Text conversation support device, method, and program
The text conversation support device addresses the inefficiencies of text chat by classifying conversation states and generating supportive messages to encourage user participation, thereby enhancing the naturalness and efficiency of text conversations.
Patent Information
- Application Number
- PCT/JP2023/019138
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-23
- Publication Date
- 2024-11-28
AI Technical Summary
Text chat systems lack the natural interaction and efficiency of face-to-face communication, as users may hesitate to speak due to uncertainty about the conversation state and appropriateness of small talk, leading to unnatural pauses and inefficient conversations.
A text conversation support device that classifies conversation states based on recent utterances and generates speech support messages to encourage user participation, using a classification model trained with conversation data to determine when to prompt users to speak and maintain conversation flow.
The solution promotes natural and efficient text conversations by eliminating unnatural silences and improving user engagement through targeted speech support messages, enhancing the overall communication experience in text chat environments.
Smart Images

Figure JP2023019138_28112024_PF_FP_ABST
Abstract
Description
Text conversation support device, method, and program
[0001] One aspect of the present invention relates to a text conversation support device, method, and program that supports a user's comments in, for example, a text chat.
[0002] In recent years, systems that use text chat to communicate instead of face-to-face or voice calls have become widespread, and the use of such systems in business situations has become particularly common recently. For example, Non-Patent Document 1 describes a system in which multiple channels corresponding to agendas are set up within a workspace, which is the area in which information is shared, and participating users are assigned to each channel as appropriate. This system makes it possible to organize various agendas in business, quickly exchange necessary information, and provide a forum for interaction between users.
[0003] “What is Slack?”, Slack, Internet <URL: https: / / slack.com / intl / ja-jp / help / articles / 115004071768-What-is-Slack>, accessed March 27, 2023
[0004] However, when using text chat tools for telecommuting or remote work, it can be difficult to convey the other person's facial expressions and situation, which can make communication less smooth than it would be in person. For example, when it comes to work-related communication, greetings and casual conversations are often natural when meeting colleagues face-to-face. However, with text chat, factors such as the inflexible reach of the recipient, difficulty in determining whether the other person is busy, and uncertainty about whether casual conversation is permitted can lead to people refraining from speaking. Furthermore, in face-to-face discussions and Q&A sessions, participants must read the other participants' thoughts and whether they are about to speak. However, such readings are difficult to discern with text chat. As a result, it can be difficult to determine when to ask a question or encourage others to speak, leading to unnatural pauses in the conversation.
[0005] The present invention has been made in light of the above circumstances, and aims to provide a technology that encourages users to speak in text chat, thereby enabling natural and efficient text conversation.
[0006] In order to solve the above problem, one aspect of a text conversation support device or support method according to the present invention, when supporting a conversation taking place in a text chat, acquires correspondence information in advance that indicates a correspondence relationship between a statement written in the text chat and a type of conversation state at the time the statement was made, and when a new statement is written in the text chat under this condition, classifies the type of conversation state corresponding to the new statement based on the correspondence information, generates a statement support message for encouraging the user to make a statement according to the classified type of conversation state, and outputs the generated statement support message.
[0007] According to one aspect of the present invention, the type of conversation state is classified based on the content of the most recent comments in text chat, and a comment support message that encourages comments according to the conversation state at that time is generated and posted in the text chat based on the classification result. Therefore, comment support is provided according to the type of conversation state at that time, thereby eliminating unnatural periods of silence and making it possible to improve the efficiency of conversations using text chat.
[0008] That is, according to one aspect of the present invention, it is possible to provide a technology that encourages users to speak in text chat, thereby enabling natural and efficient text conversation.
