Artificial intelligence device, method of operating artificial intelligence device, and non-transitory storage medium

Abstract

This disclosure relates to artificial intelligence devices, methods of operating artificial intelligence devices, and non-transitory storage media. An artificial intelligence device may include a display and at least one processor configured to receive user input based on a model code identifying a product in response to a search for user input failing to match one of a plurality of pre-stored model codes, obtain a first product search result based on the similarity between the user input and one of the plurality of pre-stored model codes, and display the first product search result.

Description

Technical Field

[0001] This invention relates to an artificial intelligence device, and more specifically, to an artificial intelligence device configured to provide product search services. Background Technology

[0002] Search term suggestion service is a feature that automatically suggests relevant search terms as the user begins to type a search term, or recommends correct search terms if the user has entered a misspelled or incorrect search term.

[0003] Search term suggestion services are primarily provided by search engines, shopping websites, and content platforms, and help users quickly and accurately find the information they want.

[0004] When Korean is incorrectly entered into English, when a typo is entered, when spelling is incorrect, when searching for synonyms, or when a name or term has been changed, the existing search term suggestion service provides the function of recommending the correct word.

[0005] However, existing search term suggestion services may only suggest general natural language search terms and have the problem of not working in specific domains (such as identifying model names of specific products), which may involve incorrectly entered numbers and digits instead of text words and sentences.

[0006] Furthermore, existing search term suggestion services use known natural language processing (NLP) models, posing a security risk to user-inputted search terms. Additionally, there is a need for artificial intelligence that can determine a user's true meaning and desired outcome, even if the user incorrectly enters a model number in their query. Summary of the Invention

[0007] The purpose of this disclosure may be to effectively present the correct model code or a similar model code to the user.

[0008] Another objective of this disclosure is to improve the user's search experience and improve product accessibility.

[0009] Another object of this disclosure is to enable users to easily search for products that suit their preferences through interactive business services.

[0010] The purpose of this disclosure is to make it easy to search for products that reflect attributes, even if only part of the model code is entered.

[0011] An artificial intelligence device according to an embodiment of the present disclosure may include a display and at least one processor, the at least one processor being configured to receive a model code identifying a product via user input, and if a search for the model code fails, to obtain a first product search result based on the similarity between the model code and a pre-stored product model code, and to display the obtained first product search result on the display.

[0012] A method for operating an artificial intelligence device according to an embodiment of the present disclosure may include: if a search for a model code fails, receiving a model code identifying a product by user input, obtaining a first product search result based on the similarity between the model code and a pre-stored product model code, and displaying the obtained first product search result.

[0013] According to embodiments of the present disclosure, a computer-readable non-transitory storage medium having a program recorded thereon for performing a method of operating an artificial intelligence device, wherein the method includes: if a search for the model code fails, receiving a model code identifying a product by user input, obtaining a first product search result based on the similarity between the model code and a pre-stored product model code, and displaying the obtained first product search result.

[0014] According to the embodiments of this disclosure, the problem of incorrect model code input can be solved more effectively, and the user can be recommended a model that is most similar in function to the input model code.

[0015] According to embodiments of this disclosure, products that are most similar in function to the desired product can be easily searched, thereby improving the experience in the product search and selection process.

[0016] According to embodiments of this disclosure, the searched products are categorized based on their functions, thus allowing users to find products with the desired specific functions more quickly and accurately.

[0017] According to embodiments of this disclosure, users can efficiently search for products that match their preferences through interactive business services, and even if the product model code is entered incorrectly or only partially, similar products can be recommended.

[0018] According to embodiments of this disclosure, even if a user enters only part of the model code or enters part of the model code incorrectly, the user can effectively search for products that reflect the desired attributes through the tag list and product images.

[0019] According to embodiments of this disclosure, even if a user only enters a portion of the model code, the user can easily search for products that reflect desired attributes through a chatbot application. Attached Figure Description

[0020] The above and other objects, features and advantages of this disclosure will become more apparent to those skilled in the art from the following detailed description of exemplary embodiments with reference to the accompanying drawings, which are briefly described below.

[0021] Figure 1 This is a block diagram illustrating the elements of an artificial intelligence device according to embodiments of the present disclosure.

[0022] Figure 2 This is a diagram illustrating the configuration of an artificial intelligence server according to an embodiment of the present disclosure.

[0023] Figure 3 This is a sequence diagram illustrating the operation method of an artificial intelligence system according to embodiments of the present disclosure.

[0024] Figures 4 to 8 This is a diagram illustrating a process for providing a first product search result when a model code is incorrectly entered, according to an embodiment of the present disclosure. The first product search result includes information about products corresponding to model codes similar to the incorrectly entered model code.

[0025] Figures 9A to 9C This is a diagram illustrating a process, according to an embodiment of the present disclosure, of providing product information with a product model code similar to the incorrectly entered model code when an incorrect model code is entered.

[0026] Figure 10 and Figure 11 This is a diagram illustrating a process for obtaining a second product search result according to an embodiment of the present disclosure.

[0027] Figures 12A to 12C This is a diagram illustrating an example of providing second product search results in response to a search request for a model code correctly entered by a user, according to an embodiment of this disclosure.

[0028] Figures 13A to 13L This is a diagram illustrating a process for providing an interactive business service for product search according to an embodiment of the present disclosure.

[0029] Figure 14A and Figure 14B This is a diagram illustrating the process for determining detailed entries for a product when only a portion of the model code is entered into the search window according to an embodiment of the present disclosure.

[0030] Figure 15 This is a diagram illustrating a method of operating an artificial intelligence device according to an embodiment of the present disclosure. Detailed Implementation

[0031] Artificial intelligence refers to the field of research on artificial intelligence or the methodology of creating artificial intelligence, while machine learning refers to the field of defining the various problems dealt with in the field of artificial intelligence and studying the methodology of solving these problems.

[0032] Machine learning is also defined as an algorithm that improves task performance through consistent experience.

[0033] Artificial neural networks (ANNs) are a type of model used in machine learning. They can refer to a holistic model with problem-solving capabilities, which is formed by artificial neurons (nodes) combining through synapses to form a network.

[0034] Artificial neural networks can be defined by the connection patterns between neurons in different layers, the learning process for updating model parameters, and the activation function that produces output values.

[0035] An artificial neural network may include an input layer, an output layer, and optionally one or more hidden layers. Each layer may include one or more neurons, and the artificial neural network may include synapses connecting the neurons. In an artificial neural network, each neuron can output the input signal, weights, and activation function values ​​used for bias, which are input through the synapse.

[0036] Model parameters refer to parameters determined through learning, including the weights of synaptic connections and the biases of neurons. Hyperparameters refer to parameters set before learning in a machine learning algorithm, including the learning rate, number of repetitions, mini-batch size, and initialization function.

