JP6909189B2 - Programs, servers and methods to switch agents according to user utterance text - Google Patents

Description

The present invention relates to a technique of an agent dialogue system that interacts with a user using a scenario.

In smartphones and tablet terminals, dialogue systems that naturally interact with users are becoming widespread. In particular, the function of a computer graphics character interacting with a user by voice or text is called an "agent". From the user's point of view, the agent establishes a dialogue according to the user's situation, hobbies, and emotions without giving the user a special consciousness.

Conventionally, there is a technology of a chat dialogue system in which a user can select one agent from a plurality of agents imitating characters such as cartoons (see, for example, Patent Document 1). According to this technique, the dialogue system acquires chat data based on the selected agent and uses the chat data at the time of dialogue to generate a response. During the dialogue, the appearance of the agent is displayed on the smartphone display (chatbot type). However, it is not possible to generate a response to a utterance from a user that is not included in the agent's chat data.

There is also a document that investigates the impression of the user regarding the dialogue system (see, for example, Non-Patent Document 1). According to this document, the user uses a dialogue system with one agent having a character and a dialogue system with three agents having a character. However, it is only an experiment using the Wizard of Oz method and has not been systematized.

Japanese Unexamined Patent Publication No. 2014-98844

Ana Paula Chaves, Marco Aurelio Gerosa, "Single or Multiple Conversational Agents ?: An Interactional Coherence Comparison" ACM CHI Conference on Human Factors in Computing Systems. (2018, April)

According to the above-mentioned Patent Document 1, when a user interacts with one agent as in the chatbot type, if the agent has no response candidate to the user's utterance, the dialogue breaks down. It will be.

Therefore, an object of the present invention is to provide a program, a server, and a method capable of changing agents according to a user's utterance text.

According to the present invention, an agent management program that causes a computer to function to manage a plurality of agents that input a user utterance text and return an agent utterance text.
For each agent, an agent database that associates user assumption text, agent utterance text, and agent persona text,
Similarity calculated by weighting the similarity between the user utterance text of the currently selected agent and the user expected text included in all agents, and the similarity between the user utterance text and the agent persona text of the agent. Degree calculation means and
If the similarity of all user-assumed texts of the currently selected agent is not greater than or equal to a predetermined threshold, the computer should function as an agent selection means to be replaced by another agent having a similarity equal to or greater than a predetermined threshold and containing the highest user-assumed text. It is characterized by.

According to other embodiments in the program of the present invention
It is also preferable to make the computer function to return the agent utterance text corresponding to the user-assumed text having the highest similarity for the other agent when it is replaced by another agent by the agent selection means.

According to other embodiments in the program of the present invention
For each user, further function as a user profile storage means that stores the user profile text,
The similarity calculation means causes the computer to weight the similarity between the user profile text of the user and the agent persona text and / or the agent utterance text of each agent to the similarity of all the user assumed texts of the agent. It is also preferable to make it work.

According to other embodiments in the program of the present invention
It is also preferable that the user profile storage means make the computer function so as to acquire the user profile text of each user by SNS (Social Networking Service).

According to other embodiments in the program of the present invention
The agent persona text contains the presence or absence of honorific expressions and
The similarity calculation means further sets the similarity of all user-assumed texts of the agent to zero when the presence or absence of the honorific expression in the user-spoken text and the presence or absence of the honorific expression in a plurality of agents do not match. It is also preferable to have the computer function.

According to other embodiments in the program of the present invention
It is also preferable that the similarity calculation means converts both the user-spoken text and the user-assumed text into vectors based on the character elements, and causes the computer to function so as to calculate the distance between the two vectors as the cosine similarity.

According to other embodiments in the program of the present invention
The user-spoken text is preferably a text obtained by converting a voice uttered by the user by a voice recognition process, or a text input by the user.

According to other embodiments in the program of the present invention
When the currently selected agent is replaced by another agent, the agent selection means sends an agent utterance text indicating that the currently selected agent is replaced by the other agent, and then the highest degree of similarity among the other agents. It is also preferable to send the agent utterance text corresponding to the user's assumed text with a high value.

