EP4364064A1

EP4364064A1 - Chat bot for an instant messaging application

Info

Publication number: EP4364064A1
Application number: EP22744801.6A
Authority: EP
Inventors: Thierry Martinez; Christian Gregoire; Régis ESNAULT
Original assignee: Orange SA
Current assignee: Orange SA
Priority date: 2021-06-28
Filing date: 2022-06-27
Publication date: 2024-05-08
Also published as: WO2023275472A1; FR3124622A1

Abstract

The present description proposes a method for processing data of an instant messaging application in which users taking part in a discussion exchange messages via respective terminals (TER), the method, implemented by a computerized chat bot (DIS), comprising: - analysing the exchanged messages in order to identify a current discussion context, - determining a relevance of the chat bot intervening in the discussion, based on the current context and based on predetermined rules, and - based on said determination, transmitting at least one message appropriate to the current context to at least one of the participants.

Description

Title: Conversational agent in an instant messaging application Technical area

This description relates to the field of instant messaging (or "chat").

Prior technique

[0002] The term "chatroom" (or "discussion room") designates a virtual meeting place, accessible from a site, which an Internet user can choose, depending on the subject proposed or the interest of the moment, in order to dialogue via a keyboard-type interface (or by voice recognition), with a certain number of participants.

[0003] The term “chatbot” denotes an (IT) agent configured to dialogue with a user. The user is offered an interface through which the user can query the chabot on a research subject of his choice. The search that the user wishes to launch on this "person-machine" interface is then influenced by the so-called "Turing" test according to which the dialogue with the machine must give the illusion that a program thinks according to a sensible dialogue, making so natural the interaction with the machine.

[0004] The chatbot is therefore intended more for a dialogue between an IT agent and a user, whereas the chatroom is usually a dialogue between several human users via terminals available to these users (terminals of the smartphone, tablet, computer or others).

[0005] There are also "metabots" which are capable of orchestrating several chatbots at the same time and of redirecting a specific request from a user according to the intention of this user towards a given chatbot or another chatbot.

[0006] Chatbots thus offer real resources to individual users for learning information.

[0007] A limitation of existing chatbots, however, is that they only offer to enrich the knowledge of an individual user, according to their own research.

Summary

[0008] This description improves the situation.

[0009] To this end, it proposes the implementation of a computer agent, in particular of the chatbot type, configured to intervene within the framework of a virtual discussion room, in particular of the chatroom type.

[0010] The present description aims for this purpose at a method of processing instant messaging data in which users participating in a discussion exchange messages via respective terminals, the method, implemented by a conversational computer agent, comprising :

- analyze the messages exchanged to identify a current discussion context,

- determine the relevance of the conversational agent's intervention in the discussion, depending on of the current context and according to predetermined rules, and

- depending on said determination, transmit, to at least one of the participants, at least one message appropriate to the current context and to at least one of the participants.

[0011] Thus, the present description proposes a conversational agent or "chatbot" capable of addressing several users of instant messaging and participating in the same discussion (or "chatroom") and thus of integrating into this chatroom . It is thus possible to orchestrate the resolution of a problem with the assistance of several people and one or more sculpin operators, for example in a work meeting, or even to animate and/or enrich an exchange between several users, each user thus learning interesting information for the group of participants.

[0012] The aforementioned predetermined rules for the intervention of the chatbot must be adapted specifically to an application to a chatroom. Indeed, a problem with existing chatbots is inherent in the fact that they are only programmed to interact with a single interlocutor. Therefore, the implementation of such a chatbot within a chatroom can first of all pose the problem of the untimely intervention of the chatbot in a discussion, for example.

[0013] In particular, the set of predetermined rules includes predefined rules, modifiable rules, or even new rules. Thus, in one embodiment, the aforementioned predefined rules may comprise at least the detection of a wake-up word from the conversational agent in a last message of the discussion.

[0014] In addition or as a variant, the predefined rules can comprise at least one detection of a duration of stopping the exchange of messages in the discussion, greater than a (predefined) threshold.

