Apparatus and method for providing chat service based on chatbot

Abstract

An apparatus for providing a chat service based on a chatbot according to one embodiment of the present invention comprises: an utterance sentence generation unit which recognizes a user voice and generates a user utterance sentence; an utterance sentence analysis unit which analyzes the user utterance sentence and extracts at least one of an utterance intention or an entity name of a user; an answer sentence generation unit which determines whether the user utterance sentence is a sentence based on a pre-stored chat scenario on the basis of at least one of the utterance intention or the entity name of the user, and generates an answer sentence corresponding to the user utterance sentence; and an answer sentence output unit which converts the answer sentence into a chatbot voice and outputs the chatbot voice, wherein, when the user utterance sentence is not a sentence based on a pre-stored chat scenario, the answer sentence generation unit can acquire the answer sentence by correcting the user utterance sentence and then inputting the corrected user utterance sentence into a pre-trained language model.

Description

Device and method for providing chatbot-based conversation service

[0001] The present invention relates to a device and method for providing a chatbot-based conversation service.

[0002] Chatbot is a combination of the words chat and robot, and can refer to software designed to converse with users through text or voice and perform specific tasks.

[0003] These chatbots are called by various names such as chatterbot, talkbot, intelligent agent, and interactive agent, and recently, with the advancement of machine learning technologies such as deep learning, chatbots are being utilized in various industries.

[0004] Meanwhile, as an example of conventional technology, a chatbot has been introduced that recognizes the speaker's voice, selects sentences to provide to the speaker based on the recognition, and provides the selected sentences to the speaker in voice or text.

[0005] However, chatbots based on conventional technology have the disadvantage of not being able to cover the wide range of conversation topics exchanged between chatbots and users, as they conduct conversations based on response rules defined by the user.

[0006] Additionally, the recently widely used artificial intelligence (AI)-based chatbots still have the problem of not being able to maintain the desired conversation flow because there is little room for user intervention.

[0007] The present invention has been conceived in the technical background described above, and aims to provide a chatbot-based conversation service providing device and method that can provide an appropriate answer desired by a user by utilizing the user's speech intention and major keywords used in the user's speech sentences.

[0008] The technical problems of the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned can be clearly understood by a person having ordinary skill in the technical field to which the present invention belongs from the description below.

[0009] According to an embodiment of the present invention for solving the above-described technical problem, a chatbot-based conversation service providing device includes a speech sentence generation unit that recognizes a user's voice to generate a user-spoken sentence, a speech sentence analysis unit that analyzes the user-spoken sentence to extract at least one of the user's speech intention and an entity name, a response sentence generation unit that determines whether the user-spoken sentence is a sentence based on a pre-stored conversation scenario based on at least one of the user's speech intention and the entity name, and generates a response sentence corresponding to the user-spoken sentence, and a response sentence output unit that converts the response sentence into a chatbot voice and outputs it, wherein, if the user-spoken sentence is not a sentence based on a pre-stored conversation scenario, the response sentence generation unit can modify the user-spoken sentence and input it into a pre-trained language model to obtain the response sentence.

[0010] In addition, a method for providing a conversation service performed by a chatbot-based conversation service providing device according to another embodiment of the present invention for solving the above technical problem includes an operation of training a language model based on a plurality of sentences for each conversation topic stored in a database and a response sentence corresponding to each sentence, an operation of recognizing a user's voice to generate a user-spoken sentence, an operation of analyzing the user-spoken sentence to extract at least one of the user's speech intent and a named entity, an operation of determining whether the user-spoken sentence is a sentence based on a pre-stored conversation scenario based on at least one of the user's speech intent and the named entity, an operation of generating a response sentence corresponding to the user-spoken sentence, and an operation of converting the response sentence into a chatbot voice and outputting it, wherein the operation of generating the response sentence may include an operation of modifying the user-spoken sentence if the user-spoken sentence is not a sentence based on a pre-stored conversation scenario, and an operation of inputting the modified user-spoken sentence into the trained language model to generate the response sentence.

[0011] The above-described problem-solving methods are merely exemplary and should not be construed as limiting the present invention. In addition to the exemplary embodiments described above, additional embodiments may exist, as described in the drawings and detailed description of the invention.

[0012] According to a device and method for providing a chatbot-based conversation service according to one embodiment of the present invention, there is an effect of being able to provide an appropriate answer desired by the user by utilizing the user's speech intention and the main keywords used in the user's speech sentence.

