to train a machine learning model that predicts human-to-human phrasing based on receiving human-to-automation agent messages. In some cases, these purged sessions may include message portions that are entirely or primarily irrelevant.

[0059] Figure 2B The illustration shows an example according to at least one embodiment. Figure 2A The upgraded session collection process 200 may involve upgraded sessions 250 chatbot sessions. Figure 2B A text box 252 is shown that can appear on the screen of a human computer (e.g., on the screen of a first user computer 102a1). Within text box 252, a message is shown as part of a conversation when a person seeks technical support first from an automated agent, i.e., a chatbot, and secondarily from a human agent. The human at the first user computer 102a1 can type on a keyboard or other input device connected to the first user computer 102a1, which will cause an initial message 254 to appear on the screen of the first user computer 102a1. This initial message 254 may query the chatbot for help on how to handle leave in the system. In this example, the chatbot may be established by a company seeking to provide advice (e.g., human resources (HR) advice) to its own employees on how to handle various employment issues such as leave.

[0060] The upgraded session 250 includes a start portion 258 in which the chatbot communicates with the user. This chatbot can be powered by, for example... Figure 1 The chatbot is illustrated by the automated agent side 110b located at the first server 112. In this example, the chatbot does not understand human queries, such as the initial message 254, and therefore notifies the human agent that it should attempt to respond to the human query.

[0061] The upgraded session 250 thus moves from the initial section 258 to the upgraded section 260, where a human agent from the company / organization, instead of a chatbot, communicates with the user. Figure 1 In the depicted example, the human agent is using a dialog program—human agent side 110c—at human agent computer 132 to provide information for a human conversation at first user computer 102a1. The human agent can type on a keyboard connected to human agent computer 132 to generate messages for the user / client. When the human agent engages the human, the human provides a paraphrased message 256 via first user computer 102a1 seeking guidance on how to programmatically respond to and / or record changes in work status, in this case, the worker has returned from leave. Paraphrased message 256 is more precise and clearer than the initial message 254, and helps the human agent understand more specifically what help the human wants. Such paraphrasing can occur unintentionally by the human, but it happens frequently in such escalated conversations. The human agent responds to paraphrased message 256 by providing answers on how the user can reactivate their account upon returning from leave.

[0062] By performing an upgraded session collection process 200 on the upgraded session 250, the message enhancement program 120 determines that the upgraded session 250 provides valuable paraphrasing for enhancing future incoming messages. Therefore, the pair of initial message 254 and paraphrased message 256 will be suitable for storage for future use by the message enhancement program 120 to enhance future initial messages provided to automated agents such as chatbots. This message pair and its user request can be stored in the upgraded session store 125 for availability to the message enhancement program 120 to enhance future initial messages received by chatbots.

[0063] The upgraded session collection process 200, for each upgraded session within the subset identified in step 204, has natural repetitions of steps 206, 208, 210, and 212, as well as 214 or 216. Through this repetition, and optionally by performing the upgraded session collection process 200 for other sets and subsets, the upgraded session repository 125 can ultimately contain hundreds, thousands, hundreds of thousands, or more matching messages / upgraded sessions. This large number can be based on the number of times the upgraded session collection process 200 is repeated and the number of sessions fed into the upgraded session collection process 200. The longer the time required to collect such possible sessions, the more likely it is to result in an increase in the number of upgraded sessions added to the repository in the upgraded session collection process 200.

[0064] A dialogue can be represented by A:=(u_1,u_2,...,u_n) and can include ordered utterances indicated by u_i. Each utterance in the dialogue can be created by an automated agent (e.g., a bot), a human, or a human agent. The sets of utterances for the bot, human, and human agent can be represented by UB, U, and UH, respectively. F can represent a predefined set of irrelevant words and phrases such as “hello,” “hi,” “hey,” “greeting,” “thumbs up,” “morning,” “afternoon,” “my,” “name,” “yes,” “live,” “chat,” and “thank you.” Given a human-to-bot session A_b and its subsequent human-to-bot upgrade dialogue A_h, two corresponding user requests can be extracted. Extraction heuristics can include finding the first non-irrelevant user utterance in the human-to-bot session A_b. The first user utterance can be represented as r_b. Extraction heuristics can also include finding the first non-irrelevant user utterance in the human-to-human agent session A_h. The first user utterance can be represented as r_h. If the similarity between r_h and r_b is greater than (>) the minimum similarity parameter, then two requests can be returned for use in the message augmentation dataset and stored in the upgraded session store 125. The message augmentation dataset can alternatively be used to train a machine learning model that predicts message phrasing for a human receiver based on input given to the automated agent receiver. If a session does not include phrasing that meets the criteria, then that progressively upgraded session is not subsequently added to the message augmentation dataset. This heuristic can be used to extract multiple request pairs from a dataset with multiple sessions, which can then be used to augment incoming messages for the automated agent / bot.

[0065] The message enhancement program 120 can perform various aspects of natural language processing to execute various aspects of the upgraded session collection process 200, such as those steps involving identifying request parts of the session. Natural language processing may include entity extraction, which extracts named entities from text and categorizes them into predetermined categories. Entities can be considered as non-irrelevant parts. The message enhancement program 120 may also use natural language processing to extract verbs from various messages to identify the request / intent of a user query.