[0037] The purpose of learning artificial neural networks can be viewed as determining the model parameters that minimize the loss function. During the learning process of an artificial neural network, the loss function can serve as an indicator for determining the optimal model parameters.

[0038] Depending on the learning method, machine learning can be divided into supervised learning, unsupervised learning, and reinforcement learning.

[0039] Supervised learning refers to the method of training an artificial neural network with labels for given learning data. The labels can represent the correct answer (or result value) inferred by the artificial neural network when the learning data is input into the artificial neural network.

[0040] Unsupervised learning refers to the method of training artificial neural networks without providing labels for training data.

[0041] Reinforcement learning can refer to a learning method in which an agent defined in an environment learns to select actions or sequences of actions that maximize the cumulative reward in each state.

[0042] In artificial neural networks, machine learning implemented using deep neural networks (DNNs) that include multiple hidden layers is also called deep learning, and deep learning is a part of machine learning.

[0043] In the following text, machine learning is used to include deep learning.

[0044] The following embodiments may be combined or integrated with each other in part or in whole, and may be connected and operated in technically different ways. Embodiments may be performed independently or in combination. Furthermore, the term "capable" as used herein includes all the meanings and definitions of the term "can".

[0045] Figure 1 This is a block diagram illustrating the elements of an artificial intelligence device according to embodiments of the present disclosure.

[0046] Artificial intelligence device 100 can be implemented as a fixed or mobile device, such as a television, projector, mobile phone, smartphone, desktop computer, laptop computer, digital broadcasting terminal, PDA (personal digital assistant), PMP (portable multimedia player), navigation, tablet PC, wearable device and set-top box (STB), DMB receiver, radio, washing machine, refrigerator, desktop computer, digital signage, robot, vehicle, etc.

[0047] refer to Figure 1 The artificial intelligence device 100 may include a communication interface 110, an input interface 120, a learning processor 130, a sensor 140, an output interface 150, a memory 170, and a processor 180.

[0048] The communication interface 110 can use wired or wireless communication technologies to send and receive data with external devices such as another artificial intelligence device or AI server 200. For example, the communication interface 110 can send and receive sensor information, user input, learning models, and control signals with external devices.

[0049] The communication technologies used in the communication interface 110 include Global System for Mobile Communications (GSM), Code Division Multiple Access (CDMA), Long Term Evolution (LTE), 5G, Wireless LAN (WLAN), Wi-Fi, Bluetooth, RFID (Radio Frequency Identification), Infrared Data Association (IrDA), ZigBee, and NFC (Near Field Communication).

[0050] Input interface 120 can obtain various types of data.

[0051] The input interface 120 may include a camera 121 for capturing images, a microphone 122 for receiving audio signals, and a user input interface 123 for receiving information from a user.

[0052] The camera 121 or microphone 122 is regarded as a sensor, and the signal obtained from the camera 121 or microphone 122 can be called sensing data or sensor information.

[0053] Input interface 120 can obtain training data for model learning and input data to be used when obtaining output using the learned model. Input interface 120 can also obtain unprocessed input data, in which case processor 180 or learning processor 130 can extract input features by preprocessing the input data.

[0054] Camera 121 processes image frames, such as still images or moving images, acquired by the image sensor in video call mode or shooting mode. The processed image frames can be displayed on display 151 or stored in memory 170.

[0055] Microphone 122 processes external acoustic signals into electronic speech data. Depending on the function (or application) being performed by the artificial intelligence device 100, the processed speech data can be utilized in various ways. Simultaneously, various noise removal or cancellation algorithms can be applied to microphone 122 to remove noise generated during the reception of external acoustic signals.

[0056] User input interface 123 is used to receive information from the user. When information is input through user input interface 123, processor 180 can control the operation of artificial intelligence device 100 to correspond to the input information.

[0057] User input interface 123 is a mechanical input device (or mechanical key, such as a button, dome switch, scroll wheel or micro switch, etc. located on the front / back or side of the artificial intelligence device 100) and a touch input device.

[0058] As an example, touch input can be virtual keys, soft keys, or visual keys displayed on the touchscreen via software processing, or touch keys located in parts outside the touchscreen.

[0059] The learning processor 130 can use training data to train a model composed of an artificial neural network. The learned artificial neural network can be called a learning model. The learning model can be used to infer the result value of new input data other than the training data, and the inferred value can be used as the basis for decisions to perform operations.

[0060] The learning processor 130 can perform AI processing together with the learning processor 240 of the AI ​​server 200.

[0061] The learning processor 130 may include memory integrated into or implemented in the artificial intelligence device 100. The learning processor 130 may be implemented using memory 170, external memory directly connected to the artificial intelligence device 100, or memory stored in an external device.

[0062] Sensor 140 can use various sensors to obtain at least one of the following: internal information of artificial intelligence device 100, information about the surrounding environment of artificial intelligence device 100, or user information.

[0063] Sensor 140 may include at least one of a proximity sensor, a lighting sensor, an acceleration sensor, a magnetic sensor, a gyroscope sensor, an inertial sensor, an RGB sensor, an IR sensor, a fingerprint sensor, an ultrasonic sensor, an optical sensor, a microphone, a lidar sensor, or a radar sensor.

[0064] Output interface 150 can generate outputs related to vision, hearing, or touch.

[0065] The output interface 150 may include a display 151 for outputting images, an audio output interface 152 for outputting audio, a tactile device 153 for outputting tactile information, and an optical output interface 154 for outputting light.

[0066] Display 151 displays (outputs) information processed by artificial intelligence device 100. For example, display 151 may display execution screen information of an application running on artificial intelligence device 100, or display user interface (UI) and graphical user interface (GUI) information based on the execution screen information.

[0067] The display 151 can be implemented as a touchscreen by forming an interlayer structure or by integrating it with a touch sensor. The touchscreen serves as a user input interface 123, providing an input interface between the artificial intelligence device 100 and the user, and can simultaneously provide an output interface between the artificial intelligence device 100 and the user.

[0068] The audio output interface 152 can output audio data received from the communication interface 110 or stored in the memory 170 in call signal receiving, call mode or recording mode, voice recognition mode, broadcast receiving mode, etc.

[0069] The audio output interface 152 may include at least one of a receiver, a speaker, or a buzzer.

[0070] The tactile device 153 produces various tactile effects that the user can perceive. A representative example of a tactile effect produced by the tactile device 153 may be vibration.

[0071] The light output interface 154 uses light from a light source in the artificial intelligence device 100 to output a signal to notify that an event has occurred. Examples of events occurring in the artificial intelligence device 100 may include receiving a message, receiving a call signal, a missed call, an alarm, a calendar notification, receiving an email, receiving information via an application, etc.