[0013] In addition, according to the present invention, the quality of the conversation between the user and the chatbot can be improved by analyzing the conversation flow between the user and the chatbot to more accurately determine the answer the user needs.

[0014] The effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by a person having ordinary skill in the art to which the present invention pertains from the description below.

[0015] FIG. 1 is a functional block diagram of a chatbot-based conversation service providing device according to one embodiment of the present invention.

[0016] FIG. 2 illustrates a flowchart for explaining a conversation service providing method performed by a chatbot-based conversation service providing device according to one embodiment of the present invention.

[0017] FIGS. 3 to 8 are diagrams illustrating an example for explaining a process of providing a conversation service to a user through a chatbot according to one embodiment of the present invention.

[0018] Hereinafter, embodiments of the present invention will be described with reference to the attached drawings. When adding reference numerals to components in each drawing, it should be noted that identical components are given the same numerals as much as possible even if they are shown in different drawings. In addition, when describing embodiments of the present invention, if a detailed description of a related known structure or function is judged to hinder understanding of the embodiments of the present invention, a detailed description thereof will be omitted. In addition, although embodiments of the present invention will be described below, the technical idea of ​​the present invention is not limited thereto and may be modified and implemented in various ways by those skilled in the art.

[0019] Throughout the specification, when a part is said to be "connected" to another part, this includes not only cases where it is "directly connected" but also cases where it is "indirectly connected" with another element in between. Throughout the specification, when a part is said to "include" a certain component, this does not mean that other components are excluded, but rather that other components may be included, unless specifically stated otherwise. In addition, when describing components of embodiments of the present invention, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only intended to distinguish the components from other components, and the nature, order, or sequence of the components are not limited by the terms.

[0020] The conversation service providing device (100) described in this specification below can be implemented as a chatbot that analyzes a user's spoken sentences and generates and outputs an answer (or response) corresponding to the user's spoken sentences.

[0021] This chatbot-based conversation service providing device (100) is an interactive system and may include a computer program that enables interaction with a user as if having a conversation through voice.

[0022] According to various embodiments, a chatbot-based conversation service providing device (100) may be a device that recognizes a user's spoken voice, extracts a specific spoken intent or a specific entity embedded in the user's spoken sentence, analyzes it, and provides the user with an appropriate answer (or response) corresponding to the user's spoken sentence.

[0023] For example, a chatbot-based conversation service providing device (100) can extract at least one of the user's speech intention and entity name included in the user's speech sentence by performing natural language understanding (NLU) on the result of recognizing the user's speech voice, and provide an answer (or response) desired by the user based on this.

[0024] For example, a chatbot-based conversation service providing device (100) may provide a user with a predefined conversation scenario-based answer based on a specific extracted speech intent and / or a specific entity name, or may provide an answer generated by an LLM (Large Language Model), which is an artificial intelligence model trained based on a plurality of texts, through a chatbot voice.

[0025] Below, the components and operation of the chatbot-based conversation service providing device (100) will be described in more detail.

[0026] FIG. 1 illustrates a functional block diagram of a chatbot-based conversation service providing device according to an embodiment of the present invention, and FIG. 2 illustrates a flowchart for explaining a conversation service providing method performed by a chatbot-based conversation service providing device according to an embodiment of the present invention.

[0027] The operations in FIG. 2 below may be performed sequentially, but are not necessarily performed sequentially. For example, the order of the operations may be changed, and at least two operations may be performed in parallel.

[0028] Referring to FIGS. 1 and 2, a chatbot-based conversation service providing device (100) may include a spoken sentence generation unit (110), a spoken sentence analysis unit (120), a response sentence generation unit (130), a response sentence output unit (140), a database (150), and a processor (160).

[0029] The spoken sentence generation unit (110) can recognize the user's voice and generate a user-spoken sentence (S210). For example, the spoken sentence generation unit (110) can generate a user-spoken sentence using STT (Speech-to-Text) technology, which allows a computer to interpret the spoken language of the user and convert the content into text.

[0030] The speech sentence analysis unit (120) can analyze the generated user speech sentence to extract at least one of the user's speech intent and entity (S220). For example, the speech sentence analysis unit (120) can analyze phrases, words, etc. included in the user's speech sentence to extract the user's specific speech intent and a specific entity, which is a major keyword included in the user's speech sentence.

[0031] According to various embodiments, the speech sentence analysis unit (120) can extract a user's specific speech intention and / or a specific entity name from a user's speech sentence by using a preset rule or a pre-learned speech intention extraction model.