According to the present invention, it is a user program that functions a computer mounted on a device that communicates with the agent management program described above.
Displaying the agent character selected by the agent management program on the display, showing the agent utterance text to the user, and making the computer function to send the user utterance text entered or spoken by the user to the agent management program. Is also preferable.

According to the present invention, it is a dialogue server that manages a plurality of agents that return an agent utterance text according to a user utterance text to a terminal based on a user operation.
For each agent, an agent database that associates user assumption text, agent utterance text, and agent persona text,
Similarity calculated by weighting the similarity between the user utterance text of the currently selected agent and the user expected text included in all agents, and the similarity between the user utterance text and the agent persona text of the agent. Degree calculation means and
When the similarity of all user-assumed texts of the currently selected agent is not equal to or higher than a predetermined threshold value, it is characterized by having an agent selection means that substitutes for another agent having a similarity degree equal to or higher than a predetermined threshold value and includes the highest user-assumed text. And.

According to the present invention, it is an agent management method of a device that manages a plurality of agents that return an agent utterance text according to a user utterance text to a terminal based on a user operation.
The device has an agent database in which the user's assumed text, the agent utterance text, and the agent persona text are associated with each agent.
The device is
The similarity between the user utterance text from the user for the currently selected agent and the expected user text included in all agents is calculated by weighting the similarity between the user utterance text and the agent persona text of the agent. Step 1 and
If the similarity of all user-assumed texts of the currently selected agent is not greater than or equal to a predetermined threshold, perform the second step of switching to another agent having a similarity greater than or equal to a predetermined threshold and containing the highest user-assumed text. It is characterized by.

According to the program, server and method of the present invention, the agent can be changed according to the user's utterance text. In particular, for the currently selected agent, when there is no agent utterance text that is a response candidate for the user utterance text, the dialogue content seen by the user can be enhanced by replacing the agent with another agent.

It is a functional block diagram of the dialogue server in this invention. It is a table of each agent registered in the agent database. It is explanatory drawing which shows the calculation of the similarity between texts in this invention. It is explanatory drawing which shows the calculation of the degree of similarity between a user utterance text and a user assumed text. It is explanatory drawing which shows the calculation of the degree of similarity between a user utterance text and an agent persona text. It is explanatory drawing which shows the calculation of the similarity based on the honorific expression between a user and an agent. It is explanatory drawing which shows the calculation of the similarity between a user profile text and an agent persona text.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a functional configuration diagram of the dialogue server in the present invention.

According to FIG. 1, the dialogue server 1 that executes the agent is connected to the terminal 2 that executes the user program via a network.

The terminal 2 is, for example, a smartphone, and has a microphone and a speaker for the user, and a display for displaying the character of the agent.
A user program that serves as an interface for dialogue with the user is implemented in the terminal 2, and the agent character received from the dialogue server 1 is displayed on the display and the agent utterance text is clearly shown to the user. Further, the user utterance text uttered or input by the user is transmitted to the dialogue server 1.
The user-spoken text is a text obtained by converting a voice uttered by the user by a voice recognition process, or a text input by the user. When the user inputs by voice utterance, the microphone input button is displayed on the display of the terminal 2, and when the user inputs the utterance text, the input form is displayed on the display of the terminal 2.

The dialogue server 1 manages a plurality of agents that input the user utterance text and return the agent utterance text. The agent of the dialogue server 1 advances the scenario interactively with the user.

According to FIG. 1, the dialogue server 1 has a dialogue interaction unit 100 and an agent dialogue unit 101 as existing functions.
Further, the dialogue server 1 has an agent database 110, a similarity calculation unit 111, an agent selection unit 112, and a user profile storage unit 113 as functions of the present invention.
These functional components are realized by executing a program that makes the computer mounted on the device function. In addition, the processing flow of these functional components can be understood as an agent selection method.