[0009] FIG. 1 is a diagram showing an example of a text chat system including a text conversation assistance device according to an embodiment of the present invention. FIG. 2 is a block diagram showing an example of a hardware configuration of a text conversation assistance device according to an embodiment of the present invention. FIG. 3 is a block diagram showing an example of a software configuration of a text conversation assistance device according to an embodiment of the present invention. FIG. 4 is a flowchart showing an example of a processing procedure and processing content of assistance control executed by a control unit of the text conversation assistance device shown in FIG. 3. FIG. 5 is a diagram for explaining a series of operations of a conversation state classification process executed by a control unit of the text conversation assistance device shown in FIG. 3. FIG. 6 is a diagram showing a first example of a process for generating utterance support data according to the result of the classification process shown in FIG. 5. FIG. 7 is a diagram showing a second example of a process for generating utterance support data according to the result of the classification process shown in FIG. 5.
[0010] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0011] [One Embodiment] (Configuration Example) (1) System FIG. 1 is a diagram showing an example of a text chat system including a text conversation support device according to one embodiment of the present invention.
[0012] The text chat system of one embodiment includes a chat control server SV having a text chat function, and enables text conversations between users by connecting terminal devices TM1 to TMn used by each user to the chat control server SV via a network NW.
[0013] The terminal devices TM1 to TMn are, for example, personal computers or smartphones, and are equipped with an application for conducting the text chat. Note that the terminal devices TM1 to TMn may be of any type or configuration as long as they are capable of transmitting the text.
[0014] The network NW includes a wide area network with the Internet at its core, and an access network for accessing the wide area network. Examples of the access network include, but are not limited to, a public communication network using wired or wireless connections, a local area network (LAN) using wired or wireless connections, and a cable television (CATV) network.
[0015] (2) Text Conversation Support Device The text chat system according to an embodiment of the present invention newly includes a text conversation support device SP. The text conversation support device SP is composed of a server computer located on the Web or the cloud, for example, and is capable of data communication with the chat control server SV via the network NW.
[0016] 2 and 3 are block diagrams showing an example of the hardware and software configurations of the text conversation support device SP.
[0017] The text conversation support device SP has a control unit 1 that uses a hardware processor such as a central processing unit (CPU), and this control unit 1 is connected via a bus 5 to a storage unit having a program storage unit 2 and a data storage unit 3, and a communication interface (hereinafter, interface will be referred to as I / F) unit 4.
[0018] The communication I / F unit 4 uses a communication protocol defined in the network NW to send and receive data to and from the chat control server SV.
[0019] The program storage unit 2 is configured by combining, for example, a nonvolatile memory such as a solid state drive (SSD) as a storage medium that can be written to and read from at any time, and a nonvolatile memory such as a read only memory (ROM), and stores middleware such as an operating system (OS) as well as application programs required to execute various controls according to the first embodiment. Hereinafter, the OS and each application program will be collectively referred to as a program.
[0020] The data storage unit 3 is, for example, a combination of a non-volatile memory such as an SSD that can be written to and read from at any time, and a volatile memory such as a RAM (Random Access Memory), and the storage area thereof is provided with a document data storage unit 31, a classification model storage unit 32, and a utterance support data storage unit 33.
[0021] The document data storage unit 31 stores a plurality of document data used for training the classification model. The document data is composed of samples of conversation data in text chat. Each sample consists of a statement body and a label that indicates the state of the conversation when the statement was made.
[0022] The classification model storage unit 32 stores a classification model trained using machine learning. The classification model has a function of classifying the state of a conversation at that time based on the content of a user's utterance.
[0023] The utterance support data storage unit 33 stores a plurality of utterance support messages corresponding to the classified conversation states.
[0024] The control unit 1 includes the following processing functions necessary to implement one embodiment of the present invention: a learning document data acquisition processing unit 11, a learning processing unit 12, a utterance data acquisition processing unit 13, a classification processing unit 14, a utterance support control processing unit 15, and a utterance support data output processing unit 16.