[0032] For example, the spoken sentence analysis unit (120) can extract specific entities from a user-spoken sentence by predefining rules for finding main keywords in the user-spoken sentence. As an example, the spoken sentence analysis unit (120) can extract specific entities corresponding to time and place, respectively, through rules for finding main keywords such as “tomorrow 3 p.m.” and “Gangnam” in the user-spoken sentence “Let’s meet in Gangnam at 3 p.m. tomorrow.” In this case, the rules for finding main keywords can be predefined by rules input by the user, etc.

[0033] In addition, for example, the speech sentence analysis unit (120) can pre-train an speech intent extraction model by utilizing a large amount of data for classifying the user's speech intent and then identify a specific speech intent from the user's speech sentence. As an example, the speech sentence analysis unit (120) can collect at least one sentence containing "movie reservation" to classify the user's speech intent of "movie reservation" and utilize this to train the speech intent extraction model. At least one sentence containing "movie reservation", that is, training data for training the speech intent extraction model, may be, for example, sentences containing "movie reservation", "movie reservation please", "movie reservation in progress", etc. In this case, the speech intent extraction model for extracting the speech intent inherent in the user's speech sentence may use a Bert model including deep learning in natural language processing, a Recurrent Neural Network (RNN) model, or a Convolutional Neural Network (CNN) model.

[0034] The response sentence generation unit (130) can determine whether a user-spoken sentence is based on a pre-stored conversation scenario based on the user's specific speech intent and / or specific entity name (S230). Specifically, the response sentence generation unit (130) can determine whether a user-spoken sentence is a sentence capable of generating a response sentence to be provided to the user based on a pre-determined conversation scenario.

[0035] If the response sentence generation unit (130) determines that the user-spoken sentence is a sentence based on a pre-stored conversation scenario, it can determine again whether the user-spoken sentence is a single conversation sentence or a continuous conversation sentence (S241).

[0036] For example, the response sentence generation unit (130) can determine, based on a specific utterance intent and / or a specific entity name extracted from the user utterance, whether the user utterance is a single conversation sentence that ends with a chatbot's response and service provision, or a continuous conversation sentence that ends through multiple interactions between the user and the chatbot. The criteria for determining whether the user utterance is a single conversation sentence or a continuous conversation sentence can be classified by conversation scenarios pre-stored in the database (150).

[0037] In operation S242, if the response sentence generation unit (130) determines that the user-uttered sentence is a single conversation sentence, it may generate a response sentence (or a single conversation response sentence) corresponding to the user-uttered sentence based on a pre-stored conversation scenario. In this case, the chatbot may also provide the user with a service corresponding to the user-uttered sentence.

[0038] In operation S243, if the response sentence generation unit (130) determines that the user-spoken sentence is a continuous conversation sentence, it may generate a response sentence corresponding to the user-spoken sentence until a specific condition is satisfied. The specific condition may include, for example, a case where the user does not speak for a certain period of time and thus the user-spoken sentence is not recognized, or all response sentences corresponding to the user-spoken sentence have been provided, or all services corresponding to the user-spoken sentence have been provided to the user.

[0039] In one embodiment, if the response sentence generation unit (130) determines that the user-spoken sentence is not a sentence based on a pre-stored conversation scenario, the response sentence generation unit (130) may modify the user-spoken sentence and input the modified user-spoken sentence into a pre-learned language model (S251).

[0040] For example, the response sentence generation unit (130) may generate additional sentences based on specific speech intent and / or specific entity names extracted from a user speech sentence, and may generate modified user speech sentences by combining the generated additional sentences with existing user speech sentences. The additional sentences may refer to sentences generated to specify the corresponding user speech sentences based on, for example, specific speech intent extracted from the user speech sentence, specific entity names, or key information related to the user stored in the database (150).

[0041] The pre-trained language model can be implemented as an LLM (Large Language Model), an artificial intelligence model trained based on a number of sentences per conversation topic stored in a database (150) and response sentences corresponding to each sentence, but is not limited thereto.

[0042] The response sentence generation unit (130) can input the modified user utterance into the trained language model to obtain a response sentence (S252). In this case, the response sentence generation unit (130) can obtain the response sentence desired by the user by adjusting specific parameters set in the language model based on specific utterance intent and / or specific entity names extracted from the user utterance.

[0043] For example, if a user prefers concise responses in everyday conversations, but detailed and specific responses in information retrieval conversations, they can adjust a specific parameter value that specifies the maximum number of words used in the responses generated by the language model. For example, a user could set the max_tokens parameter value to 100 for everyday conversations and to 1,000 for information retrieval conversations, thereby limiting the maximum number of words in the responses.

