JP7744283B2 - Dialogue system and center device - Google Patents

Description

The present invention relates to an interactive system and a center device that use terminal devices with voice input and output functions.

Loneliness is a factor that leads to serious health risks such as dementia, and the increase in single-person households (especially elderly people living alone) combined with the recent increased risk of infectious diseases makes loneliness more likely to occur, making alleviating loneliness an important social issue. Casual conversation and other everyday interactions are thought to be effective in alleviating loneliness.

In recent years, communication services have been proposed that aim to alleviate feelings of loneliness by installing devices with voice input and output capabilities (such as interactive robots) in the homes of elderly people, particularly those living alone, and engaging in everyday conversations such as casual voice chat through these devices.

In this system, when a user speaks to a device in their home, the content is converted into text using voice recognition technology and sent as a message to the service provider's management center. At the management center, an operator who notices the message types a reply into the center device, which is then sent back to the user.

Japanese Patent Application Laid-Open No. 2021-157419 Japanese Patent Application Laid-Open No. 2007-286376

Manual dialogue services and automated dialogue services using communication robots are available. However, labor shortages due to the declining labor force are also a social issue, and there are limits to how much manual responses operators can provide. On the other hand, unmanned communication robots may be put off by users due to the low quality of their dialogue (inappropriate responses, standard responses, etc.).

Patent Document 1 proposes a UI that supports the operator's responses by displaying candidate responses for each category (affirmative, counterargument, topic change) along with an evaluation value. However, when an operator responds to a message from a user, it is a heavy burden for the operator to think of a response message for each one, and there is a risk that the quality of the response messages will vary depending on the operator. Although Patent Document 1 displays candidate responses to the operator, the operator must check each candidate and decide on a response message, which still places a burden on the operator.

Patent Document 2 proposes technology that alternates between automatic responses and responses via an operator by requesting response assistance from a remote support device when a robot is unable to determine a response message. However, in conversations where it is not easy for a robot to determine the appropriateness of a response message, such as everyday conversations such as casual conversations, switching to operator response each time could place an excessive burden on the operator.

In light of the above circumstances, the object of the present invention is to provide a dialogue system that can conduct dialogue with a certain level of quality guaranteed while reducing the burden on humans.

A dialogue system according to one aspect of the present invention comprises:
A terminal device;
a center device capable of communicating with the terminal device;
A dialogue system comprising:
The terminal device
an input means for inputting a user message that is a message from a user;
a message sending means for sending the input user message;
a response output means for receiving and outputting a response message to the user message;
and
The center device
a receiving means for receiving the user message from the terminal device;
a reply candidate generating means for generating reply candidates which are candidates for the reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
It has.

According to this embodiment, the center device generates response candidates based on feature words and phrases extracted from user messages, and determines whether to respond to the user message automatically or manually based on the evaluation of the generated response candidates. This increases the number of opportunities for automatic responses and reduces the operator's workload compared to, for example, automatically responding to simple greetings and manually responding to other messages. Furthermore, because response candidates with positive evaluations are automatically used as response messages, the quality of the automatic responses can be ensured while reducing the operator's workload. Furthermore, in the case of manual responses, the operator does not need to generate the response message themselves; they can simply select from the displayed response candidates after editing them as needed. This reduces the operator's workload while ensuring the quality of the automatically generated response messages for manual responses.

The determination means may evaluate the response candidates by determining at least whether the response candidates are positive or negative, and may determine to provide an automatic response if the response candidates are positive, or to provide a manual response if the response candidates are negative.

When responding with positive content, it is expected that the response will be sympathetic to the user, so by providing an automatic response, it is possible to facilitate smooth dialogue with the user and reduce the burden on the operator. On the other hand, when responding with negative content, there is a risk that the user will feel uncomfortable, so the response must be confirmed by an operator beforehand, i.e., a manual response can ensure the quality of the response message.

The determination means may determine whether the answer candidate has positive or negative content based on the polarity value and/or category of the evaluative phrase contained in the answer candidate.

This allows us to appropriately determine whether a candidate response contains positive or negative content based on the polarity value (positive or negative) and category (place, food, hobby, preference, etc.) of the evaluation term.

The determination means may calculate a response evaluation value for each of the response candidates, indicating their appropriateness as a response message to the user message, and determine whether to respond automatically or manually based on the response evaluation value.

For example, if the response evaluation value is low, an automatic response may cause discomfort to the user, so by not using an automatic response and instead having an operator confirm the response, the quality of the response message can be ensured.

The determination means determines a positive/negative index value that evaluates the response candidate as positive or negative,
The response evaluation value may be calculated based on a sensitivity index value that evaluates sensitivity, and/or based on whether or not the candidate response appears in a dialogue history that includes the user message and past response messages.

By calculating the response evaluation value by taking into account not only the polarity values contained in the evaluation words of the response candidate but also other evaluation indicators, it is possible to more appropriately determine whether to respond automatically or manually.

The determination means determines that an automatic response should be made if the response evaluation value is equal to or greater than a threshold value,
The response transmitting means may transmit, to the terminal device, a response candidate whose response evaluation value is equal to or greater than a threshold value as the response message.

If the response evaluation value is above the threshold, it is assumed that the user has a high level of empathy, so an automatic response can be provided, reducing the operator's workload.

The response candidate generation means
generating a plurality of candidate responses to one of the user messages;
the determination means calculates the response evaluation value for each of the plurality of response candidates, and determines that an automatic response should be made when a difference between the response evaluation value of the response candidate with the largest response evaluation value and the response evaluation value of the other response candidates is equal to or greater than a predetermined value;
The response transmitting means may transmit the response candidate with the highest response evaluation value to the terminal device as the response message.