[0025] The processing units 11 to 16 are all realized by causing a hardware processor in the control unit 1 to execute an application program stored in the program storage unit 2. Note that some or all of the processing units 11 to 16 may be realized using hardware such as an LSI (Large Scale Integration) or an ASIC (Application Specific Integrated Circuit).
[0026] In the learning phase, the learning document data acquisition processing unit 11 reads out learning document data stored in advance from the document data storage unit 31 and provides the learning document data to the learning processing unit 12 .
[0027] The learning processing unit 12 constructs a classification model by having the learning model learn using the learning document data provided by the learning document data acquisition processing unit 11, and stores data relating to the constructed, trained classification model in the classification model memory unit 32.
[0028] In the operation phase, the utterance data acquisition processing unit 13 acquires the most recent utterance data and subsequent utterance data from the chat control server SV, and passes them to the utterance support control processing unit 15 .
[0029] The classification processing unit 14, under the instruction of the utterance support control processing unit 15, acquires a classification result of the conversation state corresponding to the most recent utterance data, using the trained classification model stored in the classification model storage unit 32. Note that an example of the conversation state classification process will be described in the operation example.
[0030] The utterance support control processing unit 15 instructs the classification processing unit 14 to classify the conversation states, receives the classification results from the classification processing unit 14, and selects a utterance support message corresponding to the type of conversation state represented by the received classification results from among the plurality of utterance support messages stored in the utterance support data storage unit 33. An example of the utterance support message selection process will be described in the operation example.
[0031] The utterance support data output processing unit 16 transmits the utterance support message selected by the utterance support control processing unit 15 from the communication I / F unit 4 to the chat control server SV and writes it in the chat.
[0032] (Example of Operation) Next, an example of operation of the text conversation support device SP configured as above will be described. Figure 5 shows an example of operation relating to the conversation state classification process for the learning phase and the operational phase.
[0033] (1) Learning Phase First, in the learning phase, the control unit 1 of the text-based conversation support device SP executes a learning process to build a classification model as follows.
[0034] That is, the control unit 1 of the text conversation support device SP first sequentially reads out a plurality of samples of document data stored in advance for learning from the document data storage unit 31 under the control of the learning document data acquisition processing unit 11, and provides these to the learning processing unit 12.
[0035] Each sample consists of the text of the utterance data and a label that indicates the type of conversation state when the utterance data appeared. In this example, four types of labels are defined: "conversation start," "conversation continue," "conversation end," and "solo." "Conversation start" means that the conversation has started with that utterance. "conversation continue" means that the conversation is in progress. "conversation end" means that the conversation has ended with that utterance. "solo" means that the conversation is completed with just one utterance.
[0036] Therefore, for example, if the text of a sample is "Good morning. Were you affected by the train delay?", the label "conversation started" will be associated with it. If the text of another sample is "Resolved," the label "conversation ended" will be associated with it. If the text of yet another sample is "Notice: The deadline for the survey is today," the label "single" will be associated with it.
[0037] The learning processing unit 12 uses a plurality of samples of the learning document data provided above to learn a classification model as follows.
[0038] That is, as shown in FIG. 5, the learning processing unit 12 first converts the text of each sample of the learning document data that has been read into a document vector. Known methods are used for the conversion into document vectors and for designing a classification model. For example, Doc2Vec or BERT can be used as a method for converting into document vectors. Furthermore, support vector machines (SVM) or BERT can be used as a method for designing a classification model.
[0039] Doc2Vec and BERT are described in detail in the following documents 1 and 2, respectively.
[0040] Reference 1: Le, Quoc V. and Tomas Mikolov. “Distributed Representations of Sentences and Documents.” International Conference on Machine Learning (2014). Reference 2: Devlin, Jacob, Ming-Wei Chang, Kenton Lee and Kristina Toutanova. “BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding.” ArXiv abs / 1810.04805 (2019): n. pag.
[0041] Hereinafter, in one embodiment, a description will be given assuming that BERT is used for both the conversion to document vectors and the design of a classification model.