[0066] Figure 3A A flowchart illustrating the operation of a message enhancement process 300 according to at least one embodiment is provided. The message enhancement process 300 can access... Figure 2A The image shows the repository created during the previously described upgraded session collection process 200. To further illustrate some steps of the message enhancement process 300, Figure 3B The illustration shows a chatbot session 350, which is an example of a message enhancement process 300 being implemented.

[0067] In step 302 of the message enhancement process 300, a first message is received from the first user as part of a human-to-automation agent session. This first message can be sent by the user... Figure 1 The first message is generated by the user at the dialog program 110a2 on the second user computer 102a2 shown. While this first message can be generated from any user computer involved in the automation agent, and even from the first user computer 102a1, which is described as being involved in the upgraded session collection process 200, typically a single user will not identify the automation agent and query it with the same question on different occasions. The computers shown as generating various messages are examples of implementing this embodiment and do not limit the various possible implementations of this embodiment. This display of both the first user computer 102a1 and the second user computer 102a2 helps illustrate how a user's previous session with the automation agent can be used via the upgraded session collection process 200 and message enhancement process 300 to help the automation agent better participate in subsequent sessions with another user.

[0068] In step 302, the message enhancement program 120 at the second server 122 can receive this first message. The dialog program 110b may have a redirection feature, causing incoming messages received via the dialog program 110b—representing the automation agent—to be redirected to the message enhancement program 120 through the communication network 116. Such redirection, forwarding, or bypassing allows the message enhancement program 120 to be able to receive messages submitted to automation agents such as chatbot 366 (see...). Figure 3BThe message enhancement program 110a2 can also initiate such a redirection. Message enhancement program 120 is shown in a second server 122, separate from the first server 112. In other embodiments, message enhancement program 120 may be configured together with conversation program 110b—the automation agent side—within the first server 112.

[0069] Figure 3B An example of a first message 354 generated by a human as part of a chatbot session 350 is shown. A user at a second user computer 102a2 can type words on a keyboard connected to the second user computer 102a2, which generates the words in a graphical user interface text box 352. This graphical user interface text box 352 can be generated when a user participates in the conversation program 110a2, for example, when visiting an employer's website and accessing the employee section and / or feedback section. Figure 3B The first message 354 is shown in the image. The first message 354, received by the message enhancement program 120, uses... Figure 3B The first transport arrow 356 indicates this. This first transport arrow 356 begins at the first message 354 in the graphical user interface text box 352, and... Figure 3B The process runs symbolically and ends at message enhancement 358. The dashed line of the first transfer arrow 356, while still within the area of ​​the graphical user interface text box 352, symbolizes that this first transfer arrow 356 is not displayed on the screen of the second user computer 102a2. Instead, this first transfer arrow 356 is for... Figure 3B It is shown for illustrative purposes. Figure 3B The box used for message enhancement 358 symbolically represents Figure 3A The steps shown in the diagram and described below for the message enhancement process 300, or alternatively represented, are as follows. Figure 3C The steps shown are for replacing message enhancement process 370.

[0070] In step 304 of the message enhancement process 300, the first message 354 is submitted to the storehouse of messages matched between Human-to-Automation Agent (H2AA) and Human-to-Human Agent (H2HA). This first message 354 may be the message received in step 302. This storehouse may be... Figure 1 The upgraded session store 125 shown herein may include upgrade messages collected via the upgrade session collection process 200. For example, the upgraded session store 125 may contain links to... Figure 2BThe combination of the initial message 254 shown and previously described, and the paraphrased message 256. Based on the number of times the upgraded session collection process 200 is repeated and the number of sessions fed into the upgraded session collection process 200, the upgraded session store 125 can contain hundreds, thousands, hundreds of thousands, or more matching messages. For Figure 1 The embodiment shown has both a message enhancement program 120 and an upgraded session store 125 configured in the second server 122. This submission of the first message 354 can occur via a transfer along the bus of the second server 122 from one memory storage area to another within the second server 122. This step 304 of the message enhancement process 300 can be... Figure 3B This is part of message enhancement 358 shown in the image.

[0071] Some embodiments may include an additional step between steps 304 and 306. This additional step may include an additional filtering step that analyzes the topic of the first message. After identifying the topic of the first message, the topic is compared with a list of topics that have been identified as challenging for detecting user intent. This list may be provided by the administrator of message enhancement program 120. In response to the identified topic being in the list of topics challenging for detecting user intent, subsequent steps of message enhancement process 300 may be performed. Thus, through this additional filtering, the administrator of automation agent / message enhancement program 120 may selectively select to apply message enhancements implemented by message enhancement program 120 when discussing topics that may pose particular challenges to intent detection by the automation agent. This additional filtering step may also be applied to topics that will be described later and Figure 3C The alternative message enhancement process 370 is described in the text. Similarly, gatekeeping steps can be applied to enable the message enhancement process 300 to be applied to automated agents operating in a specific country or human language, where the automated agent faces challenges in understanding incoming customer messages in that specific country or human language.

[0072] In step 306 of the message enhancement process 300, a determination is made regarding whether the first message has semantic similarity to any human-to-automation agent messages stored in the repository. If the determination in step 306 is affirmative, and the first message matches one or more human-to-automation agent messages stored in the repository, the message enhancement process 300 then proceeds to step 308. If the determination in step 306 is negative, and the first message does not match any human-to-automation agent messages stored in the repository, the message enhancement process 300 then proceeds to step 314.