This allows appropriate response candidates to be automatically provided, reducing the human workload on operators.

The determination means may determine whether to respond automatically or manually to the user message based on the evaluation of the reply candidate and the evaluation of the user message corresponding to the reply candidate.

For example, if a user message contains negative content, responding with a positive message may cause the user to feel uncomfortable, so instead of responding automatically, it is better to have an operator confirm the response before responding.

A dialogue system according to one aspect of the present invention comprises:
A terminal device;
a center device capable of communicating with the terminal device;
A dialogue system comprising:
The terminal device
an input means for inputting a user message that is a message from a user;
a message sending means for sending the input user message;
a response output means for receiving and outputting a response message to the user message;
and
The center device
a receiving means for receiving the user message from the terminal device;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the user message;
and
The determination means determines, as the evaluation, at least whether the user message has positive or negative content, and determines to make an automatic response if the user message is positive, and determines to make a manual response if the user message is negative.

A center device according to one aspect of the present invention comprises:
a receiving means for receiving a user message from a terminal device that exchanges messages with a user in a two-way manner;
a reply candidate generating means for generating reply candidates which are candidates for a reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
It is equipped with:

The present invention provides a dialogue system that can conduct dialogue with a certain level of quality guaranteed while reducing the burden on humans.

Note that the effects described here are not necessarily limited to those described herein and may be any of the effects described in the present invention.

1 is a diagram showing a configuration of a dialogue system according to an embodiment of the present invention. FIG. 2 is a diagram showing a hardware configuration of a center device of the dialogue system. The operation flow of the dialogue system is shown below. 10 shows an example of a display mode of selectable phrases by the phrase display means. 10 shows an example of a display mode of selectable phrases by the phrase display means.

Embodiments of the present invention will be described below with reference to the drawings.

1. Dialogue system configuration

Figure 1 shows the configuration of the dialogue system according to this embodiment.

The dialogue system 10 provides a communication service in which everyday conversations such as casual chatting are primarily conducted. The dialogue system 10 is difficult to standardize responses to, anticipates a wide variety of responses, and enables continuous everyday conversations such as casual chatting with multiple users. Unlike automated response systems for product information, contract procedures, etc., which can be implemented without the need for human intervention, the dialogue system 10 requires human intervention from an operator to ensure the quality of the dialogue, as emotions such as empathy and intentions are important.

The dialogue system 10 is composed of a center device 100 as a server device managed by a business providing services through the dialogue system 10, an operator device 160 connected to the center device 100, and multiple terminal devices 200 connected to a network so as to be able to communicate with the center device 100.

The terminal device 200 is installed in the home of a user of this system (e.g., an elderly person). The terminal device 200 has an input means 201, a message sending means 202, and a response output means 203. The input means 201 inputs a user message in the form of voice data from the user via a microphone M. The message sending means 202 sends the input user message to the center device 100 via the network. The response output means 203 receives a response message to the user message from the center device 100 via the network, and outputs it as voice to the user via a speaker S.

The terminal device 200 may have at least the above components, but may also be a conversational robot modeled after a small doll, with an appearance that makes the user feel familiar and reminds them of conversations with people (especially everyday conversations such as casual chatting).

The center device 100 is managed by a business operator that provides services using the dialogue system 10, and includes at least a receiving means 101, an extraction means 102, a response candidate generation means 103, a determination means 104, a phrase display means 105, a response candidate display means 106, and a response sending means 107. It also includes a control means for controlling these means. These are realized as appropriate by well-known hardware (such as a server computer or personal computer) or software stored in storage means 140.

The storage means 140 stores a dictionary 141 and a dialogue history 142. The dictionary 141 stores categories (places, foods, hobbies, preferences, etc.) and polarity values (semantic flags such as positive or negative) in association with words. The dictionary 141 may be set for each user, may be user-independent, or both. The dialogue history 142 may be data that accumulates past dialogues for each user, or user-independent data that accumulates aggregating the dialogue histories of multiple users (multiple terminal devices 200), or both.

The operator device 160 is managed by the business providing the service through the dialogue system 10, and displays candidate responses to user messages received by the center device 100 on a web browser for the operator. The center device 100 and the operator device 160 are in a web server-web client relationship.

The center device 100 and the operator device 160 may also be integrated into hardware.

2. Center device hardware configuration

Figure 2 shows the hardware configuration of the center device 100.

As shown in the figure, the center device 100 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM (Random Access Memory) 13, an input/output interface 15, and a bus 14 that connects these together.

The CPU 11 accesses the RAM 13 and other memory as needed, and performs various calculation processes while providing overall control over each block of the center device 100. The ROM 12 is a non-volatile memory that permanently stores firmware, such as the OS, programs, and various evaluation indices, that are executed by the CPU 11. The RAM 13 is used as a working area for the CPU 11, and temporarily stores the OS, various applications currently being executed, and various data being processed.

The input/output interface 15 is connected to a display unit 16, an operation reception unit 17, a memory unit 18, a communication unit 19, etc.

The display unit 16 is a display device that uses, for example, an LCD (Liquid Crystal Display), an OELD (Organic ElectroLuminescence Display), or a CRT (Cathode Ray Tube).

The operation reception unit 17 is, for example, a pointing device such as a mouse, a keyboard, a touch panel, or other input device.

The storage unit 18 is a non-volatile memory such as a hard disk drive (HDD), flash memory (SSD; solid state drive), or other solid-state memory. In addition to the OS, the storage unit 18 stores schedule information, various flags, evaluation indicators, and software that implements the various means of the center device 100. The storage unit 18 also stores applications and other programs and databases for exchanging messages with the terminal device 200, including response message generation.