[0042] When BERT is used, for example, the learning model described in Reference 3 is read as the learning model from the classification model storage unit 32. Then, by having the learning model learn using input data as fine tuning, a classification model is constructed that, when the text of a utterance is given, estimates the most likely conversation state for that utterance out of four conversation states.
[0043] Reference 3: BERT base Japanese (unidic-lite with whole word masking, jawiki-20200831), https: / / huggingface.co / cl-tohoku / bert-base-japanese-v2.
[0044] The learning processing unit 12 stores data relating to the learned classification model constructed as described above in the classification model storage unit 32 .
[0045] (2) Operation Phase When the operation phase is set, the control unit 1 of the text conversation support device SP executes text conversation support control as follows.
[0046] FIG. 4 is a flowchart showing an example of the procedure and content of the text conversation support control executed by the control unit 1 of the text conversation support device SP.
[0047] (2-1) Acquisition of utterance data That is, in step S10, the control unit 1 of the text conversation support device SP, under the control of the utterance data acquisition processing unit 13, first acquires new utterance data in the chat from the chat control server SV via the communication I / F unit 4.
[0048] The above-mentioned utterance data can be acquired, for example, by using an API (Application Programming Interface) provided by the chat control server SV. There are two possible acquisition methods. The first method is to receive the fact that a utterance has been made and its content in real time when a user participating in a monitored chat channel makes a utterance. The second method is to issue a request periodically or at a specific date and time and receive the most recent utterances written in the chat channel. It is also possible to acquire the utterance history of a monitored chat channel all at once at any time.
[0049] (2-2) Classification of Conversation State Next, under the control of the utterance support control processing unit 15, the control unit 1 of the text conversation support device SP classifies the type of conversation state corresponding to the acquired new utterance data X as follows:
[0050] That is, when the utterance support control processing unit 15 receives new utterance data X from the utterance data acquisition processing unit 13, first in step S11, it passes the new utterance data X to the classification processing unit 14 as a query and instructs it to classify the conversation state.
[0051] In response to this, the classification processing unit 14 first converts the text of the new utterance data X passed as a query into a document vector, as shown in, for example, FIG. 5, and then inputs the converted document vector into the trained classification model stored in the classification model storage unit 32.
[0052] The trained classification model outputs the classification result of the conversation state corresponding to the input document vector. In this example, four conversation state types are defined: "conversation start," "conversation continue," "conversation end," and "alone," so the trained classification model outputs the probability when classified into these four states.
[0053] When the classification processing unit 14 receives the classification probability for each conversation state output from the trained classification model, it selects the conversation state with the highest probability based on this classification probability, and sets the selected type of conversation state as the classification result.
[0054] For example, suppose that utterance data "Thank you" is acquired as a query, and the classification probability into four conversation states "conversation start," "conversation continue," "conversation end," and "alone" output from the trained classification model in response to the input document vector of this utterance data is [0.2, 0.2, 0.5, 0.1]. In this case, the classification processing unit 14 selects "conversation end," which is the conversation state with the highest probability among the above classification probabilities, and outputs the selected "conversation end" as the conversation state classification result.
[0055] When the utterance support control processing unit 15 receives the classification result from the classification processing unit 14 in step S11, it then determines in step S12 whether the classified conversation state is "conversation start" or "conversation continuation", or "conversation end" or "alone".
[0056] (2-3) Generation of utterance support data If the classified conversation state is "conversation start" or "conversation continue" as a result of the above determination, the utterance support control processing unit 15 sets standby pattern "A" in the timer in step S13 and monitors for timeout in step S14. On the other hand, if the classified conversation state is "conversation end" or "alone", it sets standby pattern "B" in the timer in step S15 and monitors for timeout in step S16.