[Dialogue Interaction Unit 100]
The dialogue interaction unit 100 serves as an interface between the user program of the terminal 2 and the agent of the agent dialogue unit 101. The dialogue interaction unit 100 transmits the character and the agent utterance text in the agent dialogue unit 101 to the terminal 2, receives the user utterance text from the terminal 2, and outputs the user utterance text to the agent dialogue unit 101.

[Agent Dialogue Unit 101]
The agent dialogue unit 101 manages a plurality of agents and activates the agent instructed by the agent selection unit 112.
Then, the agent dialogue unit 101 outputs the user utterance text input from the dialogue interaction unit 100 to the similarity calculation unit 111. Further, the agent dialogue unit 101 outputs the agent utterance text to the dialogue interaction unit 100 based on the agent instructed by the agent selection unit 112.
The agent dialogue unit 101 changes agents in response to an instruction from the agent selection unit 112.

[Agent database 110]
The agent database 110 associates the "user assumed text" with the "agent utterance text" for each agent.

FIG. 2 is a table of each agent registered in the agent database.

According to the agent database 110 of FIG. 2, when the user's assumed text is spoken by the user, the agent responds with the corresponding agent spoken text.
Further, it is also preferable that the agent database 110 further associates the "agent persona text" with each agent. Since the agent is anthropomorphic, it has an agent persona text as its character characteristic.

In FIG. 2, four agents are registered.
For example, Agent 1 has "age: 22 years old", "occupation: student", "hobby: motorcycle", etc. registered as an agent persona text.
When the user utters the user's assumed text "Canceled", the agent 1 returns the agent utterance text "Lucky! This game is interesting" to the user.
Further, when the user utters the user's assumed text "searching for a byte", the agent 1 returns the agent utterance text "Saitama has this job offer" to the user.
In this way, agent 2 "occupation: office worker", agent 3 "gym trainer", and agent 4 "occupation: part-time job" are also registered.

The agent database 110 may be a set of user-assumed texts and agent utterance texts arranged in a tree shape. Depending on the response from the user, it is possible to follow the tree and proceed with the dialogue with the user.

[User profile storage unit 113]
The user profile storage unit 113 stores the user profile text for each user.
The user profile text may be acquired for each user by the SNS (Social Networking Service). For example, the SNS profile may be used as the user profile text.
Further, for example, information on the user's hobbies / preferences and the latest situation may be extracted from the posted text of the SNS and used as the user profile text. For example, a word may be extracted from the user's posted sentence by morphological analysis, and a word characteristic by TF-IDF (Term Frequency --Inverse Document Frequency) may be used as a user profile text. preferable.

[Similarity calculation unit 111]
<Similarity between user-spoken text and user-assumed text>
The similarity calculation unit 111 calculates the similarity between the "user utterance text" from the user for the currently selected agent and the "user assumed text" included in all the agents including the currently selected agent.
The similarity calculation unit 111 converts both the user-spoken text and the user-assumed text into vectors based on the character elements, and calculates the distance between the two vectors as "cosine similarity".

Specifically, the user-spoken text is decomposed into a plurality of words by morphological analysis and converted into a feature vector using Bag Of Words. "Bag-of-Words" refers to a vector that expresses only the frequency of occurrence of each word contained in the text. Here, the order of appearance of words is ignored. This feature vector positions the text at one point in the space with the word as the axis and the frequency of occurrence as the value.
Similarly, each user-assumed text contained in all agents is decomposed into a plurality of words by morphological analysis and converted into feature vectors using Bag Of Words.

Then, the similarity calculation unit 111 calculates the cosine similarity S (io, aj) between the average feature vector of the user utterance text io and the average feature vector of each user assumed text aj included in all the agents. do.
i: user
io: User utterance text identifier (user i's o-th utterance)
a: Agent identifier
aj: User-assumed text identifier For example, it is calculated by the following conceptual formula.
Vio: Characteristic vector of word group in user utterance text io Vaj: Characteristic vector of word group in user assumed text aj of agent a S (io, aj) = cosθ = (Vio · Vaj) / (| Vio || Vaj |)
The cosine similarity S (io, aj) is a value of 0 to 1, and the higher the similarity, the closer to 1.