The communication unit 19 is, for example, a NIC (Network Interface Card) for Ethernet (registered trademark) or various modules for wireless communication such as wireless LAN, and is responsible for communication processing with the terminal device 200.

The hardware configuration of the operator device 160 and terminal device 200 is basically the same as that of the center device 100, except that the terminal device 200 has a microphone M and speaker S on the front, as described above.

3. Operational flow of the dialogue system 10

Figure 3 shows the operational flow of the dialogue system.

The input means 201 of the terminal device 200 inputs a user message from the user via the microphone M (Step S1, Yes). The user message is voice data. The message transmission means 202 of the terminal device 200 transmits the input user message to the center device 100 via the network. The message transmission means 202 may perform voice recognition on the voice data user message and convert it into a text data user message to transmit to the center device 100, or may transmit the voice data user message to the center device 100.

The receiving means 101 of the center device 100 receives the user message from the terminal device 200 via the communication unit 19 (step S2). The receiving means 101 may receive the user message in voice data and convert it into a user message in text data, or may receive a user message in text data converted from voice data by the terminal device 200.

The extraction means 102 of the center device 100 performs morphological analysis on the user message, which is text data, to divide it into multiple words and phrases, and extracts one or more characteristic words and phrases contained in the user message from these multiple words and phrases (step S3). For example, the extraction means 102 registers a dictionary 141 in which categories are assigned to each word and phrase, and extracts words from a predetermined category (place, food, hobbies, preferences, etc.) as characteristic words and phrases. The extraction means 102 may also extract different characteristic words and phrases for each user. Specifically, words (words) related to each user's hobbies and preferences are registered in the dictionary 141 as individual characteristic words and phrases for each user, and characteristic words and phrases are extracted based on the individual characteristic words and phrases related to the user with whom the conversation is taking place. In extracting characteristic words and phrases, only individual characteristic words and phrases may be extracted, or individual characteristic words and phrases may be extracted preferentially. Additionally, for example, words related to a user's hobbies and preferences may be registered in advance in dictionary 141 as characteristic words and phrases, or conversations with users may be accumulated as conversation history 142, and each user's hobbies and preferences may be identified from the conversations and registered as individual characteristic words and phrases. Furthermore, each user's hobbies and preferences may be learned from their past user messages using AI or machine learning, and words from the user messages that match the hobbies and preferences may be registered as individual characteristic words and phrases. Note that in this embodiment, "words" include single words such as nouns, and phrases containing multiple words.

The response candidate generation means 103 of the center device 100 generates one or more response candidates that are candidates for response messages that resonate with the user message based on one or more characteristic phrases. The response candidate generation means 103 may generate response candidates by using AI or machine learning to learn various user messages and appropriate response messages for them. In addition, the response candidate generation means 103 generates rule-based, standard response candidates (e.g., "Welcome home") for pre-registered phrases such as greetings (e.g., "I'm home").

As a first example, the extraction means 102 of the center device 100 performs morphological analysis on the user message "I'm back. When I went to the park, there were beautiful camellias in bloom." to extract one or more characteristic phrases "park" and "camellia" (step S3). The response candidate generation means 103 generates a response candidate "Welcome back. There were beautiful camellias in bloom. That's wonderful." based on the characteristic phrase "camellia," and further generates another response candidate based on the characteristic phrase "park" (step S4).

As a second example, the extraction means 102 of the center device 100 performs morphological analysis on the user message "I'm back. Today's day service was fun." to extract the characteristic phrases "day service" and "fun" (step S3). The response candidate generation means 103 generates the response candidate "Welcome back. You had fun at the day service. That's great." based on the characteristic phrases "day service" and "fun" (step S4).

As a third example, the extraction means 102 of the center device 100 performs morphological analysis on the user message "I'm not feeling well today" to extract the characteristic phrases "health" and "bad" (step S3). The response candidate generation means 103 generates the response candidate "You're not feeling well. That's unfortunate" based on the characteristic phrases "health" and "bad" (step S4).

The determination means 104 of the center device 100 calculates a response evaluation value, which is an evaluation value that evaluates the appropriateness of the response candidate, and determines whether to respond automatically or manually to the user message based on the response evaluation value (step S5). The determination method used by the determination means 104 will be explained in detail later.

(1) Automatic response

If it is determined that an automatic response should be made, the response sending means 107 of the center device 100 sends one of the one or more response candidates generated by the response candidate generation means 103 (step S4) as a response message to the terminal device 200 (step S9). In other words, when an automatic response is made, the center device 100 does not need to display the user message or response candidates on the operator device 160. The response sending means 107 may convert the text data response message into a voice data response message and send it to the terminal device 200, or may send the text data response message to the terminal device 200.

The response output means 203 of the terminal device 200 receives a response message from the center device 100 and outputs the response message from the speaker S at a predetermined timing (for example, immediately after reception, when a human presence sensor detects a person around the terminal device 200). The response output means 203 may receive and output a voice data response message from the center device 100, or may receive a text data response message from the center device 100 and convert it into a voice data response message using voice synthesis technology before outputting it.

(2) Manual response

If it is determined that a manual response is required, the phrase display means 105 of the center device 100 displays the user message on the operator device 160 used by the operator, and also displays one or more selectable phrases as selectable phrases. The one or more selectable phrases are at least a portion of the one or more characteristic phrases extracted from the user message (step S3). The phrase display means 105 may display all of the one or more characteristic phrases as one or more selectable phrases.

