KR20240002766A - Apparatus and method for supporting memory-based conversations - Google Patents

Abstract

A memory-based conversation support device and method are disclosed to support users in creating conversations more suitable for users by utilizing the meaning inherent in accumulated memory elements. A memory-based conversation support device according to an embodiment manages memory information including speech information received from a user, response information corresponding to speech information received from an interactive system server, and analysis information obtained by analyzing the user's speech information. a storage management unit that stores; a conversation management unit that generates conversation inducing information to induce conversation in the user before the user speaks, based on the stored memory information; and a statistical analysis unit that performs statistical analysis on the user speech information based on the stored memory information.

Description

Apparatus and method for supporting memory-based conversations}

The present invention relates to a memory-based conversation support device and method to support users in creating more suitable conversations by utilizing the meaning inherent in accumulated memory elements.

Application systems based on natural language conversations, such as chatbots, can be considered one of the representative research areas of Natural Language Processing (NLP) technology. In particular, in recent years, research on Natural Language Understanding (NLU) technology has been growing to not only apply NLP to analyze conversational text, but also understand the context of the conversation to enable more natural communication with users. .

The purpose is to provide a memory-based conversation support device and method to help users create more suitable conversations by utilizing the meaning inherent in accumulated memory elements.

According to one aspect, a memory-based conversation support device manages and stores memory information including speech information received from a user, response information corresponding to speech information received from an interactive system server, and analysis information obtained by analyzing the user's speech information. a storage management unit that stores; a conversation management unit that generates conversation inducing information to induce conversation in the user before the user speaks, based on the stored memory information; and a statistical analysis unit that performs statistical analysis on the user speech information based on the stored memory information.

The memory information includes a conversation history schema for the user's utterance sentences and response sentences and meta information of the interactive system server, a history-domain schema for the ratio distribution of conversation topic domains extracted from conversation sentences, and At least one schema among the history-emotion schema for the ratio distribution of emotions and emotion labels, the history-harmfulness schema for the ratio distribution of harmfulness and harmfulness labels extracted from conversation sentences, and the history-keyword schema for keywords extracted from sentences. It can be classified, stored and managed.

The conversation management unit may determine whether to generate conversation guidance information based on at least one of a designated time, a predetermined time elapsed since the last conversation, and a time extracted from at least one of utterance information and response information conversed within a predetermined period.

The conversation management unit may generate conversation guidance information based on at least one of interest information, emotion information, and keyword information extracted from at least one of utterance information and response information that occurred within a predetermined period of time.

The statistical analysis unit may perform statistical analysis on at least one of the average number of utterances per day/week/month, number of words used, conversation time, domain of interest, and keywords based on the stored memory information.

According to one aspect, a memory-based conversation support method manages memory information including speech information received from a user, response information corresponding to speech information received from an interactive system server, and analysis information obtained by analyzing the user's speech information. A storage management step for storing; A conversation management step of generating conversation inducing information to encourage the user to have a conversation before the user speaks, based on the stored memory information; and a statistical analysis step of performing statistical analysis on the user speech information based on the stored memory information.

The conversation management step may determine whether to generate conversation guidance information based on at least one of a designated time, a predetermined time elapsed since the last conversation, and a time extracted from at least one of utterance information and response information conversed within a predetermined period.

The conversation management step may generate conversation guidance information based on at least one of interest information, emotion information, and keyword information extracted from at least one of utterance information and response information that occurred within a predetermined period of time.

The statistical analysis step may perform statistical analysis on at least one of the average number of utterances per day/week/month, number of words used, conversation time, domain of interest, and keywords based on the stored memory information.

By utilizing the meaning inherent in continuously accumulated memory elements, the system or language model can support the user to provide more suitable dialogue creation results.

Figure 1 is an example diagram to explain an environment in which a memory-based conversation support device operates according to an embodiment.
Figure 2 is a configuration diagram of a memory-based conversation support device according to an embodiment.
Figure 3 is an example diagram for explaining a schema for memory information according to an example.
Figure 4 is a flowchart showing a memory-based conversation support method according to one embodiment.

Hereinafter, an embodiment of the present invention will be described in detail with reference to the attached drawings. In describing the present invention, if it is determined that a detailed description of a related known function or configuration may unnecessarily obscure the gist of the present invention, the detailed description will be omitted. In addition, the terms described below are terms defined in consideration of functions in the present invention, and may vary depending on the intention or custom of the user or operator. Therefore, the definition should be made based on the content throughout this specification.