Here, when the user assumed text whose cosine similarity S exceeds a predetermined threshold value is not detected from the user utterance text, Word2vec is used to calculate the cosine similarity S again. "Word2vec" is a technology that compresses dimensions by vectorizing words in order to capture the meaning and grammar of words.
Specifically, the user-spoken text is decomposed into a plurality of words by morphological analysis and converted into a feature vector using Word2vec. Similarly, each user-assumed text included in all agents is decomposed into a plurality of words by morphological analysis and converted into feature vectors using Word2vec.
Of course, it is not limited to Bag Of Words and Word2vec, and it is sufficient if the part of speech or meaning of each word can be converted into an analyzed feature vector.

[Agent selection unit 112]
If the similarity of all user-assumed texts of the currently selected agent is not equal to or higher than a predetermined threshold value, the agent selection unit 112 is replaced by another agent having a similarity degree equal to or higher than a predetermined threshold value and includes the highest user-assumed text.
Further, when the agent selection unit 112 substitutes for another agent, the agent dialogue unit 101 is instructed to return the agent utterance text corresponding to the user assumed text having the highest similarity for the replaced other agent.

FIG. 3 is an explanatory diagram showing the calculation of the similarity between texts in the present invention.

First, it is assumed that the agent 2 is selected for the user.
(S31) It is assumed that the user utters the user utterance text "I went to the office today, but I'm tired".
At this time, the cosine similarity between the user utterance text and all the user assumed texts registered in the agent 2 is calculated.
User utterance text: "I went to work today, but I'm tired."
(Agent 2)
User assumption text: "Company gets tired" ★ Cosine similarity S is the highest
"There was a presentation announcement"
"There were many meetings."
・ ・ ・ ・ ・ ・
"I haven't been to the gym"
Here, it is assumed that the user's utterance text "I went to the company today but I'm tired" has the highest cosine similarity S with the user's assumed text "The company is tired". Further, it is assumed that the cosine similarity S is equal to or higher than a predetermined threshold value (for example, 0.7).

(S32) Since the highest cosine similarity S is equal to or higher than a predetermined threshold value, the agent 2 is not replaced. Then, in response to the user's assumed text "The company gets tired", the agent 2 instructs the agent dialogue unit 101 with the agent utterance text "I'm patient until the weekend".
That is, if there is at least one user-assumed text having a cosine similarity S equal to or higher than a predetermined threshold value with respect to the user-spoken text among the currently selected agents, the agent is not replaced.

(S33) On the other hand, it is assumed that the user utters the user utterance text "I have to go to the gym and improve my physical strength".
At this time, the cosine similarity between the user utterance text and all the user assumed texts registered in the agent 2 is calculated.
User utterance text: "I have to go to the gym and get fit"
(Agent 2)
User assumption text: "Company gets tired"
"There was a presentation announcement"
"There were many meetings."
・ ・ ・ ・ ・ ・
"I haven't been to the gym" ★ The cosine similarity is the highest. Here, the cosine similarity S of the user's assumed text "I haven't been to the gym" is compared to the user utterance text "I have to go to the gym to improve my physical strength". The highest. However, the cosine similarity S is assumed to be lower than a predetermined threshold value (for example, 0.7).

In this case, the cosine similarity S between the user utterance text and all user assumed texts in agents other than agent 2 is calculated.

FIG. 4 is an explanatory diagram showing the calculation of the degree of similarity between the user utterance text and the user assumed text.