The phrase display means 105 may display some of the one or more characteristic phrases as one or more selectable phrases based on the number of times the one or more characteristic phrases appear and/or their position in the user message. For example, the phrase display means 105 may display, as selectable phrases, only characteristic phrases that appear more frequently or that appear later in the user message. The number of times the one or more characteristic phrases appear may be the number of times they appear in the current user message, or the number of times they appear in the dialogue history 142 including past user messages and response messages.

Furthermore, the phrase display means 105 may prioritize individual feature phrases related to the user with whom the conversation is taking place as selectable phrases out of one or more feature phrases. For example, only individual feature phrases may be displayed as selectable phrases, or individual feature phrases may be displayed as selectable phrases with priority.

In a first example, the phrase display means 105 may display all of the one or more characteristic phrases "park" and "camellia" as selectable phrases. "Displaying selectable" means that the center device 100, which is a web server, displays them with hyperlinks on the web browser of the operator device 160, which is a web client, so that they can be selected by manual operation by the operator (clicking, hovering the mouse over, tapping, etc.).

Figures 4 and 5 show examples of how selectable phrases are displayed by the phrase display means.

As shown in FIG. 4, the phrase display means 105 may display one or more selectable phrases 111 within the user message 112 in a manner that allows them to be distinguished from the one or more non-selectable phrases. As shown in FIG. 5, the phrase display means 105 may display one or more selectable phrases 111 together outside the user message 112 in a manner that allows them to be distinguished from the one or more non-selectable phrases. A distinguishable manner means, for example, bolding, a different color, underlining, boxed text, etc. As shown in FIG. 5, displaying the selectable phrases 111 together outside the user message 112 with a heading 115 that reads "Keywords" is also one type of distinguishable manner. By prominently displaying the selectable phrases 111, the operator can easily select the selectable phrases 111. The phrase display means 105 further displays a cursor 113 for selecting the selectable phrase 111.

The phrase display means 105 may highlight some of the one or more selectable phrases. Highlighting may, for example, be a process of further changing the color of selectable phrases that are displayed in bold, further bolding selectable phrases that are displayed in a different color from the other phrases, or thickening or changing the color of the boxed frame of selectable phrases that are displayed in boxed text, but is not limited to these. This makes more important words (what the user wants to talk about) stand out among the selectable phrases, allowing the operator to naturally select response candidates that are in line with the user's wishes.

For example, some of the one or more selectable phrases may be highlighted based on the number of times that one or more characteristic phrases appear and/or their position in the user message. Specifically, the phrase display means 105 may highlight selectable phrases that appear more frequently or that appear later in the user message, among the one or more selectable phrases, to make it easier for the operator to select them compared to other selectable phrases. The number of times that one or more characteristic phrases appear may be the number of times they appear in the current user message, or the number of times they appear in the dialogue history 142 that includes past user messages and response messages. In this way, by setting selectable phrases for each user, it becomes possible to naturally have a dialogue that is tailored to the user.

Furthermore, the phrase display means 105 may highlight individual characteristic phrases related to the user with whom the conversation is taking place, out of one or more selectable phrases. For example, only the individual characteristic phrases may be highlighted, or at least the individual characteristic phrases may be highlighted. In this way, by setting the selectable phrases to be highlighted for each user, it becomes possible to naturally have a conversation that is tailored to the user.

The operator refers to the user message and one or more selectable phrases displayed on the operator device 160 and selects one of the selectable phrases by operating it (clicking, hovering the mouse, tapping, etc.). This allows the operator to intuitively determine the content of the response that best suits their intentions.

The answer candidate display means 106 of the center device 100 determines one selectable phrase selected by the operator via the operator device 160 from one or more selectable phrases in the user message (Step S6, Yes). The answer candidate display means 106 displays one or more answer candidates generated by the answer candidate generation means 103 based on the selected selectable phrase (Step S4) on the operator device 160 in a selectable and editable manner (Step S7). If there are multiple answer candidates generated based on the selected selectable phrase, the answer candidate display means 106 may display the multiple answer candidates in descending order of response evaluation value (described below). This allows the operator to select a response candidate that suits their wishes.

In the first example, the answer candidate display means 106 determines one selectable word "camellia" selected from one or more selectable words "park" and "camellia" displayed on the operator device 160. The answer candidate display means 106 displays the answer candidate 114 "Welcome home. Beautiful camellias are blooming. That's wonderful." (see Figures 4 and 5) generated by the answer candidate generation means 103 based on the selected selectable word "camellia" (step S4) on the operator device 160 in a selectable and editable manner (step S7).

The operator uses the cursor 113 to select one of the one or more response candidates displayed on the operator device 160. This allows the operator to edit the response candidates as needed and select the edited response candidate. This allows the operator to intuitively decide on response candidates, and by editing response candidates individually for each user, it becomes possible to naturally have a conversation that is more in tune with the user.

The response sending means 107 of the center device 100 determines one response candidate selected via the operator device 160 from one or more response candidates displayed on the operator device 160 (step S8). The response sending means 107 sends the selected response candidate to the terminal device 200 as a response message (step S9). Note that, when the selected response candidate is edited, the response sending means 107 sends the edited response candidate to the terminal device 200 as a response message. The response sending means 107 may convert the text data response message into a voice data response message using voice synthesis technology and send it to the terminal device 200, or may send the text data response message to the terminal device 200.

The response output means 203 of the terminal device 200 receives a response message from the center device 100 and outputs the response message from the speaker S at a predetermined timing (for example, immediately after reception, when a human presence sensor detects a person around the terminal device 200). The response output means 203 may receive and output a voice data response message from the center device 100, or may receive a text data response message from the center device 100 and convert it into a voice data response message using voice synthesis technology before outputting it.

