Generative artificial intelligence-based information providing device and information providing system comprising same

Abstract

A generative artificial intelligence-based information providing system according to one embodiment of the present invention comprises: a rule-based database for storing product information; a vector database for storing embedding vectors of the product information; and an information providing device configured to receive a prompt including search request information, search for at least one embedding vector associated with the search request information among the embedding vectors stored in the vector database, obtain first result information corresponding to the prompt by inputting the prompt and the searched at least one embedding vector into a generative artificial intelligence model, and if second result information corresponding to the search request information exists in the product information stored in the rule-based database, provide final result information obtained on the basis of the first result information and the second result information.

Description

Generative artificial intelligence-based information provision device and information provision system including the same

[0001] The present invention relates to a generative artificial intelligence-based information providing device and an information providing system including the same.

[0002] Generative artificial intelligence (GAI) is a type of AI technology that generates or creates new data based on given input data. Because generative AI models generate new data based on their understanding of training data, they can be used in a variety of fields, including document writing, art creation, voice synthesis, game character creation, and engineering design.

[0003] In particular, chatbots, one of the examples of generative artificial intelligence models, are systems that provide necessary information through conversations with users, and are being used in various fields for purposes such as user support, information provision, and task automation.

[0004] Rule-based chatbots provide accurate information, but simple natural language processing (NLP) technology alone struggles to accurately understand and interpret all user input. Furthermore, they are limited in their ability to respond beyond predefined question-and-answer formats. Consequently, interest is growing in developing chatbots that leverage generative AI technology to provide more creative and sophisticated responses.

[0005] However, generative AI models are experiencing the so-called hallucination phenomenon, which provides incorrect information about unlearned content, raising concerns that the accuracy and usefulness of generated content cannot be guaranteed.

[0006] To address this, a new architecture called Retrieval-Augmented Generation (RAG) was developed that integrates the Large Language Model (LLM) of a generative AI model with an information retrieval model to improve the performance of text generation.

[0007] Meanwhile, with the recent advancements in related technologies like the Internet of Things (IoT) and Artificial Intelligence (AI), smart factories are becoming a hot topic. A smart factory is a manufacturing system that operates by applying Information and Communications Technology (ICT) combined with digital automation solutions throughout the production process, from design and development to manufacturing.

[0008] One of the key requirements in operating a smart factory is to understand information that is closely related to products or businesses and is highly important and practical.

[0009] To this end, it is necessary to establish an information provision system utilizing rule-based algorithms and RAG architecture so that relevant parties, including smart factory operations managers and production planners, can accurately provide or receive necessary information.

[0010] The purpose of the present invention is to provide a generative artificial intelligence-based information providing device that provides more accurate information and an information providing system including the same.

[0011] The purpose of the present invention is to provide a generative artificial intelligence-based information providing device that complements a rule-based chatbot system and a generative artificial intelligence-based chatbot system, and an information providing system including the same.

[0012] In one embodiment of the present invention, a generative artificial intelligence-based information providing system may include: a rule-based database storing product information; a vector database storing an embedding vector of the product information; and an information providing device that receives a prompt including search request information, searches for at least one embedding vector associated with the search request information among embedding vectors stored in the vector database, inputs the prompt and the searched at least one embedding vector into a generative artificial intelligence model to obtain first result information corresponding to the prompt, and, if second result information corresponding to the search request information exists among product information stored in the rule-based database, provides final result information obtained based on the first result information and the second result information.

[0013] The above information providing device can extract product information including product attribute information and response information corresponding to the product attribute information using source information, and store the product attribute information and the response information to construct the rule-based database.

[0014] The above information providing device can construct the vector database by vector embedding the product attribute information and the response information.

[0015] The above information providing device can search for corresponding product information among product information stored in the rule-based database based on the search request information, and generate the second result information using the searched product information.

[0016] The above information providing device can generate the first result information based on the similarity between the embedding vector of the prompt and the embedding vector in the vector database using a generative artificial intelligence model.

[0017] The above information providing device may provide the final result information using the numerical information of the second result information when the numerical information included in the first result information and the numerical information included in the second result information do not match.