[0015] Thus, the conversational agent can restart the discussion if it runs out, or at least revitalize it if the exchanges of messages between users begin to be spaced out by a duration corresponding to the aforementioned threshold.

[0016] In addition or as a variant, the predetermined rules may comprise at least one detection of a mood of at least one of the participants and a taking into account of said mood to transmit or not a response message to a message from this participant.

[0017] Indeed, it is necessary to overcome the problem of a response from the chatbot that is not appropriate to the discussion context. For example, the chatbot intended for a single interlocutor can react as if all the interlocutors constitute only one person. Thus, the chatbot may have a reaction that would be appropriate for one of the interlocutors, but inappropriate for the others (manifestation of a lack of empathy, failure to take into account the knowledge or language of one of the interlocutors , etc.). The taking into account of the mood of the participants by the conversational agent, in a distinct way between the participants, thus makes it possible to overcome this problem.

In such an embodiment, in the event of transmission of the response message by the conversational agent, the response message transmitted is then adapted to said mood of said at least one of the participants. [0019] In an embodiment where the messages exchanged are vocal (in a context of voice messaging for example), the detection of the mood of said at least one of the participants is enriched by a detection of vocal components in the voice message of said at least one of the attendees.

[0020] For example, an increase in the tonal frequency and/or the sound level in the speech signal of a participant characterizes an irritation of this participant.

[0021] In an embodiment implementing participant mood detection, it is also possible to:

- analyze each message of a participant before publication for the other participants in the discussion, to detect a mood of the said participant according to the message to be published, and

- in the event of detection of a mood reflecting a predefined emotion (typically a negative emotion such as anger for example) with a score above a threshold, correcting the message before publication to reduce said score, and publishing the corrected message to participants in the discussion.

[0022] Such an achievement makes it possible, for example, to clean up relations between work colleagues in the case in particular of an application of the method to the implementation of a conversational agent in a collaborative professional context, as described in detail below.

[0023] In a further embodiment, the method may comprise, to develop a message to be transmitted:

- access a knowledge database,

- identify a topic in the database linked to at least one last message in the discussion, and

- form an informative message, to be transmitted, and comprising data from the knowledge base.

Thus, the conversational agent here contributes to providing informative content to all the participants and can thus enrich their general knowledge or their professional knowledge.

[0025] For example, in the case of a discussion planned in a professional, collaborative work environment, the aforementioned knowledge database may include information specific to a professional sector of this collaborative work environment.

[0026] The messages transmitted by the conversational agent can then comprise professional tasks assigned to at least some of the participants in the discussion, the method then comprising in this embodiment a consultation of the knowledge base to organize a sequential coordination of the tasks assigned professionals, based on an identification of the participants in the discussion.

In this application, the conversational agent not only participates in the exchange of messages with the participants in the discussion, but also offers an assignment of tasks with possible organizational assistance provided to each of the participants. According to another aspect, there is provided a computer program comprising instructions for implementing the method as defined herein when this program is executed by a processor. In another aspect, there is provided a non-transitory, computer-readable recording medium on which such a program is recorded. [0029] According to another aspect, a computer device is proposed (reference DIS in FIG. 1 commented on in detail below) comprising a communication interface with an instant messaging server (SER) and a processing circuit configured to access to a memory storing predetermined rules (RP) of a conversational agent for the implementation of a method according to the present. According to another aspect, there is proposed a conversational agent server (DIS) processing communication data between a plurality of terminals (TER) running an instant messaging application in which users participating in a discussion exchange messages via respective terminals (TER), the conversational agent server comprising:

- a message generator able to analyze the messages exchanged to identify a current discussion context, to determine the relevance of intervention of the conversational agent in the discussion according to the current context and according to predetermined rules, to provide a message appropriate to the current context and to at least one of the participants, and

- a communication interface with an instant messaging server (SER) capable of transmitting to at least one of the participants the at least one appropriate message.

Brief description of the drawings

[0031] Other characteristics, details and advantages will appear on reading the detailed description below, and on analyzing the appended drawings, in which:

Fig. 1 [0032] [Fig. 1] shows an example of a system with a chatroom-type messaging server, user terminals, and a computing device according to one embodiment.