4. Judgment means

The determination means 104 of the center device 100 determines whether to respond to the user message automatically or manually based on the evaluation of the response candidates (step S5). The response candidates are evaluated for their appropriateness as a response message to the user message, taking into consideration factors such as whether they are considerate to the user and whether they cause discomfort. Specific examples of the determination method used by the determination means 104 are described below.

(1) Example 1

For example, the evaluation of the response candidate may be binary, positive or negative, and the determination means 104 determines that an automatic response will be made if the evaluation of the response candidate is positive, and determines that a manual response will be made if the evaluation of the response candidate is negative. In other words, if the response is positive, it is expected to be a response that is considerate of the user. For this reason, the determination means 104 determines that an automatic response will be made to facilitate smooth dialogue with the user and reduce the burden on the operator. On the other hand, if the response is negative, there is a risk that it will cause discomfort to the user. For this reason, the determination means 104 determines that an automatic response will not be made, but that a response will be made after the operator has performed an operation to determine the response message, i.e., that a manual response will be made.

For example, the determination means 104 extracts evaluative words from the answer candidates. "Evaluative words" are words and individual characteristic words associated with polarity values (semantic flags such as positive or negative) and categories (places, foods, hobbies, preferences, etc.) in the dictionary 141. The evaluation of the answer candidate may be determined to be negative or positive based on the polarity value that appears most frequently among the evaluative words with positive polarity values and the evaluative words with negative polarity values included in the answer candidate. The determination means 104 may determine the evaluation of the answer candidate to be negative or positive based on the polarity value of the evaluative word that appears last among multiple evaluative words included in the answer candidate. If the answer candidate includes an evaluative word with a negative polarity value, the determination means 104 may determine the evaluation of the answer candidate to be negative regardless of the number or position of positive evaluative words.

The determination means 104 may also determine whether the evaluation of a response candidate is negative or positive based on the category of the evaluative words included in the response candidate. For example, if the response candidate contains words from a specific category (places, food, hobbies, etc.), it may be determined to be positive. Furthermore, based on the category of individual characteristic words related to the user with whom the conversation is taking place, the response candidate may be determined to be positive if it contains evaluative words related to the preferences of the user with whom the conversation is taking place, and may be determined to be negative if it contains evaluative words related to things the user with whom the conversation is taking place dislikes.

(2) Example 2

Alternatively, the determination means 104 may calculate a response evaluation value for each reply candidate that indicates its appropriateness as a response message to the user message, and determine whether to respond automatically or manually to the user message based on the response evaluation value. The response evaluation value may be calculated to be higher for reply candidates that are more likely to empathize with the user message. For example, the determination means 104 may calculate a higher response evaluation value if the reply candidate includes the user's hobbies and preferences, or if it includes words with positive polarity values, emotive words, or words from a specific category.

For example, the determination means 104 extracts evaluative words from the response candidates. The determination means 104 may determine the response evaluation value based on the polarity value (positive or negative), category (place, food, hobby or preference, individual characteristic phrase, etc.) and/or position within the user message (first half, second half, etc.) of the evaluative words.

Specifically, the determination means 104 calculates a response evaluation value based on a positive/negative index value that determines whether a response candidate is positive or negative from the number and position of positive or negative evaluation words in the response candidate. For example, the determination means 104 uses 0 as a base point and adds or subtracts 1 point for each evaluation word with a positive or negative polarity value, and calculates the total score as the response evaluation value of the response candidate. Note that the determination means 104 may determine that a response candidate is positive if the total score is equal to or greater than a predetermined value (e.g., a positive value), or that a response candidate is negative if the total score is less than the predetermined value (e.g., a negative value).

(3) Example 3

The determination means 104 may weight (e.g., convert to 1.5 points) words that appear later in a response candidate (e.g., evaluative words used in the last part of a response candidate, evaluative words used in the last sentence of a response candidate). The determination means 104 may add an additional point (e.g., +1 point) if the positive evaluative words are words in a category related to the user's hobbies and preferences. If the determination means 104 evaluates a response candidate as including a negative evaluative word, the determination means 104 may set the response evaluation value of the response candidate to the lowest point, regardless of whether the response candidate includes a positive evaluative word.

The determination means 104 may determine that an automatic response should be made if the calculated response evaluation value is equal to or greater than a threshold value. If the response evaluation value is equal to or greater than the threshold value, it is assumed that the degree of empathy with the user is high, and therefore an automatic response can be made, thereby reducing the human workload of the operator.

When the response candidate generation means 103 generates multiple response candidates, the determination means 104 may calculate a response evaluation value for each of the multiple response candidates, and determine to issue an automatic response if there is one response candidate whose response evaluation value differs from the other response evaluation values by a predetermined value or more. For example, if the response candidate with the highest response evaluation value is greater than the evaluation values of all other response candidates by a predetermined value or more, the highest response candidate is automatically used as a response message. This allows the appropriate response candidate to be automatically provided, reducing the human workload on the operator.

The determination means 104 may determine whether to respond automatically or manually to a user message based on the evaluation of the user message in addition to the evaluation of the response candidates. That is, the determination means 104 may evaluate the user message in the same manner as for evaluating the response candidates, taking into account the polarity value and category of the user message, the position of the evaluative phrase in the user message, and the like. For example, if the user message contains negative content, responding with positive content may cause discomfort to the user, so an automatic response may not be performed, and the response may be confirmed by an operator beforehand.

Specifically, the determination means 104 may further add or subtract points from the response evaluation value of the response candidate, taking into account the positive and negative relationship between the user message and the response candidate. For example, when the user message is positive and the response candidate is positive, it is assumed that the degree of empathy is high, so the response evaluation value is added. When the user message is negative and the response candidate is positive, no addition is made. When the user message is positive and the response candidate is negative, it is likely to cause discomfort to the user, so confirmation by an operator is required and points are subtracted. When the user message is negative and the response candidate is negative, it is assumed that the degree of empathy is not low, so no deduction is made.