Hereinafter, embodiments of a memory-based conversation support device and method will be described in detail with reference to the drawings.

Figure 1 is an example diagram illustrating an environment in which a memory-based conversation support device operates according to an embodiment.

According to one embodiment, the memory-based conversation support device 100 includes a user terminal 10 capable of receiving conversation information from a user and outputting response information, and an interactive system server capable of generating a response according to the user's conversation. It can be connected to (20).

According to one embodiment, the memory-based conversation support device 100 is designed as a memory element with the history of various information such as conversation sentences, vocabulary, and emotions communicated with the user in an interactive system, and is embedded in the continuously accumulated memory elements. By utilizing the meaning provided, the system or language model can support the user to provide more suitable dialogue generation results. In addition, the memory-based conversation support device 100 can identify the user's system usage pattern or major conversation tendencies by statistically analyzing memory elements and user history information at a certain period, and the system or language model is generated at an appropriate time. First, you can actively encourage the user to speak.

According to one example, the interface of the user terminal 10 for conducting a conversation with the user may be developed with the Unity engine, and the interactive system server 20 is a chatbot server based on a language model such as Rasa or GPT-3. It can be configured.

According to one example, the memory-based conversation support device 100 may be located at an intermediate stage between the Unity interface and the chatbot server, stores user utterance sentences input from the interface as memory information, and then transfers them to the chatbot server. The generated response sentences and various information (e.g., emotions, domains, etc.) can also be stored and accumulated as a history of memory information. Thereafter, the memory-based conversation support device 100 may update logging results, such as system utilization patterns, by statistically analyzing elements of the memory information.

According to one embodiment, the memory-based conversation support device 100 may be a system that operates through interaction between a user terminal 10 and an interactive system server 20 connected through network communication. As an example, the user terminal 10 may be an application that operates based on various environments such as a desktop PC, laptop, tablet, or mobile, and receives the user's desired sentence through a microphone or text input device and receives the interactive system server ( 20). The interactive system server 20 can generate an appropriate response corresponding to the input user utterance by introducing various NLP/NLU technologies such as morphological analysis, syntactic analysis, emotional analysis, and intention recognition. At this time, the memory-based conversation support device 100 can systematically store and manage the information acquired during the conversation in a memory information DB (not shown) as a memory element.

Figure 2 is a configuration diagram of a memory-based conversation support device according to an embodiment.

According to one embodiment, the memory-based conversation support device 100 may include a storage management unit 110, a conversation management unit 120, and a statistical analysis unit 130.

According to one embodiment, the storage management unit 110 manages memory information including speech information received from the user, response information corresponding to speech information received from the interactive system server, and analysis information obtained by analyzing the user's speech information. and can be saved. As an example, the storage management unit 110 may store and manage analysis results such as user utterances, response sentences of conversation models, and domains/emotions.

According to one embodiment, the memory information includes a conversation history schema for the user's utterance sentences and response sentences and meta information of the interactive system server, and a history-domain schema for the ratio distribution of conversation topic domains extracted from conversation sentences. , the history-emotion schema for the percentage distribution of emotions and emotion labels extracted from conversation sentences, the history-harmfulness schema for the percentage distribution of harmfulness and harmfulness labels extracted from conversation sentences, and the history-keyword schema for keywords extracted from sentences. It can be classified into at least one schema and stored and managed.

Referring to Figure 3, schemas specifying memory elements can be divided into five types.

As an example, the Chat History schema can accumulate related meta information, including the user's utterance sentences and the response sentences of the conversation model.

As an example, the History-Domain schema can accumulate the ratio distribution of conversation topic domains extracted from conversation sentences. For example, the history-domain schema can utilize eight domains: science & technology, politics, movies, sports, music, books, fashion, and general entertainment.

As an example, the History-Emotion schema can accumulate the ratio distribution of emotions extracted from conversation sentences and the finally classified emotion labels. For example, the history-emotion schema may include the six emotions joy, anger, surprise, sadness, disgust, and fear, as well as neutral.

As an example, the History-Toxicity schema can accumulate the ratio distribution of toxicity and the final label extracted from conversation sentences. For example, a history-harm schema can utilize six hazards: identity_attack, insult, obscene, severe_toxicity, sexual_explicit, and threat, and two final labels: safe or unsafe.