[0073] The determination of step 306 can be performed by calculating the word vector of the first message 354 and the word vector of the Human-to-Automation Agent (H2AA) message portion stored in the repository, and comparing the vector of the first message 354 with the vector of the stored H2AA message portion. Therefore, through this embodiment, the message enhancement program 120 can include a language-based machine learning model capable of generating vectors of various phrases, sentences, and paragraphs and comparing the calculated vectors. For natural language processing, sentences can be represented as numeric vectors based on the semantic meaning of each word in the sentence and the relationship of each word to other words in the sentence. Two vectors that are close to each other can be semantically similar. For example, if a user writes (r_b) "How to handle maternity leave in the system?", the closest semantic sentence in the repository could be "How to handle vacation in the system?". Therefore, the mapping (r_h) could be semantically close to "How to react when my user returns from LOA?". When using a language-based model, such as BERT or a BERT-like system, to generate human-to-human requests (r_h), the result of the sentence could be "How do I reactivate my user account when I return from maternity leave?"

[0074] Step 306 can alternatively determine whether the first message matches any human-to-automation agent stored in the repository. This matching may include performing a text comparison of the first message 354 with portions of human-to-automation agent messages stored within the repository (e.g., within the upgraded session repository 125). Thus, in this embodiment, the message enhancement program 120 may include a comparator capable of comparing text and navigating in a lookup table using text comparison. A match may be an approximate match greater than a predetermined threshold. A match may be a root word match rather than a whole word match. One or more requests of the first message 354 may be compared with stored messages to find a match. When multiple repository entries have partial matches, the repository entry with the largest match percentage may be selected as long as it is greater than a predetermined threshold. Statistical analysis may be used to determine the optimal match.

[0075] Step 306 of the message enhancement process 300 can be... Figure 3B This is part of message enhancement 358 shown in the image.

[0076] In step 308 of the message enhancement process 300, human-to-human agent messages linked to matching human-to-human agent messages are retrieved. Matching human-to-automation agent (H2AA) messages may refer to the messages identified in step 306. For Figure 2BFor example, the start portion 258 is stored in an upgraded session store 125 in a manner that links to the upgraded portion 260, which is also stored in the upgraded session store 125. Therefore, if in step 306 a human-to-automation agent message (e.g., the initial message 254 (from...)... Figure 2B , and now in the repository)) with the first message 354 (in Figure 3B If the modified message 256 is linked to the initial message 254 in the upgraded session store 125, then the modified message 256 can be retrieved from the upgraded session store 125. The modified message 256 will be the human-to-human agent (H2HA) message retrieved in step 308. This step 308 of the message enhancement process 300 can be... Figure 3B This is part of message enhancement 358 shown in the diagram. In this embodiment, the upgraded session store 125 is used as a lookup table for semantically similar matches or word matches.

[0077] In step 310 of the message enhancement process 300, a human-to-human proxy message is added to the first message to form a combined message, such as an enhanced message. This human-to-human proxy message in step 310 may be the message retrieved in step 308. Figure 3B This demonstrates how the first message 354 is combined with the retrieved upgrade message, i.e. Figure 2B The combined message 362 is formed by combining the paraphrased message 256 shown in the diagram, but modified thereto. This modification includes extending the acronym "LOA" to be read as "vacation" to protect against situations where the automated agent does not understand the acronym. Therefore, in some embodiments, the combined or enhanced message may include a modified form of the retrieved message for combining with the first message 354. The upgraded session store 125 may include an additional table of acronym meanings and may replace each acronym with the corresponding word in the retrieved message before adding the retrieved message to the first message 354 to form the combined message 362. Figure 3B The second transition arrow 360 illustrates how message enhancement 358 leads to the generation of combined message 362. Message enhancement program 120 at the second server 122 can perform this combination.

[0078] For example, when an automation agent includes a language-based machine learning model that incorporates BERT or a BERT-type model, the output can be concatenated to the end of the first message to form a combined message with a delimiter such as "$SEPERATOR$" between the original message and the added message. This addition of a delimiter can be based on the requirements of the architecture used by the specific automation agent.

[0079] In step 312 of the message enhancement process 300, the combined message is submitted to the automation agent. For Figure 1 In the network environment 100 shown, this submission in step 312 can occur via a combined message from the second server 122 to the first server 112, and specifically, via the transmission of the dialog 110b to the first server 112. The dialog 110b embodies the... Figure 3B The example in the text is the automated agent for chatbot 366. Figure 3B In the middle, when the combined message 362 is submitted to the chatbot 366, the third transmission arrow 364 symbolizes this submission in step 312.

[0080] In step 314 of the message enhancement process 300, a first message is submitted to the automation agent. This first message can be the message received in step 302. Step 314 occurs when no match for the received message is found in the upgraded session store 125. In this case, message enhancement 358 does not occur, and the message can be submitted to the automation agent as is, for example, to the conversation program 110b at the first server 112. This branch of the message enhancement process 300 allows the automation agent to continue operating (albeit with lower precision), even though the message enhancement process is ineffective in enhancing the received message because the message enhancement program 120 has not found any way to enhance the message and is not convinced of enhancement.

[0081] In step 316 of the message enhancement process 300, intent prediction is generated via an automated agent. This automated agent may be the one to which the combined message is submitted in step 312, or the one to which the first message is submitted in step 314. The dialog program 110b at the first server 112 may invoke one or more internal algorithms, decision trees, and / or machine learning models to predict the intent of the received combined message. Intent prediction may be based on pre-programmed organization, information, and / or transaction services that may be provided by the organization implementing the automated agent. The automated agent itself may include a lookup table or machine learning model that allows providing a listed intent as output when a message or enhanced message is input. This generation of intent prediction is based on a pre-existing bot architecture. The message enhancement program 120 can operate without altering the pre-established bot architecture.