In a second example, the determination means 104 determines that the evaluation of the user message "I'm back. I had fun at the day service today." contains an evaluative phrase "It was fun" with a positive polarity value, and the evaluation of the user message is positive. The determination means 104 determines that the response evaluation of the response candidate "Welcome back. I saw you had fun at the day service. That's great." contains evaluative phrases "It was fun" and "Good" with a positive polarity value, and the response evaluation of the response candidate is positive. The determination means 104 may determine that the response candidate "Welcome back. I saw you had fun at the day service. That's great" should be an automatic response based on a combination of the evaluation of the user message (positive) and the evaluation of the response candidate (positive).

In a third example, the determination means 104 determines that the evaluation of the user message "I'm not feeling well today" is negative because the evaluative phrase "bad" contained in the user message has a negative polarity value. The determination means 104 determines that the evaluation of the response candidate "You're not feeling well. That's unfortunate" is negative because the evaluative phrases "bad" and "unfortunate" contained in the response candidate "You're not feeling well. That's unfortunate" have negative polarity values. The determination means 104 may determine that the response candidate "You're not feeling well. That's unfortunate" should be a manual response (i.e., a response after confirmation by an operator) based on a combination of the evaluation of the user message (negative) and the evaluation of the response candidate (negative).

(4) Example 4

The determination means 104 may calculate the response evaluation value by taking into account not only the polarity value included in the evaluation words of the response candidate, but also other evaluation indicators (a) to (c). The determination means 104 may add points when each evaluation indicator is met. The determination means 104 may weight and add points for some of the evaluation indicators (for example, (a) the item with the sensitivity index value of the response candidate may be given 2 points and other items 1 point). The determination means 104 may calculate each evaluation indicator as a numerical value between 0 and 1, for example, and take a weighted average. The determination means 104 may also use a point deduction system instead of adding points to the evaluation value.

The determination means 104 may calculate the response evaluation value by adding or subtracting points based on the above-mentioned separate evaluation index: (a) the sensitivity index value of the response candidate, (b) whether or not the candidate appears in the dialogue history 142 including the user message and the response message, and/or (c) the number of times the candidate appears in the dialogue history 142 over a predetermined period. The determination means 104 may determine that if the total score is a positive value, the response evaluation value is positive and therefore an automatic response should be made; if the total score is a negative value, the response evaluation value is negative and therefore a manual response should be made; or may determine that if the response evaluation value is equal to or greater than a threshold, an automatic response should be made.

(a) When the sensitivity index value, which indicates whether the content of a candidate response is sensitive and may cause discomfort to the user, indicates a non-sensitive value, the determination means 104 may add points to the response evaluation value of the candidate response, making it more likely to be determined to be an automatic response. For example, if a candidate response is composed of standard phrases (such as greetings) or phrases that tend to co-occur (be used together) with standard phrases, the candidate response is not sensitive and therefore may be added points. This determination may be made by estimating the degree of co-occurrence from a dictionary in which co-occurring phrases are registered or from the dialogue history.

(b) When a response candidate has a history of appearing in the dialogue history 142, the determination means 104 may add points to the response evaluation value of the response candidate, making it more likely to be determined to be an automatic response. For example, the determination means 104 may compare the generated response candidate with the user-specific or user-independent dialogue history 142, and if the response candidate has been used as a response message, determine that the response candidate has a history of appearing in the dialogue history 142 and add points. This is because similar dialogues often generate similar response messages, and if a response has been used in the past, it is assumed that there is a low risk in responding.

(c) When a phrase or expression used in a reply candidate appears infrequently in the dialogue history 142 over a predetermined period, the determination means 104 may increment the response evaluation value of the reply candidate to make it more likely to be determined as an automatic response. That is, the determination means 104 may compare the dialogue history 142 with the user over a certain period (e.g., one day or one week) with the generated reply candidate to determine whether equivalent phrases or expressions are used, and may deduct points if they are used, or may increment points if they are not. The determination means 104 may increment the response evaluation value of a reply candidate when it does not include an expression used in a dialogue with the user over the most recent predetermined period, making it more likely to be determined as an automatic response. This is because, while a reply candidate may include a convenient phrase such as "That's good" or "That's fun," repeated use of the same expression can become formulaic and undesirable. For this reason, the determination means 104 may increment a response candidate that uses a different expression and decrement a response candidate that uses an expression used in a dialogue over the most recent predetermined period.

5. Conclusion

The dialogue system 10 provides a communication service in which everyday conversations such as casual chatting are primarily conducted. The dialogue system 10 is difficult to standardize responses to, anticipates a wide variety of responses, and enables continuous everyday conversations such as casual chatting with multiple users. Unlike automated response systems for product guidance, contract procedures, etc., which can be implemented without the need for human intervention, the dialogue system 10 requires human operators to ensure the quality of the dialogue, as emotions such as empathy and intentions are important. Meanwhile, labor shortages are also a social issue, and there are limits to what operators can do manually.

Therefore, first, according to this embodiment, the center device 100 generates reply candidates (step S4) based on feature words and phrases extracted from the user message (step S3), and determines whether to automatically or manually respond to the user message based on a response evaluation value, which is an evaluation value of the generated reply candidates (step S5). This increases the number of opportunities for automatic responses and reduces the operator's workload compared to, for example, automatically responding with a simple greeting and manually responding to everything else. Furthermore, because reply candidates with a positive response evaluation value are automatically used as a response message, the quality of the automatic response can be ensured while reducing the operator's workload. Furthermore, in the case of manual responses, the operator does not need to generate the response message themselves; they simply select from the displayed reply candidates after editing them as necessary. This ensures the quality of the automatically generated response message for manual responses while reducing the operator's workload.