[0044] In addition, users can adjust specific parameter values ​​to increase or decrease the diversity of response sentences, or to restrict or allow the use of duplicate words in response sentences.

[0045] For example, a user can set the temperature parameter value for daily conversation to 1 and the temperature parameter value for information retrieval conversation to 0.5, so that the response sentence generation unit (130) generates relatively diverse response sentences for daily conversation and generates relatively uniform and consistent response sentences for information retrieval conversation.

[0046] Additionally, for example, a user can set the penalty parameter value for everyday conversations to 1 and the penalty parameter value for information retrieval conversations to -1, thereby restricting the use of duplicate words in everyday conversations and allowing the use of duplicate words in information retrieval conversations through the response sentence generation unit (130). In this way, a user can obtain a highly satisfactory answer from the chatbot by setting specific parameter values ​​of the language model that generates the response sentences in various ways depending on the type of conversation.

[0047] The response sentence output unit (140) can convert the response sentence generated by the response sentence generation unit (130) into a chatbot voice and output it (S260). At this time, the response sentence output unit (140) can convert the text-type response sentence into a voice signal using TTS (Text-to-Speech) technology and control the converted voice signal to be spoken through the chatbot's speaker (not shown), etc.

[0048] The database (150) can store data to support various functions of the chatbot-based conversation service providing device (100). For example, the database (150) can store a number of application programs, data, commands, etc. implemented in the conversation service providing device (100). As is known to those skilled in the art, the database (150) can be implemented as various types of storage devices capable of inputting and outputting information, such as a hard disk drive (HDD), a read only memory (ROM), a random access memory (RAM), an electrically erasable and programmable read only memory (EEPROM), a flash memory, a compact flash (CF) card, a secure digital (SD) card, a smart media (SM) card, a multimedia (MMC) card, or a memory stick, and can be installed within the chatbot-based conversation service providing device (100) or installed separately externally and connected to a network.

[0049] The processor (160) is implemented as a controller, microprocessor, or microcontroller, and can operate at least one or two or more of the components included in the chatbot-based conversation service providing device (100) by combining them.

[0050] According to various embodiments, when at least one of a specific user utterance intent and a specific entity name is recognized in a user utterance sentence, the processor (160) may control the response sentence generation unit (130) to generate a response sentence based on a dialogue scenario based on the specific utterance intent and / or the specific entity name. For example, when a specific utterance intent called “weather request” is recognized through a user utterance sentence called “tell me the weather,” the processor (160) may control the response sentence generation unit (130) to generate a sentence based on a dialogue scenario based on the utterance intent.

[0051] In addition, according to various embodiments, if a specific utterance intent or a specific entity name is not recognized in a user utterance sentence, the processor (160) may control the answer sentence generation unit (130) to modify the user utterance sentence and input it into a pre-learned language model to obtain an answer sentence corresponding to the user utterance sentence. For example, if the user answers “Ugh, what’s the point of saying it” to the chatbot’s question “Where does it hurt?”, the processor (160) may determine that a specific utterance intent or a specific entity name is not recognized in the user utterance sentence, and control the answer sentence generation unit (130) to generate the same answer sentence “Where does it hurt?” or generate another answer sentence through the language model.

[0052] In addition, according to various embodiments, if the processor (160) determines that a specific user utterance is not a sentence based on a predefined conversation scenario while a conversation based on a conversation scenario is in progress between a user and a chatbot, the processor (160) may control the answer sentence generation unit (130) to generate an identical answer sentence or to input the user utterance into a pre-learned language model to obtain a corresponding answer sentence. For example, if a specific entity name 'family' is recognized in the user utterance sentence "I haven't heard from my family", but there is no conversation scenario matching it, the processor (160) may control the answer sentence generation unit (130) to obtain a answer sentence corresponding to the user utterance sentence by a pre-learned language model.

[0053] As another example, if the processor (160) determines that the user utterance sentence is not suitable for the conversation topic, conversation situation, or conversation flow, even if a specific utterance intention and / or a specific entity name is recognized in the user utterance sentence, the processor (160) may control the answer sentence generation unit (130) to generate the same answer sentence or obtain the answer sentence through a language model. For example, if the user answers “I want to eat chicken” to the chatbot’s question “Where does it hurt?”, the processor (160) may determine that the user utterance sentence is not suitable for the conversation flow, and control the answer sentence generation unit (130) to generate the same answer sentence “Where does it hurt?” or generate another answer sentence through a language model.