According to FIG. 4, the similarity between the following texts is calculated.
User utterance text: "I have to go to the gym and get fit"
(Agent 1)
User assumption text: "Canceled"
"Looking for a part-time job"
"Report is hard"
・ ・ ・ ・ ・ ・
"It was hard because the train was crowded."
(Agent 3)
User assumption text: "Muscle training is good"
"I go to the gym" ★ Cosine similarity S is the highest
"A lot of overtime"
・ ・ ・ ・ ・ ・
"The train is crowded"
(Agent 4)
User assumption text: "It's a dry season"
"I'm going shopping with my friends"
"I was working part-time until late at night."
・ ・ ・ ・ ・ ・
"It's hard because the train is crowded"
Here, it is assumed that the cosine similarity S of the user assumed text "I am going to the gym" of the agent 3 is the highest with respect to the user utterance text "I have to go to the gym to improve my physical strength". Further, it is assumed that the cosine similarity S is equal to or higher than a predetermined threshold value (for example, 0.7).

(S34) Since the agent 2 does not have a user-assumed text in which the cosine similarity S is equal to or higher than a predetermined threshold value, the agent 2 is switched to the agent 3. Then, in response to the user's assumed text "I'm going to the gym", the agent 3 instructs the agent dialogue unit 101 with the agent utterance text "Is Jim going home from work?".
That is, none of the currently selected agents has a user-assumed text having a cosine similarity equal to or higher than a predetermined threshold for the user-spoken text, and among other agents, the user-spoken text has a cosine similarity equal to or higher than the predetermined threshold. If there is even one user-assumed text that has cosine similarity, it will be replaced by that agent.

As another embodiment, according to FIG. 3, the agent utterance text of S331 is returned after S33.
From the user's point of view, after replying to the agent utterance text "I'm patient until the weekend" from agents 2 to S32, the user utterance text "I have to go to the gym and get fit" in S33, suddenly another agent 3 When asked by the agent's utterance text "Is Jim going home from work?" In S34, there is a sense of discomfort, and it is not a natural flow of dialogue. This is because the agent 2 who has been interacting with until now suddenly disappears, and another agent 3 suddenly speaks.

Therefore, when the currently selected agent is replaced by another agent, the agent selection unit 112 transmits an agent utterance text indicating that the currently selected agent is replaced by the other agent, and then the other agent. Sends the agent utterance text corresponding to the user-assumed text with the highest degree of similarity in.

According to S331 in FIG. 3, the agent 2 before the change tells the user the agent 3 of the gym trainer in advance by saying the agent utterance text "By the way, Mr. Jim trainer said the same thing." You can be aware of it.

According to the above-described embodiment, the similarity calculation unit 111 calculates <similarity S between the user utterance text and the user assumed text>.
On the other hand, as another embodiment, it is also preferable to "weight" the similarity as follows.
<Similarity between user utterance text and agent persona text W ₁ >
<Relationship between the honorific expression of the user utterance text and the honorific expression of the agent utterance text K>
<Similarity between user profile text and agent persona text W ₂ >
<Similarity between user profile text and user assumed text W ₃ >

<Similarity between user utterance text and agent persona text W ₁ >
FIG. 5 is an explanatory diagram showing the calculation of the degree of similarity between the user utterance text and the agent persona text.
_{The similarity calculation unit 111 weights the similarity W 1} (cosine similarity) between the user utterance text io and the agent persona text Pa of each agent a to the similarity S of each user assumed text of the agent.
Pa: Agent persona text of agent a W ₁ (io, Pa)

<Relationship between the honorific expression of the user utterance text and the honorific expression of the agent utterance text K>
FIG. 6 is an explanatory diagram showing the calculation of the degree of similarity between the user and the agent based on the honorific expression.
As another embodiment, it is also preferred that the agent persona text includes the presence or absence of honorific expressions.