Fig. 2

[0033] [Fig. 2] shows an example of message exchange involving a conversational agent within the meaning of this description. Fig. 3

[0034] [Fig. 3] shows illustrates by way of example the steps of a method according to one embodiment.

Description of embodiments

In the description below, the intervention of a virtual conversational agent (or chatbot) is proposed, specifically here in a collective discussion, thus allowing several users to interact with a digital environment. This virtual agent relies on a database of knowledge to propose appropriate responses to a context detected by application of predetermined rules. Depending on the contexts detected, it may be proposed to guide operations involving several users in a work context, with a sequencing of the tasks to be carried out, or even, as in the simple examples of implementation presented for didactic purposes below, to share universal knowledge with several users by intervening in a conversation in an appropriate way.

The conversational agent is able to distinguish between the different participants and to adapt the message to each one, or to the whole group at the same time, or even to certain people specifically. Thus, the predetermined rules make it possible to manage the interaction in the sense of individualized user experience in a collective discussion. The individualization of the conversational agent's messages is notably based on individualization criteria, such as accreditation, age (child/adult), competence in relation to the technicality of the subject of discussion, emotional state , a cultural reference... of the participant(s) recipient(s) of the conversational agent's message. [0037] Reference is now made to FIG. 1 in which each user has a TER terminal running a “Chat Client” computer application making it possible to send and receive messages spoken and/or written in natural language via a network extended RES (Internet type) and intended for an instant messaging server SER (designated by the term “Chat Room” in FIG. 1) which then routes the messages in natural language, coming from a “Chat Client” of a user to the "Chat Clients" of other users. The messages can indeed be spoken by the users and identified as to their content by voice recognition and/or be written directly by the users.

[0038] In addition, these statements are also routed (here by the chatroom server SER) to a device, also called conversational agent server, DIS within the meaning of the present disclosure which can take, in the example described, the form of a second server designated by the term "Chat Room Servant". The device DIS implements a chatbot application and can thus access a storage memory of a database DB listing the "knowledge" on which the chatbot can rely to provide informational messages appropriate to a conversation context between Chat Client users. The device DIS can also access a memory storing predetermined rules RP (for example in the form of computer instruction codes) defining the contexts of possible interventions of the chatbot (when to intervene during a conversation), as well as the appropriate ways to intervene for the chatbot (how to intervene during the conversation, depending for example on the mood of each participant). Thus, the Chat Room Servant can, to respond to a request for information for example, return a statement from its RP knowledge base. To this end, it implements:

- an AM application for analyzing messages in natural language (or "Natural Language Understanding"), allowing the chatbot to interpret the messages exchanged between the different participants during the discussion, - a GM application for generating messages in natural language (or "Natural Language Generator"), allowing the chatbot to intervene in the conversation,

- an MR rules engine application (or "Rule Engine"), allowing the chatbot to rely on the aforementioned predetermined rules to detect a conversation context and form an appropriate response to this context.

In particular, the natural language message generator GM may comprise the message analyzer AM and/or the rules engine MR making it possible to generate messages according to predetermined rules.

In a particular embodiment, the conversational agent server (DIS) processes communication data between a plurality of terminals (TER) running an instant messaging application in which users participating in a discussion exchange messages via terminals respective (TER), the conversational agent server comprising:

The rules of the rules engine can be modified by a user such as an authorized administrator, or even by learning with respect to typical discussion situations. Thus, the "Chat Room Servant" application can, by taking into account a specific set of rules, adopt a particular global behavior such as "dominant", "expressive", "analytical", "sympathetic", or others .

In addition, new rules can be determined prior to the transmission of a message by the second conversation server-agent DIS which enrich the rules engine, including during an ongoing discussion using a conversational agent.

[0044] It can also be implemented a targeted communication to a single or only part of a group of participants in a discussion. Thus, the “Chat Room Servant” may not provide the same information to all participants. Provision may also be made for a notion of roles such as the distribution of the communication “for information”, or “for action”, depending on the profiles.