As an example, the History-Keyword schema can store information about the user's interests by accumulating keywords extracted from sentences. For example, a history-keyword schema can utilize word entities and types corresponding to keywords.

According to one embodiment, the conversation management unit 120 may generate conversation guidance information to induce a user's conversation before the user speaks, based on stored memory information.

For example, the conversation management unit 120 may supplement the response sentences of the conversation model based on the memory information obtained in the previous step to provide an answer that a specific user may prefer. Additionally, the conversation management unit 120 can determine whether to actively induce speech by identifying user patterns or conversation tendencies.

According to one embodiment, the conversation management unit 120 generates conversation guidance information based on at least one of a designated time, a predetermined time elapsed since the last conversation, and a time extracted from at least one of utterance information and response information conversed within a predetermined period. You can decide whether to do it or not.

For example, the conversation management unit 120 presents an appropriate welcome greeting according to the morning/afternoon/evening time zone in consideration of the time zone in which the conversation is performed, or preemptively greets the user based on the user's arrival/departure from school or commuting/leaving time. You can decide to start a conversation. Additionally, the conversation manager 120 may determine to start active speech when a predetermined time has elapsed since the user last performed a conversation.

For example, the conversation management unit 120 may extract schedule information for a predetermined time from the user's conversation and determine a conversation start point using the extracted time information. For example, if there is a visit to a specific place or an appointment with a specific person at a predetermined time during the conversation, the conversation management unit 120 may decide to start the related conversation after a predetermined time has elapsed from that point or the point of termination.

According to one embodiment, the conversation management unit 120 may generate conversation guidance information based on at least one of interest information, emotion information, and keyword information extracted from at least one of utterance information and response information conversed within a predetermined period of time. .

For example, the conversation management unit 120 may search memory information and present questions about the user's recent domain of interest, or create a conversation by referring to frequently mentioned keywords. Additionally, the conversation manager 120 may attempt a conversation by simply creating a question asking about the user's current situation or opinion.

As an example, the conversation manager 120 may create a conversation using emotional information generated in the previous conversation. For example, when a predetermined type of emotion is extracted from the user's previous conversation, the conversation management unit 120 may create a conversation about whether the user's emotion has been resolved.

According to one embodiment, the statistical analysis unit 130 may perform statistical analysis on user speech information based on stored memory information.

According to one example, the statistical analysis unit 130 may perform statistical analysis on at least one of the average number of utterances per day/week/month, number of words used, conversation time, domain of interest, and keywords based on stored memory information. .

As an example, the statistical analysis unit 130 uses the memory information accumulated in the system to analyze the user's learning tendency and statistical information regarding the average number of utterances per day/week/month, number of words used, conversation time, domain of interest, etc. You can. The statistical analysis unit 130 can provide the analyzed results to users or system administrators using a visualization management tool.

For example, the statistical analysis unit 130 may provide users or administrators with intuitive information such as system utilization patterns by transmitting visualized information along with statistical elements. For example, the statistical analysis unit 130 provides a complete list of conversation history, topic/domain distribution and harmfulness distribution for each conversation, statistical information such as the average number of words used, overall number of utterances, service utilization trends, and frequently used words. Information such as word sets can be provided.

Figure 4 is a flowchart showing a memory-based conversation support method according to one embodiment.

According to one embodiment, the memory-based conversation device manages and stores memory information including speech information received from the user, response information corresponding to speech information received from the interactive system server, and analysis information obtained by analyzing the user's speech information. It can be saved (410).

According to one embodiment, the memory-based conversation device may generate conversation guidance information to induce the user's conversation before the user speaks based on the stored memory information (420). As an example, the memory-based conversation device determines whether to generate conversation guidance information based on at least one of a specified time, a predetermined time elapsed since the last conversation, and a time extracted from at least one of utterance information and response information conversationed within a predetermined period of time. You can. Additionally, the memory-based conversation device may generate conversation guidance information based on at least one of interest information, emotion information, and keyword information extracted from at least one of utterance information and response information that occurred within a predetermined period of time.

According to one embodiment, the memory-based conversation device may perform statistical analysis on user speech information based on stored memory information (430). As an example, a memory-based conversation support device may perform statistical analysis on at least one of the average number of utterances per day/week/month, number of words used, conversation time, domain of interest, and keywords based on stored memory information.