The similarity calculation unit 111 may determine whether or not the user utterance text is a honorific expression by using a predicate obtained by morphological analysis of the user utterance text, or is described in the user profile. It may be.
Similarly, in order to determine whether or not the agent utterance text is an honorific expression, the agent utterance text may be determined by a morphologically analyzed predicate, or it may be described in the agent persona text. May be good.
Then, the similarity calculation unit 111 determines whether or not the honorific expression (yes / no) of the user utterance text and the honorific expression (yes / no) of the agent utterance text of each agent match. Here, if there is an honorific expression in the user utterance text and there is no honorific expression in the agent utterance text, that agent is not selected. In that case, it is preferable that the cosine similarity of all user-assumed texts in the agent is zero. While users speak in honorifics, agents speaking in plain language is not a natural dialogue.
i: user
a: Agent K (i, a) = 0, 0 <K (i, a) ≤ 1
If the honorific expressions of the user utterance text and the agent utterance text match, any number greater than 0.00 and less than or equal to 1.00 is used for K. On the other hand, if they do not match, K = 0.

It may also compare the age of the user profile text with the age of the agent profile text.
If the user's age is older than the agent's age, the agent utterance text is preferably honorific. In that case, the cosine similarity of all user-assumed texts in such agents is set to zero so that agents without honorific expressions are not selected. This is because it is common to use honorifics when interacting with older people.

<Similarity between user profile text and agent persona text W ₂ >
FIG. 7 is an explanatory diagram showing the calculation of the degree of similarity between the user profile text and the agent persona text.
_{The similarity calculation unit 111 sets the similarity W 2} (cosine similarity) between the user profile text u of the user and the agent persona text Pa of each agent a to the similarity S of all user assumed texts of the agent. Weight.
u: User profile text
Pa: Agent persona text of agent a W ₂ (u, Pa)
The greater the similarity between the user profile text and the agent persona text, the more familiar the user is to the agent who is closer to his or her hobbies and tastes. In the case of FIG. 3 described above, the agent 2 having a similar age and occupation is selected for the user.

<Similarity between user profile text and user assumed text W ₃ >
_{The similarity calculation unit 111 sets the similarity W 3} (cosine similarity) between the user profile text u of the user and the user assumed text aj of each agent a to the similarity S of all the user assumed texts of the agent. Weight.
u: User profile text
aj: User assumption text of agent a j
W ₃ (u, aj)

Finally, the total similarity Sall obtained by weighting the similarity S (io, aj) of the user assumed text with the other similarity W is calculated as follows.
Here, there are cases where the agent is changed and cases where the initial agent is selected.
<In case of agent change>
Sall (io, aj) = S (io, aj) x W ₁ (io, Pa) x K (i, a)
When K = 0 with or without honorifics, Sall (io, aj) = 0.
Then, the user assumed text having the highest overall similarity Sall is detected, and the "agent" for registering the user assumed text is selected. Further, the agent utterance text corresponding to the user assumed text is instructed to the agent dialogue unit 101.
In addition, Sall (io, aj) = S (io, aj) × W ₁ (io, aj) × K × W ₂ (io, aj) × W ₃ (io, aj) may be set. However, it is preferable that the contribution rate of the weight of _{W 2} (io, aj) × W _{3 (io, aj) is low.} It is based on the reason that the agent change decision should not be pulled by the user profile.

<When selecting the initial agent>
Sall (io, aj) = W ₂ (u, Pa) x W ₃ (u, aj)
Since there is no user utterance text at the start of the dialogue, it is preferable to select the agent by comparing the user profile.

As described in detail above, according to the program, server and method of the present invention, the agent can be changed according to the user's utterance text. In particular, for the currently selected agent, when there is no agent utterance text that is a response candidate for the user utterance text, the dialogue content seen by the user can be enhanced by replacing the agent with another agent.

With respect to the various embodiments of the present invention described above, various changes, modifications and omissions within the scope of the technical idea and viewpoint of the present invention can be easily made by those skilled in the art. The above explanation is just an example and does not attempt to restrict anything. The present invention is limited only to the scope of claims and their equivalents.