Secondly, according to this embodiment, even when a manual response is made via an operator, appropriate selectable phrases are displayed to the operator based on feature phrases extracted from the user message (step S3), and response candidates generated (step S4) based on the selectable phrases selected by the operator (step S6) are displayed to the operator (step S7) and allowed to be selected (step S8). This minimizes the burden on the operator of selection operations while displaying appropriate response candidates in line with the user message and outputting a response message selected from the response candidates to the user, thereby ensuring the quality of the dialogue.

6. Modifications

In this embodiment, the first process determines whether to provide an automatic or manual response. The second process, in the case of a manual response, displays selectable phrases so that they can be selected, and displays selectable response candidates based on the selected selectable phrase so that they can be selected. Both the first process (determining whether to provide an automatic or manual response) (step S5) and the second process (displaying selectable phrases so that they can be selected) (steps S6 to S8) contribute to reducing the human workload on operators and ensuring the quality of responses.

Therefore, as a modified example, omitting the first process (determining whether to respond automatically or manually) and simply performing the second process (displaying selectable phrases in a selectable manner) will contribute to reducing the operator's workload and ensuring the quality of responses. In other words, the determination of whether to respond automatically or manually (step S5) may be omitted, and manual responses may be provided to almost all user messages, excluding standard phrases such as greetings. In this case, the manual response involves displaying selectable phrases in a selectable manner and selecting a response message from response candidates based on the selected selectable phrase (steps S6 to S8). Even with this method, the operator's only processing involves selecting a selectable phrase and selecting a response candidate, thereby contributing to reducing the operator's workload and ensuring the quality of responses.

If the determination of whether to use the first automatic or manual response (step S5) is omitted, the center device 100 may extract feature words from the user message (step S3), have the operator select selectable words that are at least part of the feature words (step S6, Yes), and then generate response candidates based on the selected selectable words (step S4) and display the generated response candidates in a selectable manner (step S7). In other words, the order may be reversed from the above embodiment. This reduces the processing load and processing time of the center device 100 by generating only response candidates based on the selected selectable words and eliminating the need to generate response candidates that are not displayed on the operator device 160. On the other hand, if response candidates are generated (step S4) before the operator selects a selectable word (step S6, Yes), as in the above embodiment, response candidates can be displayed in a selectable manner (step S7) in a selectable manner shortly after the operator selects a selectable word (step S6, Yes), thereby streamlining the dialogue with the user.

As another variation, omitting the second process (displaying selectable phrases in a selectable manner) and simply performing the first process (determining whether to use an automatic or manual response) can contribute to reducing the operator's workload and ensuring the quality of the response. That is, the display of response candidates based on the selectable phrases selected by the operator (steps S6 to S8) can be omitted, and response candidates generated based on characteristic phrases can be displayed. In this case, in the case of a manual response, the operator does not generate the response message himself, nor does the operator select the selectable phrases; rather, the operator simply edits the displayed response candidates as necessary and selects one. This reduces the operator's workload while ensuring the quality of the response message based on the automatically generated response candidates. Furthermore, in the case of an automatic response, response candidates with positive response evaluation values are automatically used as the response message, thereby ensuring the quality of the automatic response while reducing the operator's workload.

In addition, in this embodiment, we have described a system that determines whether to provide an automatic or manual response based on a response evaluation value, and a system that determines whether to provide an automatic or manual response based on a response evaluation value and a user evaluation value. However, a system that determines whether to provide an automatic or manual response may also be based on a user evaluation. The method for evaluating user messages is the same as that described in the embodiment, and therefore will not be described here. If the evaluation of the user message is positive or has a high evaluation value (e.g., above a threshold), it is assumed that the user message is about a fun, everyday event. In this case, an automatic response is determined to be appropriate, as there is little risk of the user feeling uncomfortable if the user responds with content consistent with the user message. On the other hand, if the evaluation of the user message is negative or has a low evaluation value, it is assumed that the user message is about a sensitive topic, such as a worry or health issue. In this case, a manual response is determined to be appropriate, as there is a risk that an automatic response may cause discomfort to the user. This ensures the quality of the automatic response while reducing the human workload on the operator.

Furthermore, in this embodiment, we have described how some feature words (for example, individual feature words) are preferentially extracted as feature words, displayed as selectable words, or highlighted. As a variation, information on recent topics such as news and disaster information is obtained from online news sites, SNS (Social Networking Services), etc., and registered in the dictionary 141 as current affairs feature words. Then, similar to the explanation for individual feature words, current affairs feature words may be preferentially extracted as feature words, displayed as selectable words, or highlighted.

Although various embodiments and variations of the present technology have been described above, the present technology is not limited to the above-described embodiments, and various modifications can of course be made without departing from the spirit of the present technology.

The dialogue system according to one embodiment of the present invention can contribute to resolving social issues such as loneliness and isolation among the elderly, extending healthy lifespan, improving the quality of life (QoL) of the elderly, and a declining labor force.

10 Dialogue system 100 Center device 101 Receiving means 102 Extracting means 103 Response candidate generating means 104 Determining means 105 Phrase displaying means 106 Response candidate displaying means 107 Response sending means 111 Selectable phrase 112 User message 113 Cursor 114 Response candidate 140 Storage means 141 Dictionary 142 Dialogue history 160 Operator device 200 Terminal device 201 Input means 202 Message sending means 203 Response output means M Microphone S Speaker

Claims (10)

A terminal device;
a center device capable of communicating with the terminal device;
A dialogue system comprising:
The terminal device
an input means for inputting a user message that is a message from a user;
a message sending means for sending the input user message;
a response output means for receiving and outputting a response message to the user message;
and
The center device
a receiving means for receiving the user message from the terminal device;
a reply candidate generating means for generating reply candidates which are candidates for the reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
and
The determination means determines whether the reply candidate has a positive content or a negative content as an evaluation of the reply candidate, and determines to make an automatic response if the reply candidate is positive, and determines to make a manual response if the reply candidate is negative.
Dialogue system. 2. The dialogue system according to claim 1 ,
The determination means determines whether the answer candidate has positive content or negative content based on a polarity value and/or a category of an evaluative phrase included in the answer candidate. A terminal device;
a center device capable of communicating with the terminal device;
A dialogue system comprising:
The terminal device
an input means for inputting a user message that is a message from a user;
a message sending means for sending the input user message;
a response output means for receiving and outputting a response message to the user message;
and
The center device
a receiving means for receiving the user message from the terminal device;
a reply candidate generating means for generating reply candidates which are candidates for the reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
and
the determination means calculates a response evaluation value indicating the appropriateness of the response candidate as a response message to the user message, determines whether to provide the automatic response or the manual response based on the response evaluation value, and calculates a positive /negative index value evaluating the positive or negative of the response candidate;
a response evaluation value calculated based on a sensitivity index value that evaluates sensitivity and/or whether or not the response candidate has appeared in a dialogue history including the user message and past response messages. Dialogue system according to claim 3 ,
The determination means determines that an automatic response should be made if the response evaluation value is equal to or greater than a threshold value,
The response sending means sends the candidate response whose response evaluation value is equal to or greater than a threshold value as the response message to the terminal device. A terminal device;
a center device capable of communicating with the terminal device;
A dialogue system comprising:
The terminal device
an input means for inputting a user message that is a message from a user;
a message sending means for sending the input user message;
a response output means for receiving and outputting a response message to the user message;
and
The center device
a receiving means for receiving the user message from the terminal device;
a reply candidate generating means for generating reply candidates which are candidates for the reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
and
the determining means calculates a response evaluation value indicating the appropriateness of the response candidate as a response message to the user message, and determines whether to make the automatic response or the manual response based on the response evaluation value;
the reply candidate generating means generates a plurality of reply candidates for one of the user messages;
the determination means calculates the response evaluation value for each of the plurality of response candidates, and determines that an automatic response should be made when a difference between the response evaluation value of the response candidate with the largest response evaluation value and the response evaluation value of the other response candidates is equal to or greater than a predetermined value;
The response sending means sends the response candidate with the highest response evaluation value to the terminal device as the response message. A terminal device;
a center device capable of communicating with the terminal device;
A dialogue system comprising:
The terminal device
an input means for inputting a user message that is a message from a user;
a message sending means for sending the input user message;
a response output means for receiving and outputting a response message to the user message;
and
The center device
a receiving means for receiving the user message from the terminal device;
a reply candidate generating means for generating reply candidates which are candidates for the reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
and
The determination means determines whether to respond automatically or manually to the user message based on the evaluation of the reply candidate and the evaluation of the user message corresponding to the reply candidate. a receiving means for receiving a user message from a terminal device that exchanges messages with a user in a two-way manner;
a reply candidate generating means for generating reply candidates which are candidates for a reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
Equipped with
The determination means determines whether the reply candidate has a positive content or a negative content as an evaluation of the reply candidate, and determines to make an automatic response if the reply candidate is positive, and determines to make a manual response if the reply candidate is negative.
Center device. a receiving means for receiving a user message from a terminal device that exchanges messages with a user in a two-way manner;
a reply candidate generating means for generating reply candidates which are candidates for a reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
Equipped with
The determination means calculates a response evaluation value indicating the appropriateness of the candidate response as a response message to the user message, determines whether to respond automatically or manually based on the response evaluation value, and
Positive/negative index value evaluated as positive or negative,
a sensitivity index value that evaluates the sensitivity, and/or
Whether or not the candidate response appears in a dialogue history including the past user messages and response messages.
The response evaluation value is calculated based on the
Center device. a receiving means for receiving a user message from a terminal device that exchanges messages with a user in a two-way manner;
a reply candidate generating means for generating reply candidates which are candidates for a reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
Equipped with
the determining means calculates a response evaluation value indicating the appropriateness of the candidate response as a response message to the user message, and determines whether to make the automatic response or the manual response based on the response evaluation value;
the reply candidate generating means generates a plurality of reply candidates for one of the user messages;
the determination means calculates the response evaluation value for each of the plurality of response candidates, and determines that an automatic response should be made when a difference between the response evaluation value of the response candidate with the largest response evaluation value and the response evaluation value of the other response candidates is equal to or greater than a predetermined value;
The response sending means sends the response candidate with the highest response evaluation value to the terminal device as the response message.
Center device. a receiving means for receiving a user message from a terminal device that exchanges messages with a user in a two-way manner;
a reply candidate generating means for generating reply candidates which are candidates for a reply message to the user message;
a determining means for determining whether to automatically or manually respond to the user message based on the evaluation of the reply candidates;
a response sending means for sending the response message determined by an operator's operation to the terminal device in the case of a manual response, and for sending the response candidate to the terminal device as the response message in the case of an automatic response;
Equipped with
The determining means determines whether to automatically respond or manually respond to the user message based on the evaluation of the reply candidate and the evaluation of the user message corresponding to the reply candidate.
Center device.