[0082] In step 318 of the message enhancement process 300, a response generated by an automated agent corresponding to the intent prediction is presented to the user in the session. This intent prediction may be the intent prediction generated in step 316. Figure 3BThe diagram shows an example of a chatbot response 369 generated by the automated agent in step 318. Data packets used to generate this chatbot response 369 can be sent via communication network 116 to the second user computer 102a2 so that the chatbot response 369 can be displayed on the screen of the second user computer 102a2 within a graphical user interface text box 352. This response can, for example, provide instructions to the user, provide links to the user, or provide requests for further information to the user. Figure 3B The chatbot response 369 shown provides instructions to the user at the second user computer 102a2 to reactivate their account after the user employee returns from leave. In this example, the instructions in chatbot response 369 include several steps for the user to take to perform the reactivation. Figure 3B The fourth transfer arrow 368, shown in the diagram, which begins at chatbot 366 and ends at chatbot response 369, symbolically represents this transmission of the generated message and its presentation to the user on the corresponding user computer. The dashed line of the fourth transfer arrow 368 after entering the graphical user interface text box 352 indicates that this fourth transfer arrow 368 is not displayed on the screen of the second user computer 102a2. Instead, this fourth transfer arrow 368 is... Figure 3B Provided for illustrative purposes.

[0083] After step 318, this iteration of the message enhancement process 300 can end. The message enhancement process 300 can be repeated for another human-to-automation session, and specifically, for another incoming first message from a human.

[0084] In some embodiments, via Figure 2A The upgraded session collection process 200 described herein collects a dataset of upgraded messages that can be used to train a machine learning model. Through this training, the machine learning model can then be used to predict human-to-human agent messages based on human-to-automation agent messages received and input into the machine learning model.

[0085] Machine learning models can be language-based training models and can be trained in a similar manner to training language translation machine learning models. For example, a machine learning model can be trained similarly to an English-to-Spanish translation machine learning model that can take English words, phrases, sentences, or paragraphs and output a Spanish translation of the input words, phrases, sentences, or paragraphs. Such models are trained via supervised learning by feeding them specific equivalent words, phrases, sentences, paragraphs, and / or documents (English and Spanish). After receiving an increasing amount of data, in many cases the model eventually becomes able to generate Spanish translations even without previously receiving an accurate set of English input.

[0086] Figure 2A The dataset collected can be similarly used for supervised training of the machine learning model. For each human-to-automation agent and human-to-human agent pair that successfully passes the filtering in the upgraded session collection process 200, the human-to-human agent version of the message for this embodiment can serve as a labeled dataset of what the result should be for the paired human-to-automation agent message. The machine learning model is trained by using these pairs as example input-output pairs. The machine learning model can internally use the mapping between human-to-automation agent requests and human-to-human agent requests. The machine learning model can be able to learn to recognize and understand acronyms so that the output of the machine learning model includes representative complete phrases instead of acronyms. Therefore, in Figure 3B In the example, the supplementary part of combined message 362 could include the phrase "vacation" instead of the acronym "LOA".

[0087] Figure 3C A flowchart illustrating the operation of the alternative message enhancement process 370 according to at least one embodiment is provided. The alternative message enhancement process 370 and... Figure 3A The message enhancement processing 300 shown is similar, but instead of a store / lookup table for semantic similarity or word matching, it uses a trained machine learning model described in several previous paragraphs. Therefore, for Figure 3C The alternative message enhancement process 370 shown, steps 372, 378, 380, and 382 are essentially equivalent to the following: Figure 3A Steps 302, 312, 316, and 318 of the message enhancement process 300 are shown. Specifically, for step 372 of the alternative message enhancement process 370, the first message is received from the user as part of a human-to-automation agent session. The features and definitions explained above for step 302 apply to step 372 of the alternative message enhancement process 370.

[0088] For step 374 of the alternative message enhancement process 370, a first message is submitted to a machine learning model to generate a human-to-human agent predicted message. This step may include submitting the first message received in step 372 to the machine learning model described above, i.e., to a language-based machine learning model trained using the dataset collected in the upgraded session collection process 200. The machine learning model may be configured in a second server 122 along with the message enhancement program 120. Therefore, the message submission in step 374 may include many of the same attributes and features as described for step 304 in the message enhancement process 300, except that the design of the receiver structure will be a machine learning model rather than a lookup table. The first message in step 374 may be from... Figure 3B The first message 354 of chatbot session 350 shown in the image.

[0089] For step 376 of the alternative message enhancement process 370, the output of the machine learning model is added to the first message to form a combined message. The output can come from the machine learning model fed to the first message in step 374. The first message can be the message received in step 372. The output of the machine learning model can be a predicted human-to-human agent version of the input message. The output can be concatenated to the end of the first message to form a combined message. Figure 3B The combined message 362 shown can be an example of the combined message generated in step 376. Except for providing new supplementary words / phrases / (one or more) sentences to add to the end of the original message, step 376 can be largely equivalent to... Figure 3A Step 310 of the message enhancement process 300 shown in the figure.