[0057] Various setting examples are possible for the waiting times designated for the above waiting patterns "A" and "B," taking into account the characteristics of the target text chat, such as the average frequency of comments. For example, waiting pattern "A" assumes that comments that will lead to an ongoing conversation will be made within a short period of time, and so waiting time (A) is set to "10 minutes." On the other hand, waiting pattern "B" assumes that comments that could lead to a new conversation after an ended conversation will be made within a relatively long period of time, and so waiting time (B) is set to "30 minutes."
[0058] If the utterance support control processing unit 15 determines in step S14 or step S16 that the waiting times (A) and (B) have elapsed due to the timeout, then in step S17 it acquires an array Y of utterance data following the new utterance data from the utterance data acquisition processing unit 13. Then, in step S18, the utterance support control processing unit 15 determines whether the acquired array Y of utterance data is "empty," that is, whether or not a utterance has been made during the waiting times (A) and (B).
[0059] If the result of the above determination is that some utterance has been made during the waiting times (A) and (B), the utterance support control processor 15 determines that a message prompting the user to speak is unnecessary and terminates the process. On the other hand, if no utterance has been made during the waiting times (A) and (B), the utterance support control processor 15 proceeds to step S19, where it determines whether the waiting pattern was "A" or "B."
[0060] If the waiting pattern is "A," that is, if the waiting time is "10 minutes," the utterance support control processing unit 15 proceeds to step S20, and selectively reads out a utterance support message encouraging continuation of the conversation from the utterance support data storage unit 33, as shown in Fig. 6, for example. As the utterance support message at this time, for example, "Please leave a comment," is selected.
[0061] On the other hand, if the standby pattern is "B," that is, if the standby time is "30 minutes," the utterance support control processing unit 15 proceeds to step S21, and selectively reads out a utterance support message that prompts a new conversation from the utterance support data storage unit 33, as shown in Fig. 7, for example. As the utterance support message at this time, for example, "Is work going well today?" is selected.
[0062] (2-4) Output of utterance support data Finally, under the control of the utterance support data output processing unit 16, the control unit 1 of the text conversation support device SP transmits the utterance support data generated by the utterance support control processing unit 15 from the communication I / F unit 4 to the chat control server SV.
[0063] The chat control server SV writes the comment support message sent from the text conversation support device SP on the target chat. As a result, users participating in the chat can judge whether or not it is OK to make a comment based on the comment support message, and then make a comment. As a result, comments are encouraged, unnatural periods of silence are eliminated, and the efficiency of conversations using text chat can be improved.
[0064] (Actions and Effects) As described above, in one embodiment, a classification model is constructed that outputs a classification result for the conversation state at that time in response to input utterance data by training the learning model using multiple samples of document data as training data. When a new utterance is made, the utterance data is acquired, and the document vector of the body of the acquired utterance data is input into the trained classification model to acquire a classification result for the corresponding conversation state. Then, based on the acquired classification result, a utterance support message corresponding to the conversation state at that time is selected and written in the text chat.
[0065] Specifically, based on the most recent comment in the text chat, it determines whether the conversation is ongoing or has ended, and if the next comment is not made even though the conversation is ongoing, it displays a comment support message in the text chat encouraging the conversation to continue, and if the conversation has ended, it displays a comment support message in the text chat encouraging the start of a new conversation.
[0066] Therefore, users participating in text chat can indirectly recognize the status of other participating users through the above-mentioned speech support message and understand the timing of their speech, which encourages speech and eliminates unnatural periods of silence, thereby making it possible to improve the efficiency of conversations using text chat.
[0067] [Other Embodiments] (1) In one embodiment, an example was given in which multiple samples of learning document data are stored in advance in the document data storage unit 31, but they may also be obtained, for example, from the chat control server SV during the learning process.
[0068] (2) In one embodiment, the text conversation support device SP is installed on a server computer on the Web or on the cloud. However, it may be installed within the chat control server SV, or it may be installed on a personal computer used by a system administrator or a personal computer used by a user.