1 Dialogue server 100 Dialogue interaction unit 101 Agent dialogue unit 110 Agent database 111 Similarity calculation unit 112 Agent selection unit 113 User profile storage unit 2 Terminals

Claims (11)

An agent management program that allows a computer to function to manage multiple agents that enter user utterance text and respond to agent utterance text.
For each agent, an agent database that associates user assumption text, agent utterance text, and agent persona text,
Similarity calculated by weighting the similarity between the user utterance text of the currently selected agent and the user expected text included in all agents, and the similarity between the user utterance text and the agent persona text of the agent. Degree calculation means and
If the similarity of all user-assumed texts of the currently selected agent is not equal to or higher than a predetermined threshold, the computer should function as an agent selection means to be replaced by another agent having a similarity equal to or higher than a predetermined threshold and containing the highest user-assumed text. An agent management program that features. The claim is characterized in that when the agent is replaced by another agent by the agent selection means, the computer functions to return the agent utterance text corresponding to the user assumed text having the highest similarity for the other agent. The agent management program according to 1. For each user, further function as a user profile storage means that stores the user profile text,
The similarity calculation means is a computer so as to weight the similarity between the user profile text of the user and the agent persona text and / or the agent utterance text of each agent to the similarity of all the user assumed texts of the agent. The agent management program according to claim 1 or 2 , wherein the agent management program is made to function. The agent management program according to claim 3 , wherein the user profile storage means causes a computer to function so as to acquire a user profile text of each user by an SNS (Social Networking Service). The agent persona text includes the presence or absence of honorific expressions.
Further, when the presence / absence of the honorific expression in the user-spoken text and the presence / absence of the honorific expression in the plurality of agents do not match, the similarity calculation means sets the similarity of all the user-assumed texts of the agent to zero. The agent management program according to any one of claims 1 to 4 , wherein the computer functions as described above. The similarity calculation means is characterized in that both the user-spoken text and the user-assumed text are converted into vectors based on character elements, and the computer functions to calculate the distance between the two vectors as the cosine similarity. The agent management program according to any one of claims 1 to 5. The user-spoken text is any one of claims 1 to 6 , wherein the user-spoken text is a text obtained by converting a voice uttered by the user by a voice recognition process, or is a text input by the user. The listed agent management program. When the currently selected agent is replaced by another agent, the agent selection means sends an agent utterance text indicating that the currently selected agent is replaced by the other agent, and then the most similar in the other agent. The agent management program according to any one of claims 1 to 7 , wherein an agent utterance text corresponding to a user's assumed text with a high degree is transmitted. A user program that functions a computer mounted on a device that communicates with the agent management program according to any one of claims 1 to 8.
The computer is displayed so that the character of the agent selected by the agent management program is displayed on the display, the agent utterance text is clearly shown to the user, and the user utterance text input or spoken by the user is transmitted to the agent management program. A user program characterized by making it work. A dialogue server that manages multiple agents that respond to agent utterance texts in response to user utterance texts to terminals based on user operations.
For each agent, an agent database that associates user assumption text, agent utterance text, and agent persona text,
Similarity calculated by weighting the similarity between the user utterance text of the currently selected agent and the user expected text included in all agents, and the similarity between the user utterance text and the agent persona text of the agent. Degree calculation means and
When the similarity of all user-assumed texts of the currently selected agent is not equal to or higher than a predetermined threshold value, it is characterized by having an agent selection means that substitutes for another agent having a similarity degree equal to or higher than a predetermined threshold value and includes the highest user-assumed text. Dialogue server. It is an agent management method of a device that manages a plurality of agents that return an agent utterance text according to a user utterance text to a terminal based on a user operation.
The device includes an agent database in which a user assumption text, an agent utterance text, and an agent persona text are associated with each agent.
The device is
The similarity between the user utterance text from the user for the currently selected agent and the expected user text included in all agents is calculated by weighting the similarity between the user utterance text and the agent persona text of the agent. Step 1 and
If the similarity of all user-assumed texts of the currently selected agent is not greater than or equal to a predetermined threshold, perform the second step of switching to another agent having a similarity greater than or equal to a predetermined threshold and containing the highest user-assumed text. An agent management method characterized by.

Images