[0054] In addition, according to various embodiments, if the processor (160) determines that a specific user utterance is a dialogue scenario-based sentence during a conversation with the user based on a pre-learned language model, the processor (160) may control the answer sentence generation unit (130) to generate a dialogue scenario-based answer sentence based on a specific utterance intent and / or a specific entity name extracted from the user utterance. For example, if the user asks “Thank you. I’m going to go jogging, but what’s the weather like outside?” in response to the chatbot’s language model-based answer “When your back hurts, exercises such as yoga, stretching, and jogging are good,” the processor (160) may control the answer sentence generation unit (130) to generate a dialogue scenario-based answer sentence based on a specific utterance intent called “weather request” extracted from the user utterance.

[0055] Hereinafter, a process of providing a conversation service to a user using a chatbot-based conversation service providing device (100) according to one embodiment of the present invention will be described in more detail.

[0056] FIGS. 3 to 8 are diagrams illustrating an example for explaining a process of providing a conversation service to a user through a chatbot according to one embodiment of the present invention.

[0057] For reference, the chatbot described in FIGS. 3 to 8 may be a chatbot implemented by the conversation service providing device (100) of FIGS. 1 and 2.

[0058] Figure 3 illustrates an example of a conversation between a user (310) and a chatbot (320) when the user's utterance sentence is a single conversation sentence.

[0059] For example, in Example 1 of FIG. 3, when the speech of a user (310) saying “Run the cognitive game” is recognized through the speech sentence generation unit (110) of the conversation service providing device (100), the speech sentence analysis unit (120) can extract the user’s speech intention as “run a program” and the entity name as “game” from the user’s speech sentence.

[0060] In addition, the response sentence generation unit (130) can determine the user's utterance as a single conversation sentence and generate a corresponding response sentence (e.g., "Yes, I will run the cognitive game") and output it through the chatbot (320). In this case, the chatbot (320) can provide the user with a service for running the cognitive game.

[0061] In addition, in Example 2 of FIG. 3, when the user's (310) speech voice of "Go recharge" is recognized through the speech sentence generation unit (110) of the conversation service providing device (100), the speech sentence analysis unit (120) can extract the user's speech intention as "move to a place" and the entity name as "charge" from the user's speech sentence.

[0062] In addition, the response sentence generation unit (130) can determine the user's spoken sentence as a single conversation sentence and generate a corresponding response sentence (e.g., "I'm going to the charging station to charge") and output it through the chatbot (320). In this case, the chatbot (320) can provide a service that starts charging after moving to the charging station.

[0063] Figure 4 illustrates an example of a conversation between a chatbot (410) and a user (420) when the user's utterance sentence is a continuous conversation sentence.

[0064] Referring to FIG. 4, by executing (or setting) the conversation service providing device (100), the chatbot (410) can generate and output a sentence asking the user (420) how he or she is doing (e.g., “Did you sleep well? Are you feeling unwell anywhere?”).

[0065] Accordingly, if the user (420) responds “My stomach hurts,” the response sentence generation unit (130) can determine the user’s utterance as a continuous conversation sentence and generate a corresponding response sentence (e.g., “How much does your stomach hurt?”) and output it through the chatbot (410). Again, if the user (420) responds “My stomach hurts so much,” the database (150) can store information related to the user (e.g., pain location: stomach, pain intensity: high) in a database, and the response sentence generation unit (130) can generate a response sentence “Shall I call an ambulance?” based on a previous conversation between the chatbot (410) and the user (420) or user-related information stored in the database (150) and output it through the chatbot (410).

[0066] Next, if the user (420) responds with “Yes, call me,” the response sentence generation unit (130) can generate a corresponding response sentence (e.g., “Yes, I will call an ambulance and contact the guardian”) and output it through the chatbot (410). In this case, the chatbot (410) can provide a service of calling an ambulance and notifying the guardian of the fact stored in the database (150).

[0067] Figure 5 illustrates an example of modifying a user utterance sentence for input into a pre-trained language model when the user utterance sentence is not a sentence based on a pre-stored conversation scenario.

[0068] For example, in Example 1 of FIG. 5, when the user's spoken voice, "Search for a recipe for braised short ribs," is recognized through the spoken sentence generation unit (110) of the conversation service providing device (100), the spoken sentence analysis unit (120) can extract the user's spoken intent as 'information search' and the entity name as 'recipe' from the user's spoken sentence.

[0069] In this case, if the response sentence generation unit (130) determines that the user utterance sentence is not a sentence based on a conversation scenario, it can generate an additional sentence (e.g., “Please tell me in detail about materials, steps, time, etc.”) based on the extracted utterance intent and / or entity name.

[0070] That is, the response sentence generation unit (130) can input the modified user speech sentence (510) generated by combining the generated additional sentence and the existing user speech sentence into a pre-learned language model to obtain a response sentence corresponding to the user speech sentence.

[0071] In addition, in Example 2 of FIG. 5, when the user's speech voice, "My friend is very sick," is recognized through the speech sentence generation unit (110) of the conversation service providing device (100), the speech sentence analysis unit (120) can extract the user's speech intention as 'poor health' and the entity name as 'friend' from the user's speech sentence.

[0072] In this case, if the response sentence generation unit (130) determines that the user utterance sentence is not a sentence based on a conversation scenario, it can generate an additional sentence (e.g., “Please answer considering my feelings”) based on the extracted utterance intent and / or entity name.

[0073] That is, the response sentence generation unit (130) can input the modified user utterance sentence (520) generated by combining the generated additional sentence and the existing user utterance sentence into the pre-learned language model.

[0074] Figure 6 illustrates an example of modifying a user utterance based on the user utterance and key information related to the user.

[0075] Referring to FIG. 6, the processor (160) of the conversation service providing device (100) may store key information (e.g., female, 80 years old, high blood pressure) related to the user (610) in advance in the database (150) during a conversation between the user (610) and the chatbot (630). Key information related to the user may include, but is not limited to, information about the user's name, age, gender, disease, hobby, personality, family relationships, etc.

[0076] Afterwards, when the user's speech voice, "What should I eat for lunch today?" is recognized through the speech sentence generation unit (110), the speech sentence analysis unit (120) can extract the user's speech intent as 'information search' and the entity name as 'food' from the user's speech sentence.

[0077] In addition, the response sentence generation unit (130) can generate additional sentences (e.g., “I am a woman, I am 80 years old, and I have high blood pressure”) based on the extracted specific utterance intent, specific entity name, or user-related key information stored in the database (150), and input the modified user utterance sentences (620) obtained by combining the generated additional sentences with existing user utterance sentences into a pre-trained language model. The response sentence generation unit (130) can output the response sentences obtained through the language model (e.g., “How about a salmon steak that is good for high blood pressure today?”) through the chatbot (630).

[0078] FIG. 7 illustrates an example of a chatbot (720) that provides a response sentence to a user (710) based on a previous conversation between the user (710) and the chatbot (720).

[0079] Referring to FIG. 7, when the speech voice of a user (710) saying “Recommend an exercise that suits me (hereinafter, sentence ③)” is recognized through the speech sentence generation unit (110) of the conversation service providing device (100), the response sentence generation unit (130) can generate a corresponding response sentence (e.g., “Stretching is good when your back hurts (hereinafter, sentence ④)”) based on the previous conversation between the user (710) and the chatbot (720) (sentences ① and ② in FIG. 7) and output it through the chatbot (720).

[0080] Next, when the user (710) asks, “What kind of stretching do you like? (hereinafter, sentence ⑤),” the response sentence generation unit (130) can generate a corresponding response sentence (e.g., “When your back hurts, cat pose stretching, swan stretching, and bird dog stretching are good (hereinafter, sentence ⑥)”) based on the previous conversation (sentences ① to ⑤ in FIG. 7) and output it through the chatbot (720). In this case, the processor (160) can store the above-described sentences ① to ⑥ in the database (150) and utilize them in the next conversation with the user (710). For example, the processor (160) can store all conversations between the user (710) and the chatbot (720) in the database (150), or delete old conversations between the user (710) and the chatbot (720) and store only recent conversations in the database (150) and utilize them.

[0081] FIG. 8 illustrates an example of a chatbot (820) providing a response sentence to a user (810) based on a previous conversation between the user (810) and the chatbot (820).

[0082] For example, in Example 1 of FIG. 8, the processor (160) of the conversation service providing device (100) can store the user utterance sentence “My eldest son’s name is Kim Ki-chun” among the previous user utterance sentences in the database (150), and control the response sentence generation unit (130) to provide a sentence asking the user (810) how he is doing (for example, “How about contacting your eldest son Kim Ki-chun to see how he is doing?”) a few days later.

[0083] In addition, in Example 2 of FIG. 8, the processor (160) of the conversation service providing device (100) can store the user utterance sentence “My back hurts today” among the previous user utterance sentences in the database (150), and control the response sentence generation unit (130) to provide a sentence asking the user (810) how he / she is doing by utilizing this sentence a week later (for example, “Your back hurt a week ago, right? How are you now?”).

[0084] In addition, in Example 3 of FIG. 8, the processor (160) of the conversation service providing device (100) can control the response sentence generation unit (130) to store the user utterance sentence “I like sirutteok” among the previous user utterance sentences in the database (150), and use it a few days later to provide a sentence recommending food to the user (810) (for example, “You’re feeling depressed, would you like to eat some delicious sirutteok today?”).

[0085] Meanwhile, if the information stored in the database (150) based on the previous user utterance sentence and the information included in the most recent user utterance sentence conflict with each other, the processor (160) may update the information based on the information included in the most recent user utterance sentence and store the updated information in the database (150).

[0086] As described above, according to the chatbot-based conversation service providing device and method according to one embodiment of the present invention, there is an effect of being able to provide an appropriate answer desired by the user by utilizing the user's speech intention and the main keywords used in the user's speech sentence.

[0087] In addition, according to the present invention, the quality of the conversation between the user and the chatbot can be improved by analyzing the conversation flow between the user and the chatbot to more accurately determine the answer the user needs.

[0088] Meanwhile, the various embodiments described herein may be implemented by hardware, middleware, microcode, software, and / or a combination thereof. For example, the various embodiments may be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, other electronic units designed to perform the functions presented herein, or a combination thereof.

[0089] Additionally, for example, various embodiments may be embodied or encoded in a computer-readable medium containing instructions. Instructions embodied or encoded in the computer-readable medium may cause a programmable processor or other processor to perform a method when the instructions are executed, for example. A computer-readable medium includes a computer storage medium, which may be any available medium that can be accessed by a computer. For example, such a computer-readable medium may include a RAM, a ROM, an EEPROM, a CD-ROM or other optical disk storage medium, a magnetic disk storage medium, or other magnetic storage devices.

[0090] Such hardware, software, firmware, etc. may be implemented within the same device or within separate devices to support the various operations and functions described herein. Additionally, components, units, modules, components, etc. described as “units” in the present invention may be implemented together or individually as separate but interoperable logic devices. The depiction of different features for modules, units, etc. is intended to highlight different functional embodiments and does not necessarily imply that they must be realized by separate hardware or software components. Rather, the functionality associated with one or more modules or units may be performed by separate hardware or software components, or integrated into common or separate hardware or software components.

[0091] Although operations are depicted in the drawings in a specific order, this should not be construed as requiring that these operations be performed in the specific order depicted, or in any sequential order, or that all depicted operations be performed to achieve the desired results. In certain circumstances, multitasking and parallel processing may be advantageous. Furthermore, the distinction between the various components in the embodiments described above should not be construed as requiring such distinction in all embodiments, and it should be understood that the components depicted may generally be integrated together into a single software product or packaged into multiple software products.

[0092] As described above, the best practice embodiments have been disclosed in the drawings and specifications. While specific terminology has been used herein, it is solely for the purpose of describing the present invention and is not intended to limit the scope of the invention as defined in the claims. Therefore, those skilled in the art will understand that various modifications and equivalent embodiments are possible. Therefore, the true technical protection scope of the present invention should be determined by the technical spirit of the appended claims.

Claims

1. In a device providing a chatbot-based conversation service, A speech sentence generation unit that recognizes the user's voice and generates a user-spoken sentence; A speech sentence analysis unit that analyzes the user speech sentence and extracts at least one of the user's speech intention and entity name; A response sentence generation unit that determines whether the user utterance sentence is a sentence based on a pre-stored conversation scenario based on at least one of the user's utterance intention and entity name, and generates a response sentence corresponding to the user utterance sentence; and Includes a response sentence output section that converts the above response sentence into a chatbot voice and outputs it, The above answer sentence generation section, If the above user utterance sentence is not a sentence based on a pre-stored conversation scenario, the user utterance sentence is modified and then input into a pre-trained language model to obtain the above response sentence. A device that provides chatbot-based conversation services.

2. In paragraph 1, The above utterance sentence analysis unit, Extracting at least one of the user's utterance intent and the entity name from the user's utterance sentence based on at least one of a preset rule and a pre-learned utterance intent extraction model. A device that provides chatbot-based conversation services.

3. In paragraph 1, The above answer sentence generation section, If the user utterance sentence above is a sentence based on a pre-stored conversation scenario, it is determined whether the user utterance sentence is a single conversation sentence or a continuous conversation sentence based on at least one of the user's utterance intention and entity name. A device that provides chatbot-based conversation services.

4. In paragraph 3, The above answer sentence generation section, If the above user utterance sentence is determined to be a single conversation sentence, a response sentence corresponding to the user utterance sentence is generated according to a pre-stored conversation scenario. A device that provides chatbot-based conversation services.

5. In paragraph 3, The above answer sentence generation section, If the above user utterance sentence is determined to be a continuous conversation sentence, a response sentence corresponding to the user utterance sentence is continuously generated until a specific condition is satisfied. A device that provides chatbot-based conversation services.

6. In paragraph 1, The above answer sentence generation section, Generating an additional sentence based on at least one of the user's speech intention and entity name, and modifying the user's speech sentence to be input to the language model by combining the generated additional sentence and the user's speech sentence. A device that provides chatbot-based conversation services.

7. In paragraph 1, The above answer sentence generation section, Among the above user utterance sentences, user utterance sentences that include key information related to the user are stored in a database, and additional sentences generated based on the key information are combined with the user utterance sentences to modify the user utterance sentences for input to the language model. A device that provides chatbot-based conversation services.

8. In paragraph 1, The above answer sentence generation section, Obtaining the answer sentence by adjusting a specific parameter set in the language model based on at least one of the user's speech intention and entity name. A device that provides chatbot-based conversation services.

9. In paragraph 1, The above pre-trained language model is, It is learned based on a number of sentences per conversation topic stored in the database and the corresponding response sentences for each sentence. A device that provides chatbot-based conversation services.

10. In a method for providing a conversation service performed by a chatbot-based conversation service providing device, An operation to train a language model based on a number of sentences per conversation topic stored in a database and response sentences corresponding to each sentence; An action that recognizes the user's voice and generates a user-spoken sentence; An action of analyzing the user's utterance sentence and extracting at least one of the user's utterance intent and entity name; An operation of determining whether the user-spoken sentence is a sentence based on a previously stored conversation scenario based on at least one of the user's utterance intention and entity name; An action of generating a response sentence corresponding to the user utterance sentence; and Includes an action of converting the above response sentence into a chatbot voice and outputting it, The action of generating the above response sentence is: If the above user utterance sentence is not a sentence based on a pre-stored conversation scenario, an action of modifying the user utterance sentence; and Including an action of inputting the above modified user utterance sentence into the above learned language model to generate the above response sentence. Method for providing a chatbot-based conversation service.

11. In paragraph 10, The operation of extracting at least one of the user's speech intent and entity name is: An operation of extracting at least one of the user's utterance intent and the entity name from the user's utterance sentence based on at least one of a preset rule and a pre-learned utterance intent extraction model, Method for providing a chatbot-based conversation service.

12. In paragraph 10, If the user utterance sentence is a sentence based on a pre-stored conversation scenario, the method further includes an operation of determining whether the user utterance sentence is a single conversation sentence or a continuous conversation sentence based on at least one of the user's utterance intention and the entity name. Method for providing a chatbot-based conversation service.

13. In paragraph 12, The action of generating the above response sentence is: If the above user utterance sentence is determined to be a single conversation sentence, the method includes generating a response sentence corresponding to the user utterance sentence according to a pre-stored conversation scenario. Method for providing a chatbot-based conversation service.

14. In paragraph 12, The action of generating the above response sentence is: If the user utterance sentence above is determined to be a continuous conversation sentence, the method includes an action of continuously generating a response sentence corresponding to the user utterance sentence until a specific condition is satisfied. Method for providing a chatbot-based conversation service.

15. In paragraph 10, The action of modifying the above user-spoken sentence is: An action of generating an additional sentence based on at least one of the user's speech intent and the entity name; and An operation of modifying a user-spoken sentence for input into the language model by combining the generated additional sentence and the user-spoken sentence, Method for providing a chatbot-based conversation service.

16. In paragraph 10, The action of modifying the above user-spoken sentence is: An operation of storing user utterances containing key information related to the user among the user utterances in a database; and An operation of modifying a user utterance sentence for input to the language model by combining the user utterance sentence with an additional sentence generated based on key information related to the user, Method for providing a chatbot-based conversation service.

17. In paragraph 10, The action of generating the above response sentence is: An operation of generating the answer sentence by adjusting a specific parameter set in the language model based on at least one of the user's speech intention and the entity name, Method for providing a chatbot-based conversation service.

Images