[0045] As an example, a table of possible exchange of messages in natural language in which the chatbot intervenes is presented below:

[0046] [Table 1]

In particular, the second conversational server-agent DIS is able to intervene in the discussion based on predetermined rules and/or using a dynamic object-oriented language, such as “Smalltalk”. The second conversational server-agent DIS is thus able to identify the times at which it must intervene based in particular on criteria linked to one or more participants, in particular individualization criteria, including for example emotional criteria. These individualization criteria are not necessarily predefined but may depend on the context, the individuals, etc. It can be noted from the message M7 and its response M8 that the intervention of the chatbot in the exchanges can be triggered for example by a given naming of the chatbot, such as “Butler”. As illustrated in Figure 2, the chatbot can be given a name, including in exchanges with "bubbles" mentioning the names of the participants in the conversation (here Thierry for participant A and Régis for participant B, and so Butler for the chatbot).

[0049] More generally, if the participants do not want the chatbot to speak untimely, it is possible to define an exchange "protocol" in which all messages intended for the chatbot begin with a keyword (for example his name “Butler” or a wake-up word or “wake-up word”). [0050] With reference now to FIG. 3, after having analyzed the messages of human users in step S1 (via natural language analysis processing techniques using artificial intelligence for example) to detect certain About (not entirely accurate, violent, or with incitement to hatred, or the like), the chatbot can "correct" messages as a moderator, in step S2, before finally releasing each message to the set of participants in step S3. Thus the messages can be analyzed by the chatbot before being published to the participants, in particular in a work environment. The chatbot can, for example, evaluate an emotion score in step S4 that each message can elicit and, if the score exceeds a predefined threshold, correct such a message to make it more impersonal, before publishing it at step S3. [0051] Furthermore, for the intervention of the chatbot in the exchanges, the chatbot can detect, for example, a wake-up word (step S5) or even parameters indicative of an end of statement of one or more participants (step S6), or the like, to initiate a response message in step S9. For example at step S6, the chatbot analyzer can identify the end of a discussion paragraph of an interlocutor, or even the end of an exchange between two people or the end of a conversation, allowing it to Intervene at a convenient time without interrupting an ongoing discussion.

[0052] Optionally, the chatbot can distinguish the different interlocutors and their profiles to take them into account in its intervention in step S7. The chatbot can limit its intervention typically depending on the mood of the interlocutor it is addressing, for example. The different participants are distinguished in particular according to the expression (emotion in the message) of each participant (following step S4). In the event, for example, of the repeated use of certain words, or of specific expressions aimed at interrupting sentences, on the part of a given participant, the chatbot may intervene less in response to the words of this user. Thus, the chatbot can distinguish the different participants (in relation to their email account for example, or other) and limit or on the contrary increase the number of its interventions with such or such participant according to the parameters detected in the messages and reflecting the respective moods of these participants.

[0053] In the case of a voice chatroom application (the messages exchanged being audio messages), the chatbot, in addition to receiving the messages spoken by the people in the chat room (room), can measure the times of silence , especially following a question. From then on, the Chat Room Servant can, on a silence after a question, trigger a previously programmed reaction to restart the conversation, or joke, or other. For example, in a communication similar to a conference call, if a moderator says, "Does anyone want to say anything to wrap up?" “, after ten seconds of silence, the Chat Room Servant can take the floor to communicate “Do you want me to sing you a song? ".

[0054] The Chat Room Servant also has the ability, in this embodiment of a voice chat room, to enrich the detection of the mood of one or more participants by measuring in particular a degree of "nervousness" of such participants, depending for example on the sound volume and/or the voice frequency of the voice utterances (a higher volume and/or a higher voice frequency typically characterizing annoyance), these voice parameters being detected in step S8 (illustrated by dotted lines in the voice chatroom embodiment). Depending on the mood of the participants, the Chat Room Servant can then react by implementing a previously programmed behavior (broadcasting of soft music, statement recalling rules of decorum, diplomatic words, etc.) in step S9. For example, mood detection is performed from acoustic signals and/or other signals (image, physiological, semantic, etc.)