In the description of the embodiment of FIG. 4, descriptions that overlap with those described with reference to FIGS. 1 to 3 will be omitted.

An aspect of the present invention may be implemented as computer-readable code on a computer-readable recording medium. Codes and code segments implementing the above program can be easily deduced by a computer programmer in the art. Computer-readable recording media may include all types of recording devices that store data that can be read by a computer system. Examples of computer-readable recording media may include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical disk, etc. Additionally, the computer-readable recording medium may be distributed over a computer system connected to a network, and may be written and executed as computer-readable code in a distributed manner.

So far, the present invention has been examined focusing on its preferred embodiments. A person skilled in the art to which the present invention pertains will understand that the present invention may be implemented in a modified form without departing from the essential characteristics of the present invention. Accordingly, the scope of the present invention is not limited to the above-described embodiments, but should be construed to include various embodiments within the scope equivalent to the content described in the patent claims.

10: User terminal
20: Interactive System Server
100: Memory-based conversation support device
110: storage management unit
120: Dialog management unit
130: Statistical analysis department

Claims (10)

a storage management unit that manages and stores memory information including speech information received from the user, response information corresponding to the speech information received from the interactive system server, and analysis information obtained by analyzing the user's speech information;
a conversation management unit that generates conversation inducing information to induce conversation in the user before the user speaks, based on the stored memory information; and
A memory-based conversation support device comprising a statistical analysis unit that performs statistical analysis on user speech information based on stored memory information.
According to claim 1,
The memory information is
Conversation history schema for the user's utterance sentences and response sentences and meta information of the interactive system server, history-domain schema for the ratio distribution of conversation topic domains extracted from conversation sentences, and proportion of emotions extracted from conversation sentences Classified and stored in at least one of the following schemas: the history-emotion schema for distribution and emotion labels, the history-harmfulness schema for the ratio distribution of harmfulness and harmfulness labels extracted from conversation sentences, and the history-keyword schema for keywords extracted from sentences. and managed, memory-based conversation assistants.
According to claim 1,
The conversation management department
A memory-based conversation support device that determines whether to generate conversation guidance information based on at least one of a specified time, a predetermined time elapsed since the last conversation, and a time extracted from at least one of utterance information and response information conversed within a predetermined period.
According to claim 1,
The conversation management department
A memory-based conversation support device that generates conversation guidance information based on at least one of interest information, emotion information, and keyword information extracted from at least one of utterance information and response information conversational within a predetermined period.
According to claim 1,
The statistical analysis department
A memory-based conversation support device that performs statistical analysis on at least one of the average number of utterances per day/week/month, number of words used, conversation time, domain of interest, and keywords based on stored memory information.
A storage management step of managing and storing memory information including speech information received from the user, response information corresponding to the speech information received from the interactive system server, and analysis information obtained by analyzing the user's speech information;
A conversation management step of generating conversation inducing information to encourage the user to have a conversation before the user speaks, based on the stored memory information; and
A memory-based conversation support method comprising a statistical analysis step of performing statistical analysis on user speech information based on stored memory information.
According to claim 6,
The memory information is
Conversation history schema for the user's utterance sentences and response sentences and meta information of the interactive system server, history-domain schema for the ratio distribution of conversation topic domains extracted from conversation sentences, and proportion of emotions extracted from conversation sentences Classified and stored in at least one of the following schemas: the history-emotion schema for distribution and emotion labels, the history-harmfulness schema for the ratio distribution of harmfulness and harmfulness labels extracted from conversation sentences, and the history-keyword schema for keywords extracted from sentences. and a managed, memory-based conversation support method.
According to claim 6,
The conversation management step is
A memory-based conversation support method that determines whether to generate conversation guidance information based on at least one of a specified time, a predetermined time elapsed since the last conversation, and a time extracted from at least one of utterance information and response information conversed within a predetermined period.
According to claim 6,
The conversation management step is
A memory-based conversation support method that generates conversation guidance information based on at least one of interest information, emotion information, and keyword information extracted from at least one of utterance information and response information conversed within a predetermined period.
According to claim 6,
The statistical analysis step is
A memory-based conversation support method that performs statistical analysis on at least one of the average number of utterances per day/week/month, number of words used, conversation time, domain of interest, and keywords based on stored memory information.