[0018] In a generative artificial intelligence-based information providing device according to one embodiment of the present invention, a processor may be included that receives a prompt including search request information, searches for at least one embedding vector associated with the search request information among embedding vectors of product information stored in a vector database, inputs the prompt and the searched at least one embedding vector into a generative artificial intelligence model to obtain first result information corresponding to the prompt, and, if second result information corresponding to the search request information exists among the product information stored in a rule-based database, provides final result information obtained based on the first result information and the second result information.

[0019] A generative artificial intelligence-based information providing method performed by an information providing device according to one embodiment of the present invention may include the steps of: receiving a prompt including search request information; searching for at least one embedding vector associated with the search request information among embedding vectors of product information stored in a vector database; obtaining first result information corresponding to the prompt by inputting the prompt and the searched at least one embedding vector into a generative artificial intelligence model; and providing final result information obtained based on the first result information and the second result information when there is second result information corresponding to the search request information among the product information stored in a rule-based database.

[0020] According to one embodiment of the present invention, it is possible to provide creative and more accurate responses using various types of data, and can be used in various environments such as manufacturing sites, office work, and development work.

[0021] According to one embodiment of the present invention, an independent database of product information within a smart factory can be built to provide information that is closely related to a target product and has high importance and practicality.

[0022] According to one embodiment of the present invention, the user's intention can be clearly identified through a chain structure with a large language model (LLM), thereby improving the reliability and validity of information.

[0023] FIG. 1 is a schematic diagram illustrating an information provision system according to one embodiment of the present invention.

[0024] FIG. 2 is a block diagram illustrating the configuration of an information providing device according to one embodiment of the present invention.

[0025] FIG. 3 is a diagram illustrating an operation flow chart of an information providing device according to one embodiment of the present invention.

[0026] FIG. 4 is a diagram illustrating the operation of an information provision system according to one embodiment of the present invention.

[0027] FIG. 5 is a diagram illustrating a result information generation process of an information provision system according to one embodiment of the present invention.

[0028] FIG. 6 is a diagram illustrating the operation of an information provision system according to one embodiment of the present invention.

[0029] Figure 7 is a drawing illustrating the operation of an information provision system according to one embodiment of the present invention.

[0030] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. The detailed description set forth below, together with the accompanying drawings, is intended to explain exemplary embodiments of the present invention and is not intended to represent the only embodiments in which the present invention may be practiced. In the drawings, portions irrelevant to the description may be omitted for clarity in describing the present invention, and the same reference numerals may be used throughout the specification for identical or similar components.

[0031] FIG. 1 is a schematic diagram illustrating an information provision system according to one embodiment of the present invention.

[0032] An information provision system (1) (hereinafter referred to as system (1)) according to one embodiment of the present invention may include a database (10), an information provision device (100), and a user terminal (200).

[0033] A database (10) is a device that stores information required in the process of providing information by an information providing device (100). The database (10) may be implemented by dividing into two or more databases as needed, and the implementation method is not limited to either being implemented as a separate device from the information providing device (100) or being implemented within the information providing device (100).

[0034] The information providing device (100) is a device that provides information required during the operation of a smart factory, and can be implemented as a computer, server, smart phone, tablet PC, smart pad, laptop, etc.

[0035] A user terminal (200) is a device that inputs a prompt including search request information into a chatbot program implemented to search for information about a product and receives result information from an information providing device (100), and can be implemented as a computer, a PLC (Programmable Logic Controller), a server, a smart phone, a tablet PC, a smart pad, a laptop, etc.

[0036] At this time, the user terminal (200) may be a terminal of a person in charge of a smart factory, a smart factory server, or equipment within a smart factory. The user terminal (200) may be a device that has a built-in manufacturing execution system (MES), an employee assistance program (EAP), a recipe parameter management system (RPMS), etc.

[0037] In addition, although not shown, the system (1) may further include a web server for collecting information required by the information providing device (100). The information providing device (100) may perform crawling to collect information from the web server.

[0038] In the present invention, an information provision system (1) utilizing a rule-based algorithm and RAG architecture is proposed so that relevant parties, including smart factory operation managers, sales representatives, and production planners, can more quickly and accurately supply or receive necessary information.

[0039] Hereinafter, the configuration and operation of an information providing device (100) according to one embodiment of the present invention will be specifically described with reference to the drawings.