[0072] The memory 170 can store data that supports various functions of the artificial intelligence device 100. For example, the memory 170 can store input data, learning data, learning models, learning history, etc., obtained from the input interface 120.

[0073] The processor 180 can determine at least one executable operation of the artificial intelligence device 100 based on information determined or generated using data analysis algorithms or machine learning algorithms.

[0074] The processor 180 can control the components of the artificial intelligence device 100 to perform the determined operations.

[0075] For this purpose, processor 180 can request, search, receive, or utilize data from learning processor 130 or memory 170, and can control the elements of artificial intelligence device 100 to perform at least one predicted or determined operation in an executable operation.

[0076] If a link with an external device is desired to perform a defined operation, the processor 180 may generate control signals to control the external device and send the generated control signals to the external device.

[0077] The processor 180 can obtain intent information for user input and determine the user's request based on the obtained intent information.

[0078] The processor 180 may use at least one of an STT (speech-to-text) engine for converting speech input into a string or a natural language processing (NLP) engine for obtaining intent information corresponding to the user input.

[0079] At least one of the STT engine and the NLP engine may consist of at least a portion of an artificial neural network learned according to a machine learning algorithm. Furthermore, at least one of the STT engine or the NLP engine may be learned by the learning processor 130, by the learning processor 240 of the AI ​​server 200, or through its distributed processing.

[0080] Processor 180 collects historical information, including feedback from user actions on AI device 100, and stores this historical information in memory 170, learning processor 130, or AI server 200, etc. This historical information can be sent to external devices. The collected historical information can be used to update the learning model.

[0081] The processor 180 can control at least some of the components of the artificial intelligence device 100 to run applications stored in the memory 170.

[0082] The processor 180 can operate two or more components included in the artificial intelligence device 100 to run applications.

[0083] Figure 2 This is a diagram illustrating the configuration of an artificial intelligence server according to an embodiment of the present disclosure.

[0084] refer to Figure 2 AI server 200 can refer to a device that uses machine learning algorithms or trains artificial neural networks using learned artificial neural networks.

[0085] AI server 200 can consist of multiple servers to perform distributed processing and can be defined as a 5G network. AI server 200 can be included as part of artificial intelligence device 100 and can perform at least a portion of AI processing.

[0086] AI server 200 may include communication interface 210, memory 230, learning processor 240 and processor 260.

[0087] The communication interface 210 can send and receive data with external devices such as the artificial intelligence device 100.

[0088] The memory 230 may include a model memory 231. The model memory 231 may store a model (or artificial neural network 231a) that is being trained or has been learned by the learning processor 240.

[0089] The learning processor 240 can use training data to train the artificial neural network 231a. The learning model can be installed on the AI ​​server 200 of the artificial neural network or on an external device such as the artificial intelligence device 100.

[0090] The learning model can be implemented in hardware, software, or a combination of hardware and software. When part or all of the learning model is implemented as software, one or more instructions constituting the learning model can be stored in memory 230.

[0091] The processor 260 can use the learning model to infer the result value of new input data and generate a response or control command based on the inferred result value.

[0092] Figure 3 This is a sequence diagram illustrating the operation method of an artificial intelligence system according to embodiments of the present disclosure.

[0093] The artificial intelligence system may include an artificial intelligence device 100 and an artificial intelligence server 200.

[0094] In the following text, processor 180 may consist of a single processor or multiple processors.

[0095] refer to Figure 3 The processor 180 of the artificial intelligence device 100 can obtain or receive a model code through user input (S301).

[0096] The model code can be a code used to identify a product. The product can be a household appliance, such as a refrigerator, washing machine, air purifier, or air conditioner.

[0097] A model code can be a unique identifier string that identifies a product. A model code can be referred to as a unique identifier string.

[0098] User input can be keyboard input or voice input, but the implementation method is not limited to these. For example, the model code can be a part of an image or a QR code, etc.

[0099] In one implementation, the processor 180 may receive a model code entered through a search term input window included on a webpage.

[0100] In another embodiment, processor 180 may receive a model code input via the execution screen of a chatbot application. Artificial intelligence device 100 may provide interactive business services. Conversational business services may be services that allow users to explore products or make purchases through messaging applications, chatbot applications, voice assistants, etc.

[0101] The processor 180 can send a search request for the obtained model code to the artificial intelligence server 200 through the communication interface 110 (S303).

[0102] As an example, processor 180 can obtain user input as a search request by selecting a search button located on one side of a search term input window included on a webpage.

[0103] As another example, processor 180 can obtain search requests via user voice commands.

[0104] The processor 180 may include the model code entered in the search request and send the model code to the artificial intelligence server 200 via the communication interface 110.

[0105] The processor 260 of the AI ​​server 200 can perform a search based on the received search request and determine whether the search for the model code has failed (S305).

[0106] In one implementation, if the obtained model code is not stored, the processor 260 can determine that the search for the model code has failed, and if the obtained model code is stored, the processor 260 can determine that the search for the model code has succeeded.

[0107] If the model code received from the AI ​​device 100 is not stored in the memory 230, the AI ​​server 200 can send a search failure notification to the AI ​​device 100 indicating that the search failed. The processor 180 can determine that the search for the model code has failed based on the search failure notification received from the AI ​​server 200.

[0108] If the model code received from the AI ​​device 100 is stored in the memory 230, the AI ​​server 200 can send a search success notification to the AI ​​device 100 indicating that the search was successful. The processor 180 can determine that the search for the model code was successful based on the search success notification received from the AI ​​server 200.

[0109] If it is determined that the search for the obtained model code has failed, the processor 260 of the AI ​​server 200 can obtain a first product search result based on the similarity between the obtained model code and the previously stored product model code (S307).

[0110] If it is determined that the obtained model code is not stored, the processor 260 can determine that the search code was entered incorrectly.

[0111] If it is determined that the search for the obtained model code fails, the processor 260 can extract one or more product model codes that are similar to the obtained model code from multiple stored product model codes, and obtain product information about one or more products that identify the extracted one or more product model codes as the first product search result.

[0112] The memory 230 can store multiple product model codes, multiple product embedding vectors that match each of the multiple product model codes, and multiple product information that match each of the multiple product model codes.

[0113] Processor 260 can convert model codes into embedding vectors and obtain a first product search result based on the similarity between the converted embedding vectors and previously stored product embedding vectors. Processor 180 can extract one or more product embedding vectors from a plurality of pre-stored product embedding vectors, wherein the similarity between the product embedding vectors and the converted embedding vectors is equal to or greater than a certain similarity level. According to one embodiment, the similarity level may be a predetermined value, but the embodiment is not limited thereto. According to another embodiment, the similarity level may be dynamically adjusted, for example, based on user input preferences.

[0114] The processor 260 can obtain one or more product information corresponding to the extracted one or more product embedding vectors as the first product search result. The memory 230 of the artificial intelligence server 200 can be referred to as the database 230.

[0115] The first product search result can include product information about similar products with model codes that are similar to those entered incorrectly by the user.

[0116] Product information may include at least one of the following: product model name, product name, product manufacturer, product capacity, product power consumption, product price, product function, or product purchase location.

[0117] The processor 260 of the AI ​​server 200 can send the first product search result obtained through the communication interface 110 to the artificial intelligence device 100 (S309).

[0118] The processor 180 of the artificial intelligence device 100 can display the received first product search result on the display 151 (S311).

[0119] The processor 180 can receive the first product search result in response to the sending request of the search request, and can display the received first product search result on the display 151.

[0120] In another embodiment, the processor 180 can output the received first product search result as voice via the audio output interface 152.

[0121] In another embodiment, the processor 180 can send the received first product search result to the user terminal.

[0122] Simultaneously, when the processor 260 of the AI ​​server 200 determines that the search for the obtained model code is successful, the processor 260 can obtain a second product search result based on the functional similarity of the product with the searched model code (S313). For example, even if the obtained model code successfully matches a stored model or product, additional similar options can be provided to the user.

[0123] Processor 260 can generate a knowledge graph based on the product specification information obtained from the model code. The product specification information may include multiple specification elements. Each of the multiple specification elements can be an element representing the product specification.

[0124] For example, if the product is a vertical display device, several specification elements may include screen size, speaker output, speaker channels, product line, stand type, and resolution. Product specification information can vary depending on the product type, and even within the same type, specifications can differ.

[0125] The memory 230 can store product specification information that matches the product model code and a knowledge graph that matches the product specification information.

[0126] Processor 260 can generate unit embedding vectors for each specification element of a product from a knowledge graph through the generated knowledge graph embedding model, and can generate a final embedding vector based on the generated unit embedding vectors.

[0127] The processor 260 can obtain product information corresponding to the final embedded vector as a second product search result. The second product search result may include product information about similar products, which include products with similar functions to those of the product that matches the model code correctly entered by the user.

[0128] The second product search results can also include product information based on the model code correctly entered by the user.

[0129] The processor 260 of the AI ​​server 200 can send the second product search results obtained through the communication interface 110 to the artificial intelligence device 100 (S315).

[0130] The processor 180 of the artificial intelligence device 100 can display the received second product search results on the display 151 (S317).

[0131] In another embodiment, the processor 180 can output the received second product search results as voice via the audio output interface 152.

[0132] In another implementation, the processor 180 can send the received second product search results to the user terminal.

[0133] Figures 4 to 8 This is a diagram illustrating a process for providing a first product search result when a model code is incorrectly entered, according to an embodiment of the present disclosure. The first product search result includes information about products corresponding to model codes similar to the incorrectly entered model code.

[0134] Figure 4 It can be a manifestation Figure 3 The implementation method of step S307.

[0135] Figure 4 Each step in the process is described as being executed by the processor 260 of the AI ​​server 200, but it can also be executed by the processor 180 of the artificial intelligence device 100.

[0136] refer to Figure 4 The processor 260 of the artificial intelligence server 200 can convert each character or each unit string constituting the model code obtained through user input into a unit embedding vector (S401).

[0137] The model code may include a string used to identify the product. The string may consist of multiple characters or multiple unit strings.

[0138] A character is the smallest unit of symbol in text. A character can be any of letters, numbers, or special symbols.

[0139] A unit string can be a substring that sequentially contains one or more characters from a given string. Unit strings can represent specific attributes of a model code.

[0140] The processor 260 can recognize each character or each unit string that makes up the model code, and convert each recognized character into a unit embedding vector, or convert each recognized unit string into a unit embedding vector.

[0141] The processor 260 can recognize each character or unit string that constitutes the received model code, converting each recognized character into a unit embedding vector, or each recognized unit string into a unit embedding vector. Furthermore, it can receive characters from different languages ​​and convert them into embedding vectors.

[0142] The processor 260 of the artificial intelligence server 200 can generate an embedding vector based on multiple transformed unit embedding vectors (S403).

[0143] Processor 260 can generate an embedding vector by adding multiple transformed unit embedding vectors together.

[0144] Processor 260 can apply weights to one or more of the unit embedding vectors of multiple transformations and generate embedding vectors based on the results of applying the weights.

[0145] In one implementation, processor 260 may additionally apply weights to the unit embedding vector corresponding to the character or unit string determined to be a typo, and may generate the embedding vector based on the result of the additional application of weights.

[0146] The processor 260 of the artificial intelligence server 200 can extract one or more product embedding vectors from multiple pre-stored product embedding vectors, the similarity of which is greater than or equal to a specific similarity level with the generated embedding vector (S405).

[0147] Processor 260 can use cosine similarity to measure the similarity between the generated embedding vector and the previously stored product embedding vector, but the implementation is not limited to this. For example, other distance metrics can be used to find relevant results from a multidimensional vector embedding space, such as Euclidean distance, dot product, approximate nearest neighbor (ANN) algorithm, etc.

[0148] Cosine similarity can be an indicator of the directional similarity between two vectors.

[0149] The processor 260 can measure the similarity between the generated embedding vector and multiple product embedding vectors, and extract one or more product embedding vectors whose similarity is greater than or equal to a specific similarity level.

[0150] The processor 260 of the artificial intelligence server 200 can obtain one or more product information corresponding to one or more extracted product embedding vectors as the first product search result (S407).

[0151] In this way, using embedding vectors can help provide users with more relevant search results and product information. For example, embedding vectors can represent data in a way that captures the potential meaning and relationships between items. This allows for searches beyond simple keyword matching, which is helpful in handling numbers and digits in model codes, rather than text words. For example, vector embedding similarity search can capture semantic meaning and contextual relevance, which can provide improved accuracy and precision. Furthermore, it can discover interesting relationships in the data and better handle ambiguity and personalization, enabling more targeted recommendations.

[0152] However, according to another implementation, a hybrid approach can be used, which may include initial screening by keyword matching and then refinement by vector similarity.

[0153] Figures 5 to 8 A process for obtaining a valid model code when a user incorrectly enters a model code, according to an embodiment of the present disclosure, is shown.

[0154] exist Figures 5 to 8 In the example, suppose a user enters a search term displayed in the search window of an artificial intelligence device 100. <rf21gsg>Model code.

[0155] Artificial intelligence server 200 can be determined <rf21gsg>Whether it is stored in memory 230, and if <rf21gsg>If it is not stored in memory 230, then it can be determined. <rf21gsg>The model code was entered incorrectly.

[0156] exist Figures 5 to 8 In the middle, assuming <rf21gsg>It is an incorrectly entered model code, such as several digits or characters entered incorrectly (for example, the correct code could be RD21GSC, but the user mistakenly entered RF21GSG).

[0157] refer to Figure 5 The processor 260 can <rf21gsg>Each of the multiple unit characters in 510 is converted into multiple unit embedding vectors 511. Each unit embedding vector can consist of multiple channels with values.

[0158] Processor 260 can additionally apply weights. <rf21gsg>The unit embedding vector (position-aware adaptive weighting) at the location where the typo occurred. The processor 260 can generate the embedding vector 520 by adding the unit embedding vectors according to the weight application results.

[0159] The processor 260 can measure the similarity (similarity search) between multiple product embedding vectors 530 stored in the memory 230 and the generated embedding vector 520. For example, the similarity between embedding vector 520 and the first product embedding vector 531 can be 0.81, and the similarity between embedding vector 520 and the second product embedding vector 532 can be 0.79.

[0160] Each of the multiple product embedding vectors 530 can be matched with multiple valid model codes (product model codes).

[0161] Processor 260 can obtain a list 540 of similar embedding vectors, including the K product embedding vectors most similar to embedding vector 520, based on the similarity measurement results (top K candidates). K is illustrated using 3 as an example, but this is just an example.

[0162] Processor 260 can determine the order of the K product embedding vectors by adaptively applying a penalty (adaptive penalty ordering). Processor 260 can apply a penalty to the K product embedding vectors based on the comparison result between the erroneous input model code 510 and the K product model codes, and determine the order based on the result of the penalty.

[0163] For example, processor 260 can impose a larger penalty as the number of incorrectly entered characters increases, and a smaller penalty as the number of incorrectly entered characters decreases.

[0164] Because the product model code of the first product embedding vector 531 differs from the input model code 510 by two characters, the processor 260 can apply a penalty of -0.09 to the calculated similarity. Because the product model code of the second product embedding vector 532 differs from the input model code 510 by three characters, the processor 260 can apply a penalty of -0.16 to the calculated similarity. Because the product model code of the third product embedding vector 533 differs from the input model code 510 by three characters, the processor 260 can apply a penalty of -0.16 to the calculated similarity.

[0165] Processor 260 can obtain sorting result 550, in which K product embedding vectors are arranged according to the determined sorting.

[0166] Processor 260 can obtain the highest priority valid model code (RD21GSC)551 as the final model code based on the determined sorting.

[0167] refer to Figure 6 The processor 260 can <rf21gsg>The multiple unit strings in 510 are converted into multiple unit embedding vectors in 611. Model codes can consist of unique identifier strings, and these unique identifier strings can include multiple unit strings. Each unit string can represent a product specification element, characteristic, property, or function, etc.

[0168] Processor 260 can additionally apply weights. <rf21gsg>The unit embedding vector (position-aware adaptive weights) at the location where the typo occurred. The processor 260 can generate the embedding vector 620 by adding the unit embedding vectors according to the weight application results.

[0169] Processor 260 can measure the similarity (similarity search) between a plurality of product embedding vectors 530 stored in memory 230 and the generated embedding vector 620. Each of the plurality of product embedding vectors 530 can be matched with a plurality of valid model codes, which can correspond to a specific product (e.g., home appliances, microwave ovens, refrigerators, etc.).

[0170] The processor 260 can obtain a list 640 of similar embedding vectors, including the three product embedding vectors most similar to the embedding vector 520, based on the similarity measurement results (top K candidates).

[0171] Processor 260 can determine the order of the three product embedding vectors by adaptively applying penalties (adaptive penalty ordering). Processor 260 can generate the order of the three product embedding vectors by applying penalties to the length difference of the model code and the typo pattern.

[0172] Processor 260 can obtain sorting result 650, in which the three product embedding vectors are arranged according to the determined sorting.

[0173] The processor 260 can obtain the highest priority valid model code 651 (RD21GSC) as the final model code based on the determined sorting. In this way, even when the user enters the wrong model, the AI ​​device can effectively and accurately determine the correct model (e.g., seemingly reading the user's thoughts and knowing exactly what he or she wants or intends to want).

[0174] Figure 7 Similar to Figure 5 However, it also provides the most similar model code 551 and the category of model code 551 (dryer) based on the three product embedding vectors ordered by adaptive penalty.

[0175] In other words, processor 260 can generate a first product search result including product information and product category of model code 551, and send the generated first product search result to artificial intelligence device 100.

[0176] The AI ​​device 100 can display a first product search result that includes product information and product category. If it is not the desired product category, the user can enter the desired product category (washing machine) into the AI ​​device 100 to refine the search results.

[0177] Artificial intelligence device 100 can send the categories of products desired by the user to artificial intelligence server 200. Artificial intelligence server 200 can reorder the three product embedding vectors included in the sorting result 550 based on the received product categories (reordering according to the response).

[0178] The artificial intelligence server 200 can obtain the reordering result 560 based on the re-determination, and obtain the model code (RX25GSGR) 561 corresponding to the washing machine category in the three product embedding vectors as the final model code.

[0179] Figure 8 Similar to Figure 5 However, it also provides the most similar model code 551 among three product embedding vectors, sorted by adaptive penalty sorting, the category (dryer) of model code 551, and the specification elements (capacity, function, price) of model code 551.

[0180] In other words, processor 260 can generate a first product search result including product information, product category, and specification elements such as model code 551, and send the generated first product search result to artificial intelligence device 100.

[0181] The AI ​​device 100 can display the first product search result, including product information, product category, and specifications. If the user knows the category and specifications of the desired product, the user can input the desired product category (washing machine) and specifications into the AI ​​device 100.

[0182] Artificial intelligence device 100 can send the product category and specification elements desired by the user to artificial intelligence server 200. Artificial intelligence server 200 can reorder the three product embedding vectors included in the sorting result 550 based on the received product category and specification elements (reordering according to the response).

[0183] The artificial intelligence server 200 can obtain the reordering result 570 based on the re-determination, and obtain the model code (RX25GSGR) 571 corresponding to the washing machine category in the three product embedding vectors as the final model code.

[0184] Figures 9A to 9C This is a diagram illustrating a process, according to an embodiment of the present disclosure, of providing product information with a product model code similar to the incorrectly entered model code when an incorrect model code is entered.

[0185] exist Figures 9A to 9C In this context, the artificial intelligence device 100 can display webpage 900 on the display 151. Webpage 900 may be a page that provides product search services.

[0186] Webpage 900 may include a search window 910, which includes an input field 911 and a search button 912. The model code 913 entered by the user can be displayed on the input field 911.

[0187] After displaying model code 913 on input field 911, AI device 100 can send a search request for model code 913 to AI server 200 according to the command of selecting search button 912.

[0188] If the search for model code 913 fails in response to the search request, the artificial intelligence server 200 may send a first product search result to the artificial intelligence device 100, which includes one or more product information that matches one or more product model codes similar to model code 913.

[0189] The artificial intelligence device 100 can display a first product search result 920 in a first area of ​​a webpage 900. The first product search result 920 includes text indicating that the search for product model code 913 failed, as well as product information for product model codes similar to model code 913. The first product search result 920 may include multiple product entries 921, 922, and 923.

[0190] Each product entry may include product information with a product model code similar to the model code 913 that was mistakenly entered by the user.

[0191] Each product entry may include at least one of the following: a valid product model code, a product name, a product capacity, or a product price.

[0192] The artificial intelligence device 100 can also display a chatbot icon 930 on a webpage 900. The chatbot icon 930 can be an icon used to execute a chatbot application to guide the search, selection, or purchase of products.

[0193] Artificial intelligence device 100 can display a chatbot screen 931 on webpage 900 to guide product search based on the selection of chatbot icon 930.

[0194] In this way, according to the embodiments of this disclosure, the problem of receiving incorrect or erroneous model codes can be effectively solved, and the user can be recommended a model that is most similar in function to the input model code.

[0195] Therefore, even if users don't remember the exact model code or make a typo, they can easily find the product they want, which can greatly improve the user's search experience.

[0196] like Figure 9B As shown, the artificial intelligence device 100 can display product search category results 940 based on the selected product item 922 in a second area of ​​the webpage 900, according to a command for selecting a product item 922 included in the first product search results 920. The second area may be located at the bottom of the first area, but this is just an example.

[0197] Product search category results 940 may include multiple product entries that are added or rearranged based on the attributes or categories of the selected product entries.

[0198] Artificial intelligence device 100 can display product search classification results 940 based on the category of the second product entry 922 selected by the command to select the second product entry 922 from the first product search results 920 displayed in the first area, which includes the first product entry 921, the second product entry 922, and the third product entry 923.

[0199] When the category of the second product entry 922 is a mini washing machine, the product search classification result 940 may include multiple mini washing machine entries 941, 942 and 943 for multiple mini washing machines.

[0200] Therefore, according to embodiments of this disclosure, products that are most similar in function to the desired product can be easily searched, thereby improving the experience in the product search and selection process.

[0201] At the same time, such as Figure 9C As shown, the first product entry 921, the second product entry 922, and the third product entry 923 included in the first product search result 920 can be categorized and displayed based on attributes, categories, or whether specific functions are supported.

[0202] For example, suppose the first product item 921, the second product item 922, and the third product item 923 are washing machine items, the first product item 921 and the third product item 923 provide drying functions, while the second product item 922 does not provide drying functions.

[0203] The AI ​​device 100 can display the first product 921 and the third product 923, which provide drying functions, in a different area than the second product 922. For example, the second product entry 922, which lacks drying functions, can be spaced apart from the first product entry 921 and the third product entry 923, but the implementation is not limited to this. For example, highlighting, bolding, or different colors can be used to convey the functional differences between the product entries.

[0204] Therefore, according to embodiments of this disclosure, when searched products are categorized according to function, users can find products with desired specific functions more quickly and accurately.

[0205] To reiterate Figure 3 .

[0206] When the processor 260 of the AI ​​server 200 determines that the search for the obtained model code is successful, it can obtain a second product search result based on the functional similarity with the product of the obtained model code (S313).

[0207] If it is determined that the obtained model code has been stored, the processor 260 can determine that the search code has been entered correctly.

[0208] If the search for the obtained model code is successful, the processor 260 can obtain product information for one or more products with similar functions to the product with that model code as a second product search result.

[0209] The processor 260 of the AI ​​server 200 can send the second product search results obtained through the communication interface 110 to the artificial intelligence device 100 (S315).

[0210] The processor 180 of the artificial intelligence device 100 can display the received second product search result on the display 151 (S317). In other words, additional search results for similar and / or other products can be provided to the user, which are different from the original search product but can be well matched with the original search product (e.g., if the user searches for washing machine, the user may also be interested in dryer).

[0211] Figure 10 and Figure 11 This is a diagram illustrating a process for obtaining a second product search result according to an embodiment of the present disclosure.

[0212] Reference Figure 10 The processor 260 of the artificial intelligence server 200 can obtain product specification information that matches the model code (S1001).

[0213] The memory 230 of the artificial intelligence server 200 can store model codes and product specification information matching those model codes. The product specification information may include multiple specification elements.

[0214] refer to Figure 11 Assume the user enters the correct model code <27ART10DKPL>1110.

[0215] The processor 260 can obtain a knowledge graph based on the acquired product specification information (S1003).

[0216] A knowledge graph can be a graph that structurally represents multiple specification elements of a product.

[0217] Processor 260 can construct knowledge graph 1120 (knowledge graph construction) based on product specification information matching model code 1110. Knowledge graph 1120 can consist of multiple nodes and edges connecting the multiple nodes.

[0218] The central node of knowledge graph 1120 can be a model code, and the child nodes can be the values ​​of each specification element. Edges connecting the central node and its child nodes represent the relationships between them. These edges can correspond to specification elements.

[0219] Processor 260 can generate multiple unit embedding vectors corresponding to each of the multiple specification elements of specification information from the knowledge graph obtained through the knowledge graph embedding model (S1005).

[0220] A knowledge graph embedding model can be a model used to output embedding vectors from a knowledge graph. A knowledge graph embedding model can be the HouseE (HouseEmbedding) model

[1130] .

[0221] HouseE model 1130 can be a model used to express the hierarchical structure and sequential information of a knowledge graph.

[0222] Processor 260 can generate multiple unit embedding vectors 1140 (knowledge graph learning) by converting each of the multiple child nodes of the knowledge graph into an embedding space through the HousE model 1130.

[0223] Processor 260 can generate embedding vectors based on multiple unit embedding vectors (S1007).

[0224] The processor 260 can compute a weighted sum by considering the importance of each of the multiple unit embedding vectors 1140, and obtain the computed weighted sum as the embedding vector 1150.

[0225] Processor 260 may additionally apply weights to one or more of the multiple unit embedding vectors and generate embedding vector 1150 based on the result of applying the weights. Processor 260 may assign rank to each of the multiple specification elements and assign larger weights to the unit embedding vectors corresponding to the specification elements with higher ranks (rank-aware adaptive weighting).

[0226] Processor 260 can assign higher ratings to one or more of multiple specification elements based on user preference information. The user preference information can be preference-specific elements created based on the user's past purchase history. The user preference information can be pre-stored in memory 230.