[0055] In terms of possible applications, in a work context for example, the Chat Room Servant has the ability to orchestrate interventions from different people based on information from its comic book knowledge base, to organise, at the step S10, a distribution of roles and instructions to the participants on the chatroom according to their profiles and to collect their feedback in order to follow the progress of collaborative tasks.

[0056] The advantage of the second conversation server-agent according to the invention is its adaptability to users, to usage and to context (time, location, environment, etc.) [0057] More generally, the Chat Room Servant can be used in a domestic setting

(with voice assistants) or in the company (by teleconference with the Chat Room Servant as a chatbot mediator), or in the factory. In an industrial context, it is thus possible to control an intervention, to store written information precisely and automatically, for example by implementing a programmed memory to generate and store a fault tree and possibly enrich a resolution tree, by learning. The Chat Room Servant can be initiated by an application inviting people concerned (executive and worker) by an event or alternatively at the specific request of a person (a worker facing a problem) by inviting the Chat Room Servant and other people (other workers and executives) to participate in a spontaneous chatroom, in order to understand an event and make thoughtful decisions collectively. In another example of application, for example outside a strict working framework, a motorist who has broken down can initiate a chat room with the Chat Room Servant, a mechanic and himself in order to determine whether he is able to repair his vehicle alone or arrange for the intervention of a tow truck.

Claims

[Claim 1] A method of processing communication data between a plurality of terminals (TER) running an instant messaging application in which users participating in a discussion exchange messages via respective terminals (TER), the method implemented by a conversational computer agent (DIS), comprising:

- analyze the messages exchanged to identify a current discussion context,

- determine the relevance of the conversational agent's intervention in the discussion, according to the current context and according to predetermined rules, and

[Claim 2] Method according to claim 1, in which the predetermined rules comprise at least the detection of a wake-up word of the conversational agent in a last message of the discussion.

[Claim 3] Method according to one of the preceding claims, in which the predetermined rules comprise at least one detection of a duration of cessation of exchange of messages in the discussion, greater than a threshold.

[Claim 4] Method according to one of the preceding claims, in which the predetermined rules comprise at least one detection of a mood of at least one of the participants and a taking into account of said mood to transmit or not a response message to a message from said at least one of the participants.

[Claim 5] A method according to claim 4, wherein, in case of transmission of the response message, the transmitted response message is adapted to said mood of said at least one of the participants.

[Claim 6] Method according to one of Claims 4 and 5, in which the messages exchanged are vocal, and the detection of the mood of said at least one of the participants is enriched by a detection of vocal components in the vocal message of said at least least one of the participants.

[Claim 7] Method according to one of Claims 4 and 5, further comprising:

- analyze each message of a participant before publication for the other participants in the discussion, to detect a mood of the said participant according to the message to be published,

- in the event of detection of a mood reflecting a predefined emotion with a score above a threshold, correct the message before publication to reduce said score, and publish the corrected message to the participants in the discussion.

[Claim 8] Method according to one of the preceding claims, comprising, to produce a message to be transmitted:

- access a knowledge database,

[Claim 9] The method of claim 8, wherein said discussion is provided within a professional, collaborative work setting, and the knowledge database includes information specific to a professional sector of said collaborative work setting.

[Claim 10] Method according to claim 9, in which the messages transmitted by the conversational agent comprise professional tasks assigned to at least part of the participants in the discussion, the method comprising a consultation of the knowledge base to organize a coordination sequence of assigned professional tasks, based on an identification of the participants in the discussion.

[Claim 11] Computer program comprising instructions for implementing the method according to one of Claims 1 to 10 when this program is executed by a processor.

[Claim 12] Non-transitory recording medium readable by a computer on which is recorded a program for implementing the method according to one of Claims 1 to 10 when this program is executed by a processor.

[Claim 13] Computer device (DIS) comprising a communication interface with an instant messaging server (SER) and a processing circuit configured to access a memory storing predetermined rules of a conversational agent for the implementation of a method according to one of claims 1 to 10.

[Claim 14] Conversational agent server (DIS) processing communication data between a plurality of terminals (TER) running an instant messaging application in which users participating in a discussion exchange messages via respective terminals (TER), the server conversational agent comprising: