Chatbot service system for perceiving user intent based on knowledge graph

Abstract

The present invention relates to a chatbot service system for identifying a user's utterance intent based on a knowledge graph. When the intent of a utterance is unclear based on a pre-established knowledge graph and it is unclear what information should be provided, the system guides the flow of conversation to make the user's utterance intent more specific in order to re-identify the user's utterance intent, thereby enabling the effective provision of desired answer information that matches the user's utterance intent.

Description

Technology Field

[0001] The present invention relates to a chatbot service system for identifying a user's utterance intention based on a knowledge graph. Background Technology

[0002] Generally, while various daily complaints are increasing due to the complexity and diversity of society as a whole, counselors at actual counseling centers are facing difficulties due to limited personnel. Users want to receive the information they desire immediately at their preferred time through personalized conversations, without being constrained by time.

[0003] In order to achieve such efficient communication, dialogue systems or chatbots capable of automatically answering user questions are becoming widely available due to the recent development of information processing and artificial intelligence technologies.

[0004] These chatbots are a technology that allows users to receive necessary information or services through a familiar conversational interface without having to run a separate website or app.

[0005] Chatbots for customer consultation have been used to automate the task of responding to customer inquiries via telephone or website, and in this case, the main goal is to accurately understand the user's intent of the question and provide an appropriate answer.

[0006] In other words, chatbots for customer consultation are primarily used for consultations targeting an unspecified number of people, and they utilize a method of storing expected question / answer pairs in advance without considering individual characteristics, and then extracting the question / answer pair closest to the user's question through machine learning or pattern matching to provide an answer.

[0007] These chatbots identify the user's intent and provide the most suitable information to satisfy their needs. However, the technology behind all these chatbots fundamentally relies on big data. In other words, they utilize artificial intelligence to output answers that match the intent through big data.

[0008] However, it is difficult to identify the precise intent from a user's utterance. While the role of a chatbot is to identify appropriate intent through natural language processing and provide corresponding information, if the precise intent is not identified, the user will not receive the information they want.

[0009] In addition, in conversations between people, if one fails to grasp the other person's intent, one may use gestures or follow-up questions to re-identify that intent and facilitate a natural conversation. Similarly, regarding utterances that are not understood, chatbots can guide and re-identify the user's intent based on the situation, the flow of the conversation, and the diversity of intents, rather than simply mechanically re-questioning, thereby enabling efficient communication.

[0010] Meanwhile, cases where the user's utterance intent is inaccurate can be broadly divided into three categories. Specifically, first, issues caused by unclear pronunciation or typos; second, utterances that are ambiguous or perceived as having multiple intentions; and third, cases where the user uses similar words because they themselves do not know the exact word.

[0011] In all of the aforementioned situations, it is difficult for chatbots to provide accurate responses. This is because many chatbot technologies provide answers extracted from conversation example databases (DB) or answers generated according to predefined business rules.

[0012] Accordingly, the chatbot can exclude all of the above cases by guiding the conversation to determine what intention the user had, instead of simply responding that it did not understand the user's utterance. Prior art literature

[0013] Korean Registered Patent No. 10-1840420 (Published May 4, 2018) The problem to be solved

[0014] The present invention has been devised to solve the aforementioned problems, and the objective of the present invention is to provide a chatbot service system for identifying a user's utterance intent based on a knowledge graph, which induces the flow of conversation to make the user's utterance intent more specific in order to re-identify the user's utterance intent when the intention of the utterance is unclear regarding the information learned based on a pre-constructed knowledge graph and it is not accurate to determine what information to provide, thereby enabling the effective provision of desired answer information that matches the user's utterance intent. means of solving the problem

[0015] To achieve the aforementioned objective, one aspect of the present invention comprises: at least one user terminal that transmits chatbot question content intended by a user; and a chatbot server that receives chatbot question content intended by each user transmitted from each user terminal, matches each user’s chatbot question with each intent through natural language understanding / processing based thereon, and provides a chatbot chat service by recommending each answer corresponding to each matched intent and providing it to each user terminal. The system includes a knowledge graph construction device that constructs a knowledge graph in the form of a graph expressed based on knowledge relationships, wherein for each chatbot question intended by each user, each answer corresponding to a matched / unmatched utterance intention is connected, at least one related word node associated with each utterance intention is connected via a related edge, at least one similar word node having a similar meaning to each related word node is connected via a similar edge, or at least one homonymous word node having two meanings to each related word node is connected via a homonymous edge; wherein, if there is no matched utterance intention for each user's chatbot question, the chatbot server selects a plurality of unmatched utterance intentions corresponding to a pre-set utterance intention selection criterion matching rate for selecting unmatched utterance intentions using a machine learning utterance intention matching module, and uses the knowledge graph constructed by the knowledge graph construction device to connect a related word node, a similar edge, or a homonymous edge to at least one edge of the knowledge graph for each selected unmatched utterance intention. The present invention provides a chatbot service system for identifying a user's utterance intent based on a knowledge graph, characterized by providing a chatbot chat service that selects words corresponding to at least one word node among word nodes or homonym word nodes as recommended words and transmits them to each user terminal.

[0016] Here, it is preferable that each user terminal provides a user interface service so that each user selects and inputs one of the recommended words transmitted from the chatbot server.

[0017] Preferably, the chatbot server can provide a chatbot chat service by receiving a recommended word selected and entered by each user from each user terminal and using a knowledge graph built by the knowledge graph building device based thereon, so that an answer corresponding to an unmatched utterance intention finally connected to the recommended word selected and entered by each user is transmitted to each user terminal.

[0018] Preferably, when there are multiple nodes connected to a recommended word selected and entered by each user from each user terminal, the chatbot server may recommend a word or an unmatched utterance intention corresponding to the multiple nodes and transmit it to each user terminal so that the user can select which of the multiple nodes is the question most desired by each user or the word with the closest meaning.

[0019] Preferably, the chatbot server can provide a chatbot chat service by using a knowledge graph constructed from the knowledge graph construction device to select words corresponding to word nodes connected to the ends of knowledge graph nodes for each selected unmatched utterance intention as recommended words and transmit them to each user terminal.

[0020] Preferably, the chatbot server can provide a chatbot chat service based on a chatbot chat service for police civil complaint consultation, by comparing the content of the chatbot question intended by each user with utterance intentions learned by artificial intelligence through a pre-configured machine learning model, and recommending each answer corresponding to each utterance intention matched to each user's chatbot question and transmitting it to each user's terminal.

[0021] Preferably, the knowledge graph constructed from the knowledge graph construction device may be represented in the form of a graph based on knowledge relationships in which each answer corresponding to each utterance intention that is matched / unmatched for each chatbot question intended by each user is connected, at least one related word node associated with each utterance intention is connected to a related edge, and at least one similar word node having a similar meaning to each related word node is connected to a similar edge.

[0022] Preferably, the knowledge graph constructed from the knowledge graph construction device may be represented in the form of a graph based on knowledge relationships in which each answer corresponding to each utterance intention that is matched / unmatched for each chatbot question intended by each user is connected, at least one related word node associated with each utterance intention is connected to a related edge, and at least one homonymous word node having two meanings is connected to each related word node to a homonymous edge.

[0023] Preferably, the knowledge graph constructed from the knowledge graph construction device may be represented in the form of a graph based on knowledge relationships in which each answer corresponding to each utterance intention that is matched / unmatched for each chatbot question intended by each user is connected, at least one related word node associated with each utterance intention is connected to a related edge, at least one similar word node having a similar meaning to each related word node is connected to a similar edge, and at least one homonymous word node having two meanings to each related word node is connected to a homonymous edge.

[0024] Preferably, the knowledge graph constructed from the knowledge graph construction device may be represented in a graph form based on knowledge relationships consisting of at least one word node among related word nodes, similar word nodes, or homonymous word nodes connected to each utterance intention, including utterance intentions related to public security.

[0025] Preferably, the machine learning utterance intention matching module generates an artificial intelligence learning model for matching utterance intentions for each chatbot question by machine learning a plurality of data sets, which are composed of a plurality of chatbot questions and utterance intentions corresponding to each chatbot question, and then assigns a matching rate to pre-stored utterance intentions using each user's chatbot question as input data to the generated artificial intelligence learning model, and then matches the utterance intention for each chatbot question according to the matching rate assigned to each utterance intention, or selects a plurality of unmatched utterance intentions corresponding to a pre-set utterance intention selection criterion matching rate for selecting unmatched utterance intentions.

[0026] Preferably, the machine learning utterance intention matching module can determine and match the utterance intention as a matching utterance intention for the user's chatbot question when the matching rate assigned to each utterance intention is greater than or equal to a preset threshold, and determine the utterance intention as a non-matching utterance intention for the user's chatbot question when the matching rate assigned to each utterance intention is less than or equal to a preset threshold, thereby producing a result of non-matching.

[0027] Preferably, the matching rate of the utterance intention selection criteria can be set to be smaller than the threshold value.

[0028] Preferably, the machine learning utterance intention matching module can vectorize the nouns of each user's chatbot question using a pre-set word embedding technique to create feature vectors before inputting each user's chatbot question as input data into the generated artificial intelligence learning model.

[0029] Preferably, the word embedding technique set above may use the algorithm of a Term Frequency-Inverse Document Frequency (TF-IDF) model.

[0030] Preferably, when machine learning a plurality of data sets, the machine learning utterance intention matching module can be trained using a multi-class classification method. Effects of the invention

[0031] According to the chatbot service system for identifying a user's utterance intention based on the knowledge graph of the present invention as described above, when the intention of the utterance is unclear regarding the information learned based on the pre-established knowledge graph and it is unclear what information should be provided, the flow of conversation is guided to make the user's utterance intention more specific in order to re-identify the user's utterance intention, thereby having the advantage of effectively providing desired answer information that matches the user's utterance intention.

[0032] In addition, according to the present invention, instead of providing only a predetermined answer based on the user's speech intention, it is possible to provide effective information by inducing a conversation flow in different directions in various aspects, and there is an advantage in that one can expect the role of an efficient information provider rather than a passive chatbot. Brief explanation of the drawing

[0033] FIG. 1 is an overall block diagram illustrating a chatbot service system for identifying a user's speech intention based on a knowledge graph according to an embodiment of the present invention. FIG. 2 is a conceptual diagram showing an example of a knowledge graph constructed through a knowledge graph construction device applied to one embodiment of the present invention. FIG. 3 is a conceptual diagram showing another example of a knowledge graph constructed through a knowledge graph construction device applied to one embodiment of the present invention. FIG. 4 is a diagram showing an example of conversation content for selecting recommended words between a user terminal and a chatbot server applied in one embodiment of the present invention. Specific details for implementing the invention

[0034] The aforementioned objectives, features, and advantages are described in detail below with reference to the attached drawings, and accordingly, a person skilled in the art to which the present invention pertains will be able to easily implement the technical concept of the present invention. In describing the present invention, detailed descriptions of known technologies related to the present invention are omitted if it is determined that such descriptions may unnecessarily obscure the essence of the present invention.

[0035] Terms including ordinal numbers, such as first, second, etc., may be used to describe various components, but said components are not limited by said terms. Such terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component. The terms used in this application are used merely to describe specific embodiments and are not intended to limit the present invention. A singular expression includes a plural expression unless the context clearly indicates otherwise.

[0036] The terms used in this invention have been selected based on currently widely used general terms, taking into account their functions within the invention; however, these terms may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the relevant description of the invention. Therefore, the terms used in this invention should be defined not merely by their names, but based on their meanings and the overall content of the invention.

[0037] When a part of a specification is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "...part" or "module" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software.

[0038] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, the embodiments of the present invention exemplified below may be modified in various different forms, and the scope of the present invention is not limited to the embodiments described below. The embodiments of the present invention are provided to more completely explain the present invention to those skilled in the art.

[0039] Combinations of each block of the attached block diagram and each step of the flowchart may be executed by computer program instructions (execution engines), and since these computer program instructions may be loaded into the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing equipment, the instructions executed through the processor of the computer or other programmable data processing equipment create means to perform the functions described in each block of the block diagram or each step of the flowchart. Since these computer program instructions may also be stored in computer-available or computer-readable memory that can be directed toward the computer or other programmable data processing equipment to implement the function in a specific way, the instructions stored in computer-available or computer-readable memory may also produce a manufactured item containing instruction means to perform the function described in each block of the block diagram or each step of the flowchart.

[0040] And, since computer program instructions can be loaded onto a computer or other programmable data processing equipment, instructions that execute a computer or other programmable data processing equipment by performing a series of operation steps on the computer or other programmable data processing equipment to create a process executed on the computer can also provide steps for executing the functions described in each block of the block diagram and each step of the flowchart.

[0041] Additionally, each block or each step may represent a module, segment, or part of code containing one or more executable instructions for executing specific logical functions, and it should be noted that in some alternative embodiments, the functions mentioned in the blocks or steps may occur out of order. For example, two blocks or steps described in succession may actually be performed substantially simultaneously, and the blocks or steps may also be performed in the reverse order of the corresponding functions as needed.

[0042] First, the chatbot mentioned throughout this invention is a compound word of chat and robot, referring to a conversational system that interacts with people based on artificial intelligence, which is a technology that incorporates the artificial intelligence of a robot into a conversational interface. With the advancement of artificial intelligence technology, machine learning or deep learning technologies are applied to chatbots, enabling them to understand user questions, grasp intent, and provide appropriate answers, thereby allowing them to perform more complex tasks.

[0043] These chatbots can be classified according to their mode of operation into rule-based chatbots and machine learning-based chatbots, according to their method of information exchange into one-time question-and-answer chatbots and continuous conversational chatbots, and according to their method of answer generation into search model chatbots and generative model chatbots.

[0044] The above-mentioned rule-based chatbot operates based on predefined rules and includes interpretation rules that interpret user input, and response rules that react to user input or rules that generate a response. The above-mentioned machine learning-based chatbot uses machine learning algorithms to understand natural language and generate or select appropriate answers. The Natural Language Processing (NLP) module performs the role of decomposing the natural language input by the user using various linguistic resources, and the Natural Language Understanding (NLU) module identifies the meaning of the decomposed natural language and transmits the refined information to a final classification model or generative model for generating answers.

[0045] The above one-time Q&A chatbot does not require conversation management functions and operates in a manner similar to a search engine that presents the most relevant answer to the question. Most FAQ chatbots that provide fixed answers to frequently asked questions are designed in the form of one-time Q&A. The above continuous conversational chatbot has a start and end to the conversation, collects information from the user to achieve a specific purpose, and terminates the conversation when a final response is provided based on the collected information.

[0046] The search model chatbot described above operates by selecting and presenting the most appropriate answer from among the answers stored in an existing knowledge base in response to a user's question. The generative model chatbot described above generates and provides a new answer that is not previously defined in response to a user's question. The present invention is applicable to the various types of chatbots described above.

[0047] FIG. 1 is an overall block diagram for explaining a chatbot service system for identifying a user's utterance intention based on a knowledge graph according to an embodiment of the present invention, FIG. 2 is a conceptual diagram showing an example of a knowledge graph constructed through a knowledge graph construction device applied to an embodiment of the present invention, FIG. 3 is a conceptual diagram showing another example of a knowledge graph constructed through a knowledge graph construction device applied to an embodiment of the present invention, and FIG. 4 is a diagram showing an example of conversation content for selecting recommended words between a user terminal and a chatbot server applied to an embodiment of the present invention.

[0048] Referring to FIGS. 1 to 4, a chatbot service system for identifying a user's utterance intention based on a knowledge graph according to an embodiment of the present invention comprises, broadly, at least one user terminal (100-1 to 100-N), a chatbot server (200), and a knowledge graph construction device (300). Meanwhile, since the components shown in FIGS. 1 to 4 are not essential, the chatbot service system for identifying a user's utterance intention based on a knowledge graph according to an embodiment of the present invention may have more components or fewer components.

[0049] Hereinafter, the components of a chatbot service system for identifying a user's utterance intention based on a knowledge graph according to one embodiment of the present invention will be examined in detail as follows.

[0050] Each user terminal (100-1 to 100-N) performs the function of transmitting the content of a chatbot question intended by the user to a chatbot server (200) through a communication network (10). At this time, the communication network (10) is preferably composed of Ethernet or a mobile communication network, and may be a high-speed backbone network of a large communication network capable of providing large-capacity, long-distance voice and data services, and may be a next-generation wireless network including the Internet or Wi-Fi, WiGig, Wibro (Wireless Broadband Internet), WiMAX (World Interoperability for Microwave Access), etc., for providing high-speed multimedia services.

[0051] The above Internet refers to a global open computer network structure that provides the TCP / IP protocol and various services existing in the upper layer, namely HTTP (Hyper Text Transfer Protocol), Telnet, FTP (File Transfer Protocol), DNS (Domain Name System), SMTP (Simple Mail Transfer Protocol), SNMP (Simple Network Management Protocol), NFS (Network File Service), NIS (Network Information Service), etc., and provides an environment that enables each user terminal (100-1 to 100-N) to connect to the chatbot server (200). Meanwhile, the above Internet may be a wired or wireless Internet, and may also be a core network integrated with a wired public network, a wireless mobile communication network, or a mobile Internet.

[0052] If the communication network (10) is a mobile communication network, it may be a synchronous mobile communication network or an asynchronous mobile communication network. As an example of an asynchronous mobile communication network, a WCDMA (Wideband Code Division Multiple Access) type communication network may be cited. In this case, although not shown in the drawing, the mobile communication network may include an RNC (Radio Network Controller). Meanwhile, although a WCDMA network was given as an example, it may be a next-generation communication network such as a 3G LTE network, a 4G network, a 5G network, or other IP-based IP networks. Such a communication network (10) performs the role of mutually transmitting signals and data between each user terminal (100-1 to 100-N) and a chatbot server (200).

[0053] Additionally, each user terminal (100-1 to 100-N) can perform the function of providing a user interface service so that each user selects and inputs one of the recommended words transmitted from the chatbot server (200) through the communication network (10).

[0054] Additionally, each user terminal (100-1 to 100-N) can receive information data (i.e., question and answer content, etc.) of the answer to each user's chatbot question transmitted from the chatbot server (200) via the communication network (10) and provide a service so that each user can visually and / or audibly check it.

[0055] Each of these user terminals (100-1 to 100-N) may be a terminal equipped with a chatbot client, which is a chatbot service application that receives the content of a chatbot question intended by each user, transmits it to a chatbot server (200) through a communication network (10), and receives answer information for the chatbot question from the chatbot server (200) and provides it to each user.

[0056] Meanwhile, each of the aforementioned user terminals (100-1 to 100-N) is preferably composed of at least one mobile terminal device among a smartphone, smart pad, and / or smart note that communicates via wireless internet or mobile internet, and additionally, it may comprehensively refer to all wired / wireless home appliance / communication systems having a user interface for connecting to the chatbot server (200), such as a Palm PC, a mobile game console (Mobile PlayStation), a DMB (Digital Multimedia Broadcasting) phone with communication functions, a tablet PC, and an iPad.

[0057] The chatbot server (200) receives the content of a chatbot question intended by each user transmitted from each user terminal (100-1 to 100-N), matches each user’s utterance intent to each chatbot question through natural language understanding and / or processing based on this, recommends each answer corresponding to each matched utterance intent, and performs the function of providing a chatbot chat service so that it is provided to each user terminal (100-1 to 100-N) through the communication network (10).

[0058] Additionally, if there is no matched utterance intention for each user's chatbot question, the chatbot server (200) uses a machine learning utterance intention matching module (250) to select a plurality of unmatched utterance intentions corresponding to a pre-set utterance intention selection criteria matching rate for selecting unmatched utterance intentions, and uses a knowledge graph constructed from a knowledge graph construction device (300) to select words corresponding to at least one word node among a related word node, a similar word node, and / or a homonymous word node connected to at least one edge among a related edge, a similar edge, and / or a homonymous word edge of the knowledge graph for each selected unmatched utterance intention as recommended words, and performs the function of providing a chatbot chat service so that the words are transmitted to each user terminal (100-1 to 100-N) through a communication network (10).

[0059] Here, the machine learning utterance intention matching module (250) is a data set composed of multiple chatbot questions and utterance intentions corresponding to each chatbot question, and machine learns the multiple data sets to create an artificial intelligence learning model for matching each utterance intention for each chatbot question, and uses each user's chatbot question as input data to the created artificial intelligence learning model to assign a matching rate (or matching probability score) to the previously stored utterance intentions, and then, according to the matching rate assigned to each utterance intention, it can perform the function of matching the utterance intention for each chatbot question or selecting multiple unmatched utterance intentions corresponding to the pre-set utterance intention selection criteria matching rate for selecting unmatched utterance intentions.

[0060] That is, the machine learning utterance intention matching module (250) can determine and match the utterance intention as a matching utterance intention for the user's chatbot question if the matching rate assigned to each utterance intention is greater than or equal to a preset threshold (e.g., 90% or 0.9), and can determine the utterance intention as a non-matching utterance intention for the user's chatbot question if the matching rate assigned to each utterance intention is less than or equal to the preset threshold, thereby producing a result of non-matching. Meanwhile, it is preferable that the utterance intention selection criteria matching rate (e.g., 70% to 90% or 0.7 to 0.9) be set to be smaller than the threshold.

[0061] For example, the machine learning speech intent matching module (250) may use Deep Neural Networks (DNN), a type of machine learning, to match each user's speech intent. The neural network is trained using each user's speech intent and a chatbot question (or question text) related to that user's speech intent as data pairs.

[0062] At this time, by training using a Multi-Class Classification method and using arbitrary chatbot questions as input data for this neural network, a matching probability score (e.g., a score between 0 and 1) is assigned to each user's utterance intention. Meanwhile, before inputting each user's chatbot question, the nouns in the chatbot question can be vectorized using a pre-configured Word Embedding technique to create a feature vector.

[0063] Here, if the utterance intention of the user with the highest score exceeds the threshold set by the user, it is matched to that user's utterance intention, otherwise, the result is a mismatch.

[0064] Meanwhile, the word embedding technique set above preferably uses, for example, the algorithm of the Term Frequency-Inverse Document Frequency (TF-IDF) model, but is not limited thereto, and may also use well-known techniques such as CBOW (Continuous Bag-of-Words), Skip-gram, Word2Vec, etc.

[0065] In other words, the above word embeddings represent the meaning of the word itself as a vector in a multidimensional space, and these word embeddings are based on the Distributional Hypothesis.

[0066] The distributional hypothesis mentioned above is the assumption that words with similar contexts have similar meanings, and it is the core foundation of modern statistical natural language processing. Word embedding methods based on the distributional hypothesis can be broadly classified into count-based methods, such as Latent Semantic Analysis (LSA), and predictive methods, such as Neural Probabilistic Language Models (NPLM).

[0067] The former is a method that utilizes how often a word appears with other words in a context, while the latter is a method that directly predicts words using contextual words represented as vectors. Deep learning-based statistical language models using the aforementioned word embeddings are known to have superior performance compared to traditional statistical language models. However, word embeddings are also not immune to the problem of data scarcity. In particular, it is important to handle unknown words that do not appear in the training data. Recently, research to improve the quality of word embeddings through combination with knowledge bases, such as lexical semantic networks, has been actively conducted.

[0068] Various word embedding methods based on the above distribution hypothesis have been proposed, and recently, the following three architectures are the most widely used.

[0069] First, Continuous Bag-of-Words (CBOW) is a prediction method based on the distributional hypothesis. It is similar to the Feedforward Neural Network Language Model (NNLM) but lacks hidden layers. It predicts words from contextual words and performs well on small datasets.

[0070] Furthermore, Skip-gram is a prediction method based on the distributional hypothesis and is provided by Word2vec along with CBOW. Unlike CBOW, it predicts contextual words from words, and due to its fast training speed, it is primarily used for word embeddings based on large-scale data. Additionally, GloVe (Global Vectors for Word Representation) is a frequency-based method based on the distributional hypothesis that uses each row of the Global Co-occurrences Matrix as a word vector.

[0071] Recently, word embedding methods that are easy to use with open source libraries such as TensorFlow and exhibit fast training speeds are gaining popularity. In particular, CBOW and Skip-gram, which apply various methods (Hierarchical Softmax, Negative Sampling) that can drastically reduce the amount of Softmax computation in the output layer, are the most widely used.

[0072] Additionally, the chatbot server (200) can perform the function of providing a chatbot chat service by receiving recommended words selected and entered by each user from each user terminal (100-1 to 100-N) and using a knowledge graph built from a knowledge graph building device (300) based on this, so that an answer corresponding to an unmatched speech intention finally connected to the recommended words selected and entered by each user is transmitted to each user terminal (100-1 to 100-N) through the communication network (10).

[0073] Additionally, the chatbot server (300) can perform the function of recommending a word or a mismatched speech intention corresponding to the multiple nodes (e.g., similar word node, related word node, or homonymous word node, etc.) connected to a recommended word selected and input by each user from each user terminal (100-1 to 100-N) when there are multiple nodes (e.g., similar word node, related word node, or homonymous word node, etc.) among the multiple nodes so that each user can select which node is the question they most want (e.g., mismatched speech intention) or the word with the closest meaning (e.g., similar word, related word, or homonymous word, etc.) and transmitting it to each user terminal (100-1 to 100-N) through the communication network (10).

[0074] Additionally, the chatbot server (300) can perform the function of providing a chatbot chat service by using a knowledge graph constructed from a knowledge graph construction device (300) to select words corresponding to word nodes (e.g., similar word nodes, related word nodes, or homonymous word nodes, etc.) connected to the end of a knowledge graph node for each selected unmatched utterance intention as recommended words and transmitting them to each user terminal (100-1 to 100-N) through a communication network (10).

[0075] Additionally, the chatbot server (200) can perform the function of providing a chatbot chat service, for example, based on a chatbot chat service for police civil complaint consultation, by comparing the content of the chatbot question intended by each user with speech intentions learned by artificial intelligence through a pre-set machine learning model, recommending each answer corresponding to each speech intention matched to each user's chatbot question, and transmitting it to each user terminal (100-1 to 100-N) through a communication network (10).

[0076] Meanwhile, the aforementioned pre-configured machine learning model is an artificial intelligence model that predicts the future by analyzing vast amounts of data, and is a model in which the computer solves problems by acquiring uninputted information through a machine learning process on its own.

[0077] For such machine learning, deep learning techniques utilizing neural networks such as CNN (Convolutional Neural Network), R-CNN (Region with Convolutional Neural Network), DNN (Deep Neural Networks), or LSTM (Long Short Term Memory) can be used. Since these machine learning techniques are already well known, a detailed explanation is omitted here.

[0078] The aforementioned chatbot server (200) may be a server that provides a chatbot service web page, app page, program, or application that adaptively reuses a question-and-answer data set. This chatbot server (200) may be implemented as a computer capable of connecting to a remote server or terminal via a communication network (10). Here, the computer may include, for example, a laptop, desktop, or laptop equipped with navigation and a web browser.

[0079] And, the knowledge graph constructor (300) is connected to the chatbot server (200), and each answer corresponding to each utterance intent (e.g., 'random number generation automatic lottery', 'participation in bidding by type', 'ship bidding', 'fruit import', see FIG. 2 and FIG. 3) that is matched and / or unmatched for each chatbot question intended by each user is connected, and at least one related word node associated with each utterance intent (e.g., 'random number', 'generation', 'automatic', 'lottery', 'ship', 'auto', see FIG. 2 and FIG. 3) is connected as a related edge, and at least one similar word node having a similar meaning (i.e., synonym) to each related word node (e.g., 'random', 'random', 'auto', see FIG. 2) is connected as a similar edge, or at least one homonymous word node having two meanings to each related word node (e.g., 'pear', FIG. 3 It performs the function of constructing a knowledge graph in the form of a graph based on knowledge relationships connected by homonym edges.

[0080] Additionally, the knowledge graph constructed from the knowledge graph construction device (300) can be represented in the form of a graph based on knowledge relationships, as shown in FIG. 3, in which each answer corresponding to each utterance intention (e.g., 'ship bidding', 'fruit import') that is matched and / or unmatched for each chatbot question intended by each user is connected, at least one related word node (e.g., 'ship', 'auto') associated with each utterance intention is connected to a related edge, and at least one homonymous word node (e.g., 'boat') having two meanings is connected to each related word node to a homonymous edge.

[0081] Additionally, the knowledge graph constructed from the knowledge graph construction device (300) may be represented in the form of a graph based on knowledge relationships, in which the embodiments of FIGS. 2 and FIGS. 3 are integrated as shown in FIGS. 2 and FIGS. 3, so that each answer corresponding to each utterance intention that is matched and / or unmatched for each chatbot question intended by each user is connected, at least one related word node associated with each utterance intention is connected to a related edge, at least one similar word node having a similar meaning to each related word node is connected to a similar edge, and at least one homonymous word node having two meanings to each related word node is connected to a homonymous edge.

[0082] Additionally, the knowledge graph constructed from the knowledge graph construction device (300) may be expressed in a graph form based on knowledge relationships consisting of at least one word node among related word nodes, similar word nodes, and / or homonymous word nodes connected to each of the speech intentions, for example, speech intentions related to public security.

[0083] That is, the knowledge graph constructed from the knowledge graph construction device (300) is a collection of explanations that are mutually connected like a graph for individual knowledge. It has the effect of understanding intent by thinking like a human, rather than understanding intent by interpreting only specific words, by considering the relationships between words and the flow of sentences. In addition, it can be used in various fields of artificial intelligence (AI) as it is possible to distinguish homonyms, process vast amounts of data quickly, and understand different intents depending on the method of keyword combination and real-time processing. That is, not only is simple data stored, but information on the relationships between data is also stored, enabling pattern analysis and understanding of patterns of exceptions.

[0084] As described above, a chatbot service system for identifying a user's utterance intention based on a knowledge graph according to one embodiment of the present invention can effectively provide desired answer information that matches the user's utterance intention by guiding the flow of conversation to make the user's utterance intention more specific, by sequentially showing recommended word options (e.g., first a noun, then options of nouns related to that noun) and allowing the user to select them so that the user can approach the question closest to the desired question based on a pre-established knowledge graph when the intention of the utterance is unclear and it is not accurate to provide information.

[0085] Although a preferred embodiment of a chatbot service system for identifying a user's speech intention based on a knowledge graph according to the present invention has been described, the present invention is not limited thereto and can be implemented with various modifications within the scope of the claims, the detailed description of the invention, and the attached drawings, and such modifications also fall within the scope of the present invention. Explanation of the symbols

[0086] 100-1 to 100-N: User terminal, 200 : Chatbot server, 250 : Machine Learning Utterance Intent Matching Module, 300 : Knowledge Graph Constructor

Claims

Claim 1 At least one user terminal that transmits the content of a chatbot question intended by a user; a chatbot server that receives the content of a chatbot question intended by each user transmitted from each user terminal, matches each user's utterance intent to each chatbot question through natural language understanding / processing based thereon, and provides a chatbot chat service by recommending each answer corresponding to each matched utterance intent and providing it to each user terminal; The present invention includes a knowledge graph construction device that constructs a knowledge graph in the form of a graph expressed based on knowledge relationships in which, for a chatbot question intended by each user, each answer corresponding to each matched / unmatched utterance intention is connected, at least one related word node associated with each utterance intention is connected via a related edge, at least one similar word node having a similar meaning to each related word node is connected via a similar edge, and at least one homonymous word node having two meanings to each related word node is connected via a homonymous edge; wherein each user terminal provides a user interface service to allow each user to select and input one of the recommended words transmitted from the chatbot server; and if there is no matched utterance intention for each user's chatbot question, the chatbot server selects a plurality of unmatched utterance intentions corresponding to a pre-set utterance intention selection criterion matching rate for selecting unmatched utterance intentions using a machine learning utterance intention matching module, and uses the knowledge graph constructed from the knowledge graph construction device to [address] each selected unmatched utterance intention A chatbot chat service is provided to select words corresponding to related word nodes, similar word nodes, and homonym word nodes connected by related edges, similar edges, and homonym edge of the knowledge graph as recommended words and send them to each user terminal, andA chatbot chat service is provided to receive recommended words selected and entered by each user from each user terminal, and based on this, to utilize a knowledge graph constructed by the knowledge graph construction device to transmit to each user terminal an answer corresponding to the mismatched utterance intention finally connected to the recommended words selected and entered by each user; if there are multiple nodes connected to the recommended words selected and entered by each user from each user terminal, the service recommends words or mismatched utterance intentions corresponding to those multiple nodes and transmits them to each user terminal so that the user can select which of the multiple nodes represents the question most desired or the word with the closest meaning; and the service is provided to utilize the knowledge graph constructed by the knowledge graph construction device to select words corresponding to the word nodes connected to the ends of the knowledge graph nodes for each selected mismatched utterance intention as recommended words and transmit them to each user terminal; and the machine learning utterance intention matching module is a data set composed of multiple chatbot questions and utterance intentions corresponding to each chatbot question, and machine learns the multiple data sets to determine the utterance intention for each chatbot question An artificial intelligence learning model for matching is created, and a matching rate is assigned to pre-stored utterance intentions using each user's chatbot question as input data to the created artificial intelligence learning model. Then, based on the matching rate assigned to each utterance intention, the utterance intention for each chatbot question is matched, or multiple unmatched utterance intentions corresponding to the pre-set utterance intention selection criteria matching rate for selecting unmatched utterance intentions are selected. If the matching rate assigned to each utterance intention is greater than or equal to a pre-set threshold, the utterance intention is determined to be a matched utterance intention for the user's chatbot question and matched; if the matching rate assigned to each utterance intention is less than or equal to the pre-set threshold, the utterance intention is determined to be an unmatched utterance intention for the user's chatbot question and a result of unmatching is derived.A chatbot service system for identifying a user's utterance intention based on a knowledge graph, characterized in that the matching rate of the utterance intention selection criteria is set to be smaller than the threshold value. Claim 2 delete Claim 3 delete Claim 4 delete Claim 5 A chatbot service system for identifying a user's utterance intention based on a knowledge graph according to claim 1, wherein the chatbot server provides a chatbot chat service based on a chatbot chat service for police civil complaint consultation, wherein the chatbot server compares the content of a chatbot question intended by each user with utterance intentions learned by artificial intelligence through a pre-configured machine learning model, recommends each answer corresponding to each utterance intention matched to each user's chatbot question, and transmits it to each user terminal. Claim 6 delete Claim 7 delete Claim 8 delete Claim 9 A chatbot service system for identifying a user's utterance intention based on a knowledge graph, wherein, in claim 1, the knowledge graph constructed from the knowledge graph construction device is expressed in a graph form based on a knowledge relationship comprising at least one word node among related word nodes, similar word nodes, or homonymous word nodes connected to each utterance intention, including utterance intentions related to public security. Claim 10 delete Claim 11 delete Claim 12 A chatbot service system for identifying a user's utterance intention based on a knowledge graph, wherein, in claim 1, the machine learning utterance intention matching module vectorizes the nouns of each user's chatbot question using a pre-set word embedding technique to create feature vectors before inputting each user's chatbot question as input data into the generated artificial intelligence learning model. Claim 13 A chatbot service system for identifying a user's utterance intention based on a knowledge graph, wherein the word embedding technique set above utilizes the algorithm of a Term Frequency-Inverse Document Frequency (TF-IDF) model. Claim 14 A chatbot service system for identifying a user's utterance intention based on a knowledge graph, wherein, in claim 1, the machine learning utterance intention matching module learns using a multi-class classification method when machine learning a plurality of data sets.

Images