[0040] FIG. 2 is a block diagram illustrating the configuration of an information providing device according to one embodiment of the present invention.

[0041] An information providing device (100) according to one embodiment of the present invention may include an input unit (110), a communication unit (120), a display unit (130), a storage unit (140), and a processor (150).

[0042] The input unit (110) generates input data in response to user input of the information providing device (100). For example, the user input may be a user input that initiates the operation of the information providing device (100), a user input that stores product information, a user input required to build a database, etc. In addition, the input may be applied without limitation to a user input required to transmit result information corresponding to search request information to a user terminal (200).

[0043] The input unit (110) includes at least one input means. The input unit (110) may include a keyboard, a key pad, a dome switch, a touch panel, a touch key, a mouse, a menu button, etc.

[0044] The communication unit (120) can perform communication with external devices such as a user terminal (200) and a web server to transmit and receive search request information, product information, product attribute information and response information, numerical information, result information, etc.

[0045] To this end, the communication unit (120) can perform wireless communication such as 5G (5th generation communication), LTE-A (Long Term Evolution-Advanced), LTE (Long Term Evolution), Wi-Fi (Wireless Fidelity), Bluetooth, or wired communication such as LAN (Local Area Network), WAN (Wide Area Network), and power line communication.

[0046] The display unit (130) displays display data according to the operation of the information providing device (100). The display unit (130) can display a screen displaying search request information, a screen displaying product information, a screen displaying result information, a screen receiving user input, etc.

[0047] The display unit (130) includes a liquid crystal display (LCD), a light emitting diode (LED) display, an organic light emitting diode (OLED) display, a micro electro mechanical systems (MEMS) display, and an electronic paper display. The display unit (130) may be implemented as a touch screen by being combined with the input unit (110).

[0048] The storage unit (140) stores the operation programs of the information providing device (100). The storage unit (140) includes a non-volatile storage that can preserve data (information) regardless of whether power is supplied, and a volatile memory that loads data to be processed by the processor (150) and cannot preserve data if power is not supplied. The storage includes a flash memory, a hard-disc drive (HDD), a solid-state drive (SSD), a read-only memory (ROM), etc., and the memory includes a buffer, a random access memory (RAM), etc.

[0049] The storage unit (140) may be implemented by including any one of the databases (10) as described above. The storage unit (140) may store search request information, product information, product attribute information and response information, numerical information, result information, generative artificial intelligence models, etc., and may store operation programs required in the process of receiving prompts, collecting and searching product information, vector embedding, building / running generative artificial intelligence models, obtaining / providing result information, etc.

[0050] The processor (150) can control at least one other component (e.g., hardware or software component) of the information providing device (100) by executing software such as a program, and can perform various data processing or operations.

[0051] A processor (150) according to one embodiment of the present invention receives a prompt including search request information, searches for at least one embedding vector associated with the search request information among embedding vectors stored in the vector database, inputs the prompt and the searched at least one embedding vector into a generative artificial intelligence model to obtain first result information corresponding to the prompt, and, if second result information corresponding to the search request information exists among product information stored in the rule-based database, provides final result information obtained based on the first result information and the second result information.

[0052] At this time, the processor (150) may build a database and / or a generative artificial intelligence model, or may receive and store a previously built database and / or a generative artificial intelligence model from the outside and use it, but is not limited to either one.

[0053] Meanwhile, the processor (150) may perform at least a portion of the data analysis, processing, and result information generation for performing the above operations using at least one of a machine learning, neural network, or deep learning algorithm as a rule-based or artificial intelligence (AI) algorithm. Examples of the neural network may include models such as a convolutional neural network (CNN), a deep neural network (DNN), a recurrent neural network (RNN), and a transformer.

[0054] FIG. 3 is a diagram illustrating an operation flow chart of an information providing device according to one embodiment of the present invention.

[0055] A processor (150) according to one embodiment of the present invention can receive a prompt including search request information (S10).

[0056] Search request information is the basic information required to perform searches at the request of internal staff or clients. Search request information may include at least one of the following: product type, product manufacturer, and product model name.

[0057] The processor (150) can receive a prompt containing search request information from a user terminal (200) via a chatbot program. The prompt is a query input into a generative artificial intelligence model, for example, "Tell me about measuring instrument product A." The prompt can be a voice input in addition to text input, and the prompt format is not limited to any one.

[0058] A processor (150) according to one embodiment of the present invention can search for at least one embedding vector associated with search request information among the embedding vectors stored in a vector database (S20).

[0059] As described above, the database (10) of FIG. 1 may be implemented as two or more databases as needed. For example, the database (10) may include a rule-based database and a vector database.

[0060] The processor (150) can receive and store a pre-built database (10) from the outside and use it, and the process of the processor (150) building the database (10) is described with reference to FIG. 4.

[0061] Meanwhile, as described above, the present invention adopts the RAG architecture to improve the performance of text generation of a generative artificial intelligence model, and the RAG architecture is largely composed of a generative artificial intelligence model and a vector database.

[0062] A generative AI model can be a large-scale language model, and a vector database is constructed by vector-embedding information that serves as a reference for the large-scale language model to provide answers to queries. This reference information can be product information or information retrieved based on a prompt. This will be described in detail with reference to Figure 6.

[0063] A processor (150) according to one embodiment of the present invention can input a prompt and at least one searched embedding vector into a generative artificial intelligence model to obtain first result information corresponding to the prompt (S30).

[0064] The processor (150) can generate first result information based on the similarity between the embedding vector of the prompt and the embedding vector in the vector database. The similarity can be determined by a distance measurement method such as Euclidean distance or cosine similarity, and similarity can also be measured using various other known methods.

[0065] The processor (150) can generate first result information using embedding vectors with the highest similarity through a generative artificial intelligence model.

[0066] For example, for the prompt "Tell me about product A of measuring instrument", the first result information can be obtained as "Product A of Company B's measuring instrument consists of two main measuring modules and 25 branch measuring modules. Branch measuring modules are divided into single-phase and three-phase, and there are four types of single-phase direct connection / through-hole (100A, 200A, 325A, 500A) each, and five types of three-phase direct connection / through-hole (5A (through-hole), 30A, 100A, 125A, 250A) each. There is one type of panel temperature module, and the busbar temperature module is divided into single-phase and three-phase, and there are direct connection 32EGR, 2P 100A for single-phase, and direct connection 3P 100A, 125A, 250A for three-phase."

[0067] However, since the first result information obtained through the generative artificial intelligence model is obtained by combining embedding vectors, information may be omitted or errors may occur during the process of generating the first result information.

[0068] Among the first result information, if it contains important information, such as numerical information, it needs to be re-verified. To address this, the present invention utilizes a rule-based algorithm in parallel, as follows.

[0069] According to one embodiment of the present invention, the processor (150) can provide final result information obtained based on the first result information and the second result information when there is second result information corresponding to search request information among product information stored in a rule-based database (S40).

[0070] A rule-based database is a database that stores product information. Product information can be stored in text form and can include product attribute information and response information corresponding to the product attribute information.

[0071] The processor (150) can search for corresponding product information among product information stored in a rule-based database based on search request information, and can generate second result information using the searched product information.

[0072] For example, for the prompt "Tell me about product A of the instrument",

[0073] "1. A product composition - 2 main measurement modules, 25 branch measurement modules,

[0074] 2. Branch measurement module

[0075] - Single phase: 4 types each of direct connection / through connection (30A, 100A, 125A, 250A)

[0076] - Three phase: Direct connection - 4 types, through connection - 5 types (5A (through connection), 30A, 100A, 125A, 250A)

[0077] 3. Panel temperature module

[0078] 4. Busbar temperature module

[0079] - Single phase: Direct connection 32EGR, 2P 100A

[0080] - Three phase: You can obtain the second result information called "Direct connection 3P 100A, 125A, 250A, 400A, 630A".

[0081] The processor (150) can generate final result information by comparing the first result information with the second result information. At this time, the first result information is obtained using a vector database and can highly reflect the user's query intent and query content. The second result information is obtained using a rule-based database and can relatively accurately reflect the numerical information included in the product information. Therefore, if the numerical information included in the first result information and the numerical information included in the second result information do not match, the processor (150) can generate final result information based on the first result information but using the numerical information of the second result information.

[0082] For example, for the prompt "Tell me about product A of the measuring instrument," reflecting both the first and second result information, the final result is "Product A of Company B's measuring instrument consists of two main measuring modules and 25 branch measuring modules. Branch measuring modules are divided into single-phase and three-phase, and there are four types of single-phase direct / through (30A, 100A, 125A, 250A) each, and four types of three-phase direct / through (5A (through), 30A, 100A, 125A, 250A) each. There is one type of panel temperature module, and the busbar temperature module is divided into single-phase and three-phase, and there are direct 32EGR, 2P 100A for single-phase, and direct 3P 100A, 125A, 250A, 400A, 630A for three-phase." Information can be generated.

[0083] In another embodiment, the processor (150) may generate final result information by reflecting second result information obtained using a rule-based database before generating first result information.

[0084] According to one embodiment of the present invention, it is possible to provide creative and more accurate responses using various types of data, and can be used in various environments such as manufacturing sites, office work, and development work.

[0085] FIG. 4 is a diagram illustrating the operation of an information provision system according to one embodiment of the present invention.

[0086] Figure 4 specifically describes the database construction process of the information provision device (100). Any content that overlaps with the content described with reference to Figure 3 will be omitted.

[0087] First, the processor (150) can extract product attribute information and response information corresponding to the product attribute information using source information.

[0088] Source information refers to raw data related to a product. For example, source information can be product-related data such as product manuals, operational information, failure information, and power / process information. Source information can be applied in a variety of ways. The method or path for collecting source information is not limited to a single source information collection method.

[0089] The processor (150) can preprocess source information according to a rule-based algorithm to extract product attribute information and response information corresponding to the product attribute information.

[0090] Product attribute information refers to all items related to the product, and response information is information that includes a response to be given when a query regarding the corresponding item is received. For example, the processor (150) can extract product attribute information including the standard usage environment (temperature, humidity, usage location, etc.), rated conditions (frequency, input range of control power, power consumption, voltage input range, current input range, etc.) of Product A through source information, which is a catalog for Product A. In addition, the processor (150) can include response information for each product attribute information, for example, for temperature among the standard usage environments, “Normal usage temperature: minimum (-20°C), maximum (60°C), storage temperature: minimum (-30°C), maximum (70°C)”, and for humidity, “Maximum humidity (80%) (but no condensation)”.

[0091] The processor (150) can build a rule-based database (11) by storing product attribute information and response information, and can build a vector database (12) by vector-embedding product attribute information and response information.

[0092] According to one embodiment of the present invention, an independent database of product information within a smart factory can be built to provide information that is closely related to a target product and has high importance and practicality.

[0093] FIG. 5 is a diagram illustrating a result information generation process of an information provision system according to one embodiment of the present invention.

[0094] The information providing device (100) can select the most suitable result information (first result information) for the prompt (520) by performing embedding vector search-importance evaluation-final selection based on a generative artificial intelligence model (510).

[0095] First, the information providing device (100) can search for related embedding vectors in the vector database (13) based on the prompt (520). The information providing device (100) can evaluate importance based on the similarity between the embedding vector vectorized from the prompt (520) and the embedding vector in the vector database (13). The information providing device (100) can input the embedding vectors with the highest importance into the generative artificial intelligence model (510) to obtain first result information. At this time, the processor (150) can also obtain first result information by forming an enhanced prompt including both the embedding vector of the prompt (520) and the embedding vector in the vector database (12) and inputting the enhanced prompt into the generative artificial intelligence model (510).

[0096] According to one embodiment of the present invention, the user's intention can be clearly identified through a chain structure with a large language model (LLM), thereby improving the reliability and validity of information.

[0097] FIG. 6 is a diagram illustrating the operation of an information provision system according to one embodiment of the present invention.

[0098] The information providing device (100) can transmit final result information to the user terminal (200) based on the first result information obtained using the vector database (12) and the second result information obtained using the rule-based database (11).

[0099] Figure 7 is a drawing illustrating the operation of an information provision system according to one embodiment of the present invention.

[0100] Referring to FIG. 7, a user terminal (200) can transmit a prompt such as "At what temperature can the measuring instrument product A be used?" to an information providing device (100). The information providing device (100) can vectorize the prompt to search for related content in a rule-based database (11) and a vector database (12), and retrieve information regarding usage conditions (e.g., temperature, humidity, and usage location).

[0101] The information providing device (100) can generate a response using a RAG-based large-scale language model. For example, it can generate a response such as 1) "Measuring instrument product A can be used at temperatures ranging from -20 to 60°C" or 2) "Measuring instrument product A can be used at temperatures ranging from -25 to 65°C."

[0102] The information provision device (100) can compare the response with a rule-based database (11) and supplement the response. This is a process to verify the accuracy of the response and check for missing information.

[0103] For example, if there is a response for use temperature but information about storage temperature is missing or incorrect, the information providing device (100) can generate a supplemented response by utilizing the rule-based database (11).

[0104] For example, the generated response 1) above can be supplemented with "1) "The normal operating temperature of product A of measuring instrument is -20℃ to 60℃, and the storage temperature is -30℃ to 70℃" (added content) to create a response. In addition, the generated response 2) above can be supplemented with "2) "The normal operating temperature of product A of measuring instrument is -20℃ to 60℃, and the storage temperature is -30℃ to 70℃" (added content and modified content) to create a response.

[0105] Finally, the information providing device (100) can provide a supplemented response to the user terminal (200).

Claims

1. In the information provision system based on generative artificial intelligence, A rule-based database that stores product information; A vector database storing an embedding vector of the above product information; and Receive a prompt containing search request information, Retrieving at least one embedding vector associated with the search request information among the embedding vectors stored in the above vector database, Inputting the above prompt and at least one searched embedding vector into the generative artificial intelligence model to obtain first result information corresponding to the prompt, An information providing system including an information providing device that provides final result information obtained based on the first result information and the second result information when there is second result information corresponding to the search request information among the product information stored in the rule-based database.

2. In paragraph 1, The above information providing device, Extracting product information including product attribute information and response information corresponding to the product attribute information using source information, An information providing system that constructs a rule-based database by storing the above product attribute information and the above response information.

3. In paragraph 2, The above information providing device, An information providing system that constructs a vector database by vector embedding the above product attribute information and the above response information.

4. In paragraph 1, The above information providing device, Based on the above search request information, the corresponding product information is searched among the product information stored in the rule-based database, An information providing system that generates the second result information using the product information searched above.

5. In paragraph 1, The above information providing device, An information providing system that generates the first result information based on the similarity between the embedding vector of the prompt and the embedding vector in the vector database using a generative artificial intelligence model.

6. In paragraph 1, The above information providing device, An information providing system that provides the final result information using the numerical information of the second result information when the numerical information included in the first result information and the numerical information included in the second result information do not match.

7. In a generative artificial intelligence-based information providing device, Receive a prompt containing search request information, Retrieving at least one embedding vector associated with the search request information among the embedding vectors of product information stored in the vector database, Inputting the above prompt and at least one searched embedding vector into the generative artificial intelligence model to obtain first result information corresponding to the prompt, An information providing device including a processor that provides final result information obtained based on the first result information and the second result information when there is second result information corresponding to the search request information among the product information stored in the rule-based database.

8. In paragraph 7, The above processor, Extracting product information including product attribute information and response information corresponding to the product attribute information using source information, An information providing device that constructs the rule-based database by storing the above product attribute information and the above response information.

9. In paragraph 8, The above processor, An information providing device that constructs a vector database by vector embedding the above product attribute information and the above response information.

10. In paragraph 7, The above processor, Based on the above search request information, the corresponding product information is searched among the product information stored in the rule-based database, An information providing device that generates the second result information using the product information searched above.

11. In paragraph 7, The above processor, An information providing device that generates the first result information based on the similarity between the embedding vector of the prompt and the embedding vector in the vector database using a generative artificial intelligence model.

12. In paragraph 7, The above processor, An information providing device that provides the final result information using the numerical information of the second result information when the numerical information included in the first result information and the numerical information included in the second result information do not match.

13. In a method for providing information based on artificial intelligence, which is performed by an information providing device, A step of receiving a prompt containing search request information; A step of searching for at least one embedding vector associated with the search request information among the embedding vectors of product information stored in a vector database; A step of inputting the prompt and at least one searched embedding vector into a generative artificial intelligence model to obtain first result information corresponding to the prompt; An information providing method comprising the step of providing final result information obtained based on the first result information and the second result information, when second result information corresponding to the search request information exists among the product information stored in the rule-based database.

Images