[0069] (3) The processing functions of the text-based conversation support device SP may be distributed among multiple server computers or personal computers that can transmit data to each other. In this case, for example, the classification model learning processing unit and classification model storage unit may be located in a server computer, and the processing functions used in the other operational phases may be located in each user's personal computer, thereby reducing the load on the personal computers.
[0070] (4) In addition, various modifications can be made to the configuration of the text conversation support device SP, the processing procedures and processing contents of each processing function, the types of conversation states to be classified, and examples of utterance support messages corresponding to the types of classified conversation states, without departing from the spirit of this invention.
[0071] Although the embodiments of the present invention have been described in detail above, the above description is merely an example of the present invention in every respect. It goes without saying that various improvements and modifications can be made without departing from the scope of the present invention. In other words, when implementing the present invention, specific configurations according to the embodiments may be appropriately adopted.
[0072] In short, this invention is not limited to the above-described embodiments, and in the implementation stage, the components can be modified and embodied without departing from the spirit of the invention. Furthermore, various inventions can be formed by appropriately combining multiple components disclosed in the above-described embodiments. For example, some components may be omitted from all the components shown in the embodiments. Furthermore, components from different embodiments may be appropriately combined.
[0073] SP...Text conversation support device SV...Chat control server TM1 to TMn...User's terminal device NW...Network 1...Control unit 2...Program storage unit 3...Data storage unit 4...Communication I / F unit 5...Bus 11...Learning document data acquisition processing unit 12...Learning processing unit 13...Utterance data acquisition processing unit 14...Classification processing unit 15...Utterance support control processing unit 16...Utterance support data output processing unit 31...Document data storage unit 32...Classification model storage unit 33...Utterance support data storage unit
Claims
1. A text conversation support device that supports conversations held in text chat, comprising: a memory unit that stores correspondence information indicating a correspondence between a statement written on the text chat and a type of conversation state when the statement was made; a first processing unit that, when a new statement is written on the text chat, classifies the type of conversation state corresponding to the new statement based on the correspondence information; and a second processing unit that generates a statement support message for encouraging the statement to be made for each type of conversation state in accordance with the classified type of conversation state, and outputs the generated statement support message.
2. A text conversation support device as described in claim 1, further comprising a third processing unit that acquires learning data consisting of pairs of statements written in the text chat and labels representing the types of the conversation states corresponding to the statements, constructs a classification model of the conversation states by having a learning model learn using the acquired learning data, and stores data related to the constructed classification model in the memory unit as the correspondence information.
3. The text conversation support device according to claim 1, wherein said first processing unit classifies the types of conversation states into four types: conversation start, conversation continuation, conversation end, and alone.
4. The text conversation support device according to claim 3, wherein the second processing unit generates, when the type of the conversation state corresponding to the new utterance is classified as the start of conversation or the continuation of conversation, a message encouraging the continuation of the conversation as the utterance support message if no next utterance is made within a first waiting time after the new utterance is made.
5. The text conversation support device according to claim 3, wherein the second processing unit generates a message encouraging a new conversation as the utterance support message when the type of the conversation state corresponding to the new utterance is classified as the end of the conversation or alone and no next utterance is made within a second waiting time after the new utterance is made.
6. A text conversation support method in which an information processing device executes processing related to supporting conversations held in text chat, the method comprising the steps of: acquiring correspondence information indicating a correspondence between a statement written in the text chat and a type of conversation state when the statement was made; when a new statement is written in the text chat, classifying the type of conversation state corresponding to the new statement based on the correspondence information; and generating a statement support message for encouraging the statement according to the type of conversation state classified according to the type of the conversation state, and outputting the generated statement support message.
7. A program for causing a processor included in a text conversation support device according to any one of claims 1 to 5 to execute at least one of the processes executed by the first to fourth processing units included in said text conversation support device.
Citation Information
Patent Citations
Information processing device, information processing method, and information processing program
JP2021012303A
Business support system
JP2022049551A