[0090] For example, when an automation agent includes a language-based machine learning model that incorporates BERT or a BERT-type model, the output can be concatenated to the end of the first message to form a combined message with a delimiter such as "$SEPERATOR$" between the original message and the added message. This addition of a delimiter can be based on the needs of the language model incorporated by the automation agent.

[0091] For step 378 of the alternative message enhancement process 370, the combined message is submitted to the automation agent. This combined message can be the message generated in step 376. Step 378 can be equivalent to... Figure 3A Step 312 of the message enhancement process 300 shown in the figure.

[0092] For step 380 of the alternative message enhancement process 370, the automation agent generates an intent prediction. This intent prediction is based on the input of the combined message generated in step 376 to the automation agent, as occurs in step 378. This step 380 can be equivalent to... Figure 3A Step 316 of the message enhancement process 300 shown in the figure.

[0093] For step 382 of the alternative message enhancement process 370, a response generated by an automated agent corresponding to the intent prediction is presented to the user in the session. The intent prediction can be the intent prediction generated in step 380. This step 382 can be equivalent to... Figure 3A Step 318 of the message enhancement process 300 shown in the figure.

[0094] for Figure 3A and 3C The message augmentation model can be represented as f(r_b) = (r_h), where the human-to-bot message request is the input to the function, and the output is the equivalent request written for the predicted human-to-human message. The output can be concatenated to the original message to form a composite message (r_b + r_h), which is sent to an automated agent (e.g., a chatbot) for intent detection. In this way, the automated agent can detect intent based on (r_b + r_h) instead of just r_b.

[0095] Some embodiments may include modifications to message enhancement process 300 and / or alternative message enhancement process 370, in which an automated agent prediction confidence level is determined. If the confidence level of the automated agent's intent prediction is below a predetermined threshold, message enhancement is then invoked first. The original message is fed to the automated agent, which generates an initial intent prediction / classification, and also generates a confidence level for the generated intent prediction / classification. In response to the confidence level being below the predetermined threshold, the remaining steps of message enhancement process 300 and / or alternative message enhancement process 370 may be performed to enhance the message. These additional steps ultimately include feeding the enhanced message to the automated agent, which can increase the confidence level that the automated agent has in its predictions. Figure 1 In the embodiment shown, a message can be sent to a dialog 110b at the first server 112 for input into an automation agent before being sent to the second server 122 for message enhancement via message enhancement program 120. After the composite / enhanced message is created, it can then be sent back to the first server 112 via communication network 116 for input into the automation agent. In some embodiments, message enhancement program 120 may also be configured within the first server 112 to eliminate these two (round-trip) external message transmissions via communication network 116.

[0096] Figure 3A , 3BThe diagram in Figure 3C illustrates how, in this embodiment, after the initial dataset is collected via the upgraded session collection process 200, a human agent is no longer required to perform the method. Both the message enhancement process 300 itself and the alternative message enhancement process 370 can be performed without human involvement or without any human involvement in the loop. This automation that eliminates the need for humans is helpful because human labeling can require significant resources for model creation and can therefore be expensive. Even the dataset in the upgraded session collection process 200 can be performed without additional human labeling, although a human agent is involved in previous sessions, such as the upgraded session 250.

[0097] Understandable. Figure 2A , 2B 3A, 3B, and 3C provide illustrations of some embodiments and do not imply any limitation on how the different embodiments are implemented. Numerous modifications may be made to the depicted embodiments(s), such as the depicted sequence of steps, based on design and implementation requirements.

[0098] Figure 4 This is an illustrative embodiment of the present invention. Figure 1 Block diagram 400 depicts the internal and external components of a computer. It should be understood that... Figure 4 This illustration provides only one possible implementation and does not imply any limitation on the environment in which different embodiments may be implemented. Many modifications can be made to the depicted environment based on design and implementation requirements.

[0099] Data processing systems 402a, 402b, 404a, and 404b represent any electronic device capable of implementing machine-readable program instructions. Data processing systems 402a, 402b, 404a, and 404b can represent smartphones, computer systems, PDAs, or other electronic devices. Examples of computing systems, environments, and / or configurations that can be represented by data processing systems 402a, 402b, 404a, and 404b include, but are not limited to, personal computer systems, server computer systems, thin clients, fat clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, network PCs, minicomputer systems, and distributed cloud computing environments that include any of the above systems or devices.

[0100] First and second user computers 102a1, 102a2, first and second servers 112, 122, and first human agent computer 132 may include Figure 4The diagram illustrates corresponding sets of internal components 402a, 402b and external components 404a, 404b. Each set of internal components 402a, 402b includes one or more processors 406, one or more computer-readable RAMs 408 and one or more computer-readable ROMs 410 on one or more buses 412, one or more operating systems 414, and one or more computer-readable tangible storage devices 416. One or more operating systems 414, a message enhancement program 120 in the second server 122 (or, in another embodiment, the first server 112), dialog programs 110a1, 110a2 in the first and second user computers 102a1, 102a2, dialog program 110b in the first server 112, and dialog program 110c in the human agent computer 132 may be stored on one or more computer-readable tangible storage devices 416 for execution by one or more processors 406 via one or more RAMs 408 (which typically include cache). Figure 4 In the illustrated embodiment, each computer-readable tangible storage device 416 is a disk storage device of an internal hard disk drive. Alternatively, each computer-readable tangible storage device 416 is a semiconductor storage device, such as ROM 410, EPROM, flash memory, or any other computer-readable tangible storage device capable of storing computer programs and digital information.

[0101] Each set of internal components 402a, 402b also includes an R / W drive or interface 418 for reading from and writing to one or more portable computer-readable tangible storage devices 420 (such as CD-ROM, DVD, Memory Stick, magnetic tape, disk, optical disc, or semiconductor storage devices). Software programs, such as message enhancement program 120, may be stored on one or more of the respective portable computer-readable tangible storage devices 420, read via the respective R / W drive or interface 418, and loaded into the respective hard disk drive, such as tangible storage device 416.

[0102] Each set of internal components 402a, 402b may also include a network adapter (or switch port card) or interface 422, such as a TCP / IP adapter card, a wireless Wi-Fi interface card, or a 3G, 4G, or 5G wireless interface card, or other wired or wireless communication links. The message enhancement program 120 in the second server 122 (or, in another embodiment, the first server 112), the dialog programs 110a1, 110a2 in the first and second user computers 102a1, 102a2, the dialog program 110b in the first server 112, and the dialog program 110c in the human agent computer 132 can be accessed via a network (e.g., the Internet, a local area network, or other wide area networks). Figure 1The communication network 116 shown in the diagram and the corresponding network adapter or interface 422 are downloaded from an external computer (e.g., a server). From the network adapter (or switch port adapter) or interface 422, the message enhancement program 120 in the second server 122 (or, in another embodiment, the first server 112), the dialog programs 110a1 and 110a2 in the first and second user computers 102a1 and 102a2, the dialog program 110b in the first server 112, and the dialog program 110c in the human agent computer 132 are loaded into the corresponding hard disk drives, such as physical storage devices 416. The network may include copper wire, fiber optic, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers.

[0103] Each set of external components 404a, 404b may include a computer display monitor 424, a keyboard 426, and a computer mouse 428. External components 404a, 404b may also include a touchscreen, virtual keyboard, touchpad, pointing device, and other human-machine interface devices. Each set of internal components 402a, 402b also includes a device driver 430 for interfacing with the computer display monitor 424, keyboard 426, and computer mouse 428. Device driver 430, R / W driver or interface 418, and network adapter or interface 422 include hardware and software (stored in storage device 416 and / or ROM 410).

[0104] This invention can be a system, method, and / or computer program product at any possible level of technical detail integration. The computer program product may include a computer-readable storage medium (or media) having computer-readable program instructions thereon for causing a processor to perform aspects of the invention.

[0105] Computer-readable storage media can be tangible devices capable of retaining and storing instructions used by an instruction execution device. Computer-readable storage media can be, for example, but not limited to, electronic storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of computer-readable storage media includes the following: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable optical disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices such as punch cards or raised structures in recesses on which instructions are recorded, and any suitable combination of the foregoing. As used herein, computer-readable storage media should not be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0106] The computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to a suitable computing / processing device, or via a network, such as the Internet, a local area network (LAN), a wide area network (WAN), and / or a wireless network, to an external computer or external storage device. The network may include copper cables, optical fibers, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to a computer-readable storage medium within the respective computing / processing device.

[0107] Computer-readable program instructions used to perform the operations of this invention may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, integrated circuit configuration data, or source code or object code written in any combination of one or more programming languages ​​(including object-oriented programming languages ​​such as Smalltalk, C++, etc.) and procedural programming languages ​​(e.g., the "C" programming language or similar programming languages). The computer-readable program instructions may be implemented entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the latter scenario, the remote computer may be connected to the user's computer via any type of network (including a local area network (LAN) or a wide area network (WAN)) or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, to perform aspects of this invention, electronic circuits including, for example, programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs) may be implemented using the state information of the computer-readable program instructions to personalize the electronic circuits.

[0108] This document describes aspects of the invention with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0109] These computer-readable program instructions may be provided to a processor of a computer or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / actions specified in one or more boxes of a flowchart and / or block diagram. These computer-readable program instructions may also be stored in a computer-readable storage medium that can instruct a computer, programmable data processing apparatus, and / or other devices to operate in a particular manner, such that the computer-readable storage medium storing the instructions comprises an article of writing containing instructions that implement aspects of the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0110] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer-implemented process, such that the instructions, which execute on the computer, other programmable apparatus or other device, perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0111] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of instructions, including one or more executable instructions for implementing a specified logical function(s). In some alternative embodiments, the functions indicated in the blocks may occur in a non-linear order as shown in the drawings. For example, two blocks shown consecutively may actually be implemented as a single step, simultaneously, substantially simultaneously, in a manner that overlaps partially or entirely in time, or, depending on the functions involved, these blocks may sometimes be executed in reverse order. It will also be noted that each block illustrated in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified function or action or executes a combination of dedicated hardware and computer instructions.

[0112] It should be understood that while this disclosure includes a detailed description of cloud computing, the implementation of the teachings detailed herein is not limited to cloud computing environments. Rather, embodiments of the invention can be implemented in conjunction with any other type of computing environment now known or developed hereafter.

[0113] Cloud computing is a service delivery model that enables convenient, on-demand network access to shared pools of configurable computing resources (e.g., networks, network bandwidth, servers, processing, memory, storage devices, applications, virtual machines, and services) that can be rapidly provisioned and released with minimal management effort or interaction with service providers. This cloud model may include at least five features, at least three service models, and at least four deployment models.

[0114] The features are as follows:

[0115] On-demand self-service: Cloud consumers can unilaterally and automatically provide computing power, such as server time and network storage, as needed, without requiring human interaction with the service provider.

[0116] Extensive network access: Capabilities are available on the network and accessed through standard mechanisms that facilitate use by heterogeneous thin or thick client platforms (e.g., mobile phones, laptops, and PDAs).

[0117] Resource pooling: A provider's computing resources are pooled to serve multiple consumers using a multi-tenant model, where different physical and virtual resources are dynamically allocated and reallocated based on demand. This has significance in terms of location independence, because consumers typically do not control or know the exact location of the resources provided, but can specify the location at a higher level of abstraction (e.g., country, state, or data center).

[0118] Rapid Flexibility: In some cases, the ability to scale outwards and inwards quickly and flexibly can be provided. For consumers, the available capacity often appears unlimited and can be purchased in any quantity at any time.

[0119] Measurement services: Cloud systems automatically control and optimize resource usage by leveraging metering capabilities at certain levels of abstraction appropriate to service types (e.g., storage, processing, bandwidth, and active user accounts). Resource usage can be monitored, controlled, and reported, providing transparency for both service providers and consumers.

[0120] The service model is as follows:

[0121] Software as a Service (SaaS): This provides consumers with the ability to use a provider's applications running on cloud infrastructure. These applications can be accessed from various client devices through a thin client interface such as a web browser (e.g., web-based email). Consumers do not manage or control the underlying cloud infrastructure, including the network, servers, operating system, storage devices, or even individual application capabilities, with the possible exception of limited user-specific application configuration settings.

[0122] Platform as a Service (PaaS): This provides consumers with the ability to deploy applications created by the consumer or acquired using provider-supported programming languages ​​and tools onto cloud infrastructure. Consumers do not manage or control the underlying cloud infrastructure, including networks, servers, operating systems, or storage devices, but they have control over the deployed applications and the configuration of any application hosting environment.

[0123] Infrastructure as a Service (IaaS): This provides consumers with the capability to offer processing, storage, networking, and other basic computing resources that enable them to deploy and run any software (which may include operating systems and applications). Consumers do not manage or control the underlying cloud infrastructure, but they do have control over the operating system, storage devices, deployed applications, and limited control over networking components of their choice (e.g., host firewalls).

[0124] The deployment model is as follows:

[0125] Private cloud: Cloud infrastructure used solely for organizational operations. It can be managed by the organization or a third party and can exist inside or outside a building.

[0126] Community cloud: Cloud infrastructure shared by several organizations and supporting specific communities with shared concerns (e.g., tasks, security requirements, policies, and compliance considerations). It can be managed by an organization or a third party and can exist on-site or off-site.

[0127] Public cloud: Cloud infrastructure that is available to the general public or large industrial groups and is owned by an organization that sells cloud services.

[0128] Hybrid cloud: A cloud infrastructure is a combination of two or more clouds (private, community, or public) that remain a single entity but are bound together by standardized or proprietary technologies that enable data and application portability (e.g., cloud bursts for load balancing between clouds).

[0129] Cloud computing environments are service-oriented, focusing on statelessness, loose coupling, modularity, and semantic interoperability. At the heart of cloud computing is the infrastructure of a network of interconnected nodes.

[0130] Now for reference Figure 5 The diagram illustrates an illustrative cloud computing environment 500. As shown, the cloud computing environment 500 includes one or more cloud computing nodes 50 that can communicate with local computing devices used by cloud consumers, such as personal digital assistants (PDAs) or cellular phones 50A, desktop computers 50B, laptop computers 50C, and / or automotive computer systems 50N. The nodes 50 can communicate with each other and may include… Figure 1 The individual computers and servers shown are illustrated. They can be physically or virtually grouped (not shown) in one or more networks, such as the private, community, public, or hybrid clouds described above, or a combination thereof. This allows the cloud computing environment 500 to provide infrastructure, platform, and / or software services to cloud consumers who do not need to maintain resources on their local computing devices. It should be understood that in Figure 5 The types of computing devices 50A-50N shown are for illustrative purposes only, and computing node 50 and cloud computing environment 500 can communicate with any type of computing device via any type of network and / or network-addressable connection (e.g., using a web browser).

[0131] Now for reference Figure 6 This illustrates a set of functional abstraction layers 600 provided by the cloud computing environment 500. It should be understood beforehand that... Figure 6 The components, layers, and functions shown are for illustrative purposes only, and embodiments of the invention are not limited thereto. As depicted, the following layers and corresponding functions are provided:

[0132] Hardware and software layer 602 includes hardware and software components. Examples of hardware components include: host 604; server 606 based on RISC (Reduced Instruction Set Computer) architecture; server 608; blade server 610; storage device 612; and network and networking components 614. In some embodiments, software components include network application server software 616 and database software 618.

[0133] The virtualization layer 620 provides an abstraction layer from which the following examples of virtual entities can be provided: virtual server 622; virtual storage device 624; virtual network 626, including virtual private network; virtual application and operating system 628; and virtual client 630.

[0134] In one example, management layer 632 may provide the functionality described below. Resource provisioning 634 provides dynamic procurement of computing resources and other resources used to perform tasks within the cloud computing environment. Metering and pricing 636 provides cost tracking, such as when resources are used within the cloud computing environment, and billing or pricing for consuming these resources. In one example, these resources may include application software licenses. Security provides authentication for cloud consumers and tasks, and protection for data and other resources. User portal 638 provides access to the cloud computing environment for consumers and system administrators. Service level management 640 provides cloud resource allocation and management to ensure that required service levels are met. Service level agreement (SLA) planning and fulfillment 642 provides pre-scheduling and procurement of cloud resources, where future needs are anticipated according to the SLA.

[0135] Workload layer 644 provides examples of functionalities that can leverage a cloud computing environment. Examples of workloads and functionalities that can be provided from this layer include: mapping and navigation 646; software development and lifecycle management 648; virtual classroom education delivery 650; data analytics and processing 652; transaction processing 654; and enhanced conversation collection and message enhancement 656. Message enhancement program 120 provides a way to automatically enhance messages being sent to an automation agent using collected enhanced conversations, helping the automation agent better understand appropriate responses to queries expressed in the messages.

[0136] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used herein, the singular forms “a,” “an,” and “the” are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will also be understood that the terms “comprising,” “including,” “containing,” “having,” “having,” “with,” “using,” etc., as used in this specification, specify the presence of the stated features, integers, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0137] Various embodiments of the invention have been described for illustrative purposes, but are not intended to be exhaustive or limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope of the described embodiments. The terminology used herein is chosen to best explain the principles of the embodiments, their practical application, or improvements to technical techniques found in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A computer-implemented method, comprising: The first message is received from the first user, and the first message is generated during the first session between the first user and the first automation agent. A second message is generated by the computer. The second message includes the same request as the first message, but has a different language form than the first message. The second message and the first message are combined to form a combined message, wherein the combination occurs by concatenating the second message to the first message; as well as The combined message is input into the first automation agent, causing the first automation agent to generate an intent classification for the first message.

2. The method of claim 1, wherein the second message is generated in a previous session between another user and a human agent.

3. The method of claim 2, wherein the previous session includes an upgraded session that begins with the other user and the automated agent and upgrades to the human agent.

4. The method of claim 3, further comprising: Extract the first request from the beginning of the upgraded session, wherein the beginning occurs between the other user and the automated agent; Extract the second request from the upgrade portion of the upgraded session, wherein the upgrade portion occurs between the other user and the human agent; The first request is compared with the second request to determine the similarity between the first request and the second request; as well as In response to the similarity being greater than a predetermined threshold, the previous session is stored in an upgraded session store, wherein the previous session includes a second message.

5. The method of claim 4, wherein the second message is stored in the upgraded session store to link to the start portion of the upgraded session.

6. The method of claim 5, wherein: The generation of the second message includes performing a text comparison between the first message and the message stored in the upgraded session store; The text comparison identifies a text match between the first message and the beginning portion of the upgraded session; as well as The upgraded portion is retrieved from the upgraded session repository and generated as a second message because it is linked to the beginning portion.

7. The method of claim 1, wherein generating the second message comprises inputting the first message into a machine learning model and, in response to the input, receiving the second message as an output from the machine learning model.

8. The method of claim 1, wherein the generation of the second message includes performing a text comparison between the first message and a message stored in an upgraded session store.

9. The method of claim 1, wherein the generation of the second message includes performing a semantic similarity comparison between the first message and a message stored in an upgraded session store.

10. The method of claim 1, wherein the first automated agent comprises a natural language processor that performs natural language processing on the combined message.

11. The method of claim 1, further comprising filtering out irrelevant content from the first message to find the first utterance, wherein the request is based on the first utterance.

12. The method of claim 1, wherein the first automated agent is a chatbot.

13. The method of claim 1, further comprising: Identify the subject of the first message; as well as The identified topics are compared with a list of topics that are challenging to detect user intent; In response to the identified topic being in the list of topics that pose a challenge to detecting user intent, the generation of the second message is performed.

14. A computer system, comprising: One or more processors, one or more computer-readable tangible storage media, and program instructions stored on at least one of the one or more computer-readable tangible storage media, the program instructions being executed by at least one of the one or more processors to cause the computer system to: A first message is received from a first user, and the first message is generated during a first session between the first user and the first automation agent. A second message is generated, which has the same request as the first message, but has a different language form. The second message and the first message are combined to form a combined message, wherein the combination occurs by concatenating the second message to the first message; as well as The combined message is input into the first automation agent, causing the first automation agent to generate an intent classification for the first message.

15. The computer system of claim 14, wherein the second message is generated in a previous session between another user and a human agent.

16. The computer system of claim 15, wherein the previous session includes an upgraded session that begins with the other user and the automated agent and upgrades to the human agent.

17. The computer system of claim 14, wherein generating the second message includes performing a semantic similarity comparison between the first message and a message stored in an upgraded session store.

18. The computer system of claim 14, wherein the first automation agent includes a natural language processor that performs natural language processing on the combined message.

19. A computer program product comprising a computer-readable storage medium having program instructions implemented therewith, wherein the program instructions are executable by a computer system to cause the computer system to: The first message is received from the first user, and the first message is generated during the first session between the first user and the first automation agent. A second message is generated, which has the same request as the first message, but has a different language form. The second message and the first message are combined to form a combined message, wherein the combination occurs by concatenating the second message to the first message; as well as The combined message is input into the first automation agent, causing the first automation agent to generate an intent classification for the first message.

20. The computer program product of claim 19, wherein generating the second message includes performing a semantic similarity comparison between the first message and a message stored in an upgraded session library.

21. A system comprising modules for performing the steps of the method of any one of claims 1-13.

Images