[0227] Processor 260 can calculate the similarity between the generated embedding vector 1150 and multiple product embedding vectors 1160 stored in memory 230. Processor 260 can obtain the K most similar product embedding vectors 1170 to embedding vector 1150 based on the similarity calculation results. K can be 3, but this is just an example and the implementation is not limited to this (K can be a number greater than or equal to 1, such as 10, 100, 1,000, etc.).

[0228] The processor 260 can obtain product information of the product that matches the embedded vector as a second product search result (S1009).

[0229] The processor 260 can extract product information that matches each of the multiple product embedding vectors 1170 from the memory 230 and obtain the extracted product information as a second product search result.

[0230] The second product search results may include product information for one or more products that are functionally similar to the product whose model code the user has correctly entered.

[0231] Figures 12A to 12C This is a diagram illustrating an example of providing second product search results in response to a search request for a model code correctly entered by a user, according to an embodiment of this disclosure.

[0232] exist Figures 12A to 12C In this context, the artificial intelligence device 100 can display a webpage 1200 on a monitor 151. The webpage 1200 may be a page that provides product search services.

[0233] Web page 1200 may include a search window 910, which includes an input field 911 and a search button 912. The model code 1201 entered by the user can be displayed on the input field 911.

[0234] After the model code 1201 is displayed on the input field 911, the artificial intelligence device 100 can send a search request for the model code 1201 to the artificial intelligence server 200 according to the command used to select the search button 912.

[0235] If the search for model code 1201 is successful in response to the search request, the artificial intelligence server 200 may send a second product search result to the artificial intelligence device 100. The second product search result includes product information that matches model code 1201 and one or more product information that matches one or more product model codes similar to model code 1201.

[0236] Artificial intelligence device 100 can display a second product search result 1203 on webpage 900, including a basic product search result 1210 and similar product search results 1220. The basic product search result 1210 includes product information of products matching model code 1201, while the similar product search result 1220 includes product information of one or more product model codes similar to model code 1201. The similar product search result 1220 may include multiple similar product entries 1221, 1222, and 1223.

[0237] Each product entry may include product information for a product model code similar to the model code 1201 entered by the user.

[0238] Each product entry may include at least one of the following: a valid product model code, a product name, a product capacity, or a product price.

[0239] The artificial intelligence device 100 can also display a chatbot icon 1230 on a webpage 1200. The chatbot icon 1230 can be an icon used to execute a chatbot application to guide users to search for, select, or purchase products.

[0240] refer to Figure 12B The artificial intelligence device 100 may additionally provide functional summary information to the similar product search results 1220. The similar product search results 1220 may include functional summary information to explain the differences between the functions of the product matching the model code and the functions of similar products.

[0241] For example, the artificial intelligence device 100 may display first function summary information 1221a on one side of the first product item 1221, second function summary information 1222a on one side of the second product item 1222, and third function summary information 1223a on one side of the third product item 1223.

[0242] Each of the first functional summary information 1221a, the second functional summary information 1222a, and the third functional summary information 1223a may include a summary of the functions of each product entry.

[0243] Each of the first function summary information 1221a, the second function summary information 1222a, and the third function summary information 1223a may include information about functions compared to other products. The other product may be a product corresponding to a correctly entered model code or a similar product.

[0244] In one implementation, the artificial intelligence device 100 can generate functional summary information based on product information for each product entry using a large language model (LLM). The large language model can be stored in memory 170.

[0245] In another implementation, the artificial intelligence server 200 can generate functional summary information based on the product information of each product entry using a large language model, and send the second product search result 1203 and the functional summary information to the artificial intelligence device 100.

[0246] Users can efficiently and conveniently check information about the product through the feature summary information.

[0247] Simultaneously, a refresh button 1240 can be further displayed on webpage 1200. The refresh button 1240 can be a button used to provide a functional summary of each product entry in a new expression or according to a new / different style or format. A large language model can be used in this process.

[0248] refer to Figure 12C The artificial intelligence device 100 can also display multiple attribute buttons 1251 to 1255 on the webpage 1200 for classifying multiple similar product entries based on the product's attributes.

[0249] Each of the multiple attribute buttons 1251 to 1255 may be a button used to sort or recommend additional product entries based on the user's preferred product attributes among multiple similar product entries 1221, 1222 and 1223.

[0250] The first button 1251 can be used to sort multiple product items 1221, 1222, and 1223 or to recommend additional product items based on whether the product supports the knock-on function. For example, when the first button 1251 is selected, the artificial intelligence device 100 can prioritize product items that support the knock-on function among the multiple product items 1221, 1222, and 1223.

[0251] The second button 1252 may be a button used to categorize or recommend additional product items from multiple product entries 1221, 1222, and 1223 based on the product's color. When the second button 1252 is selected, the AI ​​device 100 may display only the product entries from the multiple product entries 1221, 1222, and 1223 that have the same color as the product with model code 1201. The second button 1252 may also be a color button representing a specific color.

[0252] The third button 1253 may be used to sort or recommend additional product items among multiple product items 1221, 1222, and 1223 based on the product's price. When the third button 1253 is selected, the artificial intelligence device 100 may display product items among the multiple product items 1221, 1222, and 1223 that have a price similar to that of model code 1201.

[0253] The fourth button 1254 may be used to categorize or recommend additional product entries from multiple product entries 1221, 1222, and 1223 based on the product's capacity. When the fourth button 1254 is selected, the artificial intelligence device 100 may display a product entry from the multiple product entries 1221, 1222, and 1223 that has the same capacity as the product with model code 1201.

[0254] The fifth button 1255 can be used to categorize or recommend additional product items from multiple product entries 1221, 1222, and 1223 based on the product's ice purification function. When the fifth button 1255 is selected, the AI ​​device 100 can display only the product entries from the multiple product entries 1221, 1222, and 1223 that have the ice purification function.

[0255] Figures 13A to 13L This is a diagram illustrating a process for providing an interactive business service for product search according to an embodiment of the present disclosure.

[0256] In the following text, it is assumed that the artificial intelligence server 200 receives a query from the artificial intelligence device 100, generates a response to the query, and sends the generated response to the artificial intelligence device 100.

[0257] However, it is not necessary to be limited to this, and the memory 170 of the artificial intelligence device 100 can (e.g., locally) store the LLM 1302, and can directly generate responses to user queries through the LLM 1302.

[0258] refer to Figure 13A The AI ​​device 100 can display the execution screen 1300 of the chatbot application on the display 151. The AI ​​device 100 can receive requests for dehumidifier recommendations 1301 through the execution screen 1300 of the chatbot application.

[0259] In one implementation, the artificial intelligence device 100 can send the received recommendation query 1301 to the artificial intelligence server 200. The artificial intelligence server 200 can obtain the recommendation query 1301 and the knowledge graph 1310 corresponding to the dehumidifier included in the recommendation query 1301 as input to the LLM 1302. The knowledge graph may be referred to as a specification knowledge graph.

[0260] The LLM 1302 can be stored in the memory 230 of the artificial intelligence server 200.

[0261] LLM 1302 can output the dehumidifier node (A) based on the recommended query 1301 and the knowledge graph 1310.

[0262] The AI ​​server 200 can record dehumidifiers in structured answers.

[0263] The artificial intelligence server 200 can check the priority of dehumidifiers based on the knowledge graph 1310. For example... Figure 13B As shown, the artificial intelligence server 200 can extract the first subgraph 1311 corresponding to the first priority <capacity>.

[0264] Server 200 can input the extracted first sub-graph 1311 into LLM 1302 to generate an inquiry response 1303 that queries the installation space of the dehumidifier in relation to its capacity. The generated inquiry response 1303 can be sent to AI device 100 and displayed on the execution screen 1300 of the chatbot application.

[0265] like Figure 13C As shown, the artificial intelligence device 100 can receive an installation space query 1304 indicating the installation space (storage room or closet) of the dehumidifier after the problem response 1303, and send the received installation space query 1304 to the artificial intelligence server 200.

[0266] The AI ​​server 200 can extract the dehumidifier's capacity (13L) from the installation space query 1304 and knowledge graph 1310 via LLM 1302.

[0267] In addition, the AI ​​server 200 can record capacity (13L) in the structured answer.

[0268] Then, as Figure 13D As shown, the AI ​​server 200 can input the second sub-graph 1312 corresponding to the function with the second priority into the LLM 1302 to generate a question response 1305 inquiring about the dehumidifier's function. The generated question response 1305 can be sent to the AI ​​device 100 and displayed on the execution screen 1300 of the chatbot application. The question response 1305 may include an answer to the installation space query 1304.

[0269] like Figure 13E As shown, the artificial intelligence device 100 can receive the function query 1306 requesting the dehumidifier's function and send the received function query 1306 to the artificial intelligence server 200.

[0270] Artificial intelligence server 200 can extract the dehumidifier's functions (UVnano) from function query 1306 and knowledge graph 1310 via LLM 1302.

[0271] The AI ​​server 200 can also record features (UVnano) in structured answers.

[0272] Then, as Figure 13E As shown, the AI ​​server 200 can input the third sub-graph 1313 corresponding to the third priority function into the LLM 1302 to generate a question response 1307 inquiring about the color of the dehumidifier. The generated question response 1307 can be sent to the AI ​​device 100 and displayed on the execution screen 1300 of the chatbot application. The question response 1307 may include an answer to the requested function query 1306.

[0273] like Figure 13F As shown, the artificial intelligence device 100 can receive a color query 1308 indicating beige and send the received color query 1308 to the artificial intelligence server 200. The artificial intelligence server 200 can additionally record the color (beige) in the structured answer.

[0274] Artificial intelligence server 200 can eventually obtain a structured answer 1320 for {Category name: dehumidifier, Capacity: 13L, Function: UVnano, Color: beige}.

[0275] Artificial intelligence server 200 can extract similar products based on structured answer 1320.

[0276] The artificial intelligence server 200 can generate a knowledge graph based on the structured answer 1320. According to... Figure 11 In this implementation, the artificial intelligence server 200 can use a knowledge graph generated based on the knowledge graph embedding model 1130 to obtain multiple product model codes. Figure 11 The description replaces the detailed description.

[0277] and Figure 11 The difference in the implementation method is that, Figure 11 In the implementation method, a knowledge graph is created based on the model code 1110 directly input by the user, and... Figure 13G In this implementation, a knowledge graph is created based on the structured answer 1320 that combines attributes obtained from the user's query, without receiving model codes.

[0278] Artificial intelligence server 200 can obtain product model code 1321 corresponding to products with attributes similar to those of structured answer 1320.

[0279] Artificial intelligence server 200 can generate product recommendation response 1309 from product model code 1321 via LLM 1302 and send product recommendation response 1309 to artificial intelligence device 100. For example... Figure 13G As shown, the artificial intelligence device 100 can display product recommendation responses 1309 on the execution screen 1300 of the chatbot application.

[0280] Product recommendation response 1309 may include the product model code and a description of the product model code.

[0281] like Figure 13H As shown, the artificial intelligence device 100 can receive a comparison query 1331 requesting a difference between products corresponding to two product model codes included in the product recommendation response 1309, and send the comparison query 1331 to the artificial intelligence server 200.

[0282] The artificial intelligence server 200 can generate a knowledge graph-specific comparison query 1331a for comparing two products based on the received comparison query 1331.

[0283] Artificial intelligence server 200 can extract subgraph 1341 describing the relationship between two products from knowledge graph 1340 representing the relationship between product model codes 1321.

[0284] The AI ​​server 200 can extract the commonalities and differences between two nodes corresponding to two products from subgraph 1341, and based on the extraction results, such as Figure 13I As shown, a comparison table 1342 is generated, which includes the commonalities and differences between the two products. Each commonality and difference can be expressed as a keyword representing a product feature, and each feature can be pre-assigned a priority.

[0285] The artificial intelligence server 200 can generate input sentences 1344 describing the similarity and differences between two products based on keywords, keyword priority, and a terminology dictionary 1343 that describes the keywords.

[0286] like Figure 13J As shown, server 200 can generate product comparison response 1352 from input sentence 1344 via LLM 1302. AI server 200 can send product comparison response 1332 to AI device 100, and AI device 100 can display product comparison response 1332 on screen 1300, such as... Figure 13K As shown.

[0287] like Figure 13L As shown, the artificial intelligence device 100 can receive queries corresponding to the model code. <dz16peca>The price query for the product is 1333, and the received price query 1333 is sent to the artificial intelligence server 200.

[0288] If the model code included in the received price query 1333 is not stored, the artificial intelligence server 200 can obtain a product model code similar to that model code. In this regard, it can be applied... Figure 5 and Figure 6 One implementation method involves obtaining product model codes that are similar to the input model codes based on the similarity of the model codes.

[0289] Artificial intelligence server 200 can extract products with product model codes (DQ163PECA) similar to the model codes included in price query 1333, and generate a price response 1352 based on the extracted product information. Artificial intelligence server 200 can generate price response 1352 by inputting a knowledge graph 1351 based on the attributes of the product model code (DQ163PECA) into LLM 1302.

[0290] The AI ​​server 200 can send the price response 1352 to the AI ​​device 100, and the AI ​​device 100 can display the received price response 1352 on the execution screen 1300 of the chatbot application.

[0291] Therefore, according to the embodiments of this disclosure, users can easily search for products that match their preferences through interactive business services, and even if the product model code is entered incorrectly, products that are as similar as possible can be recommended.

[0292] Figure 14A and Figure 14B This is a diagram illustrating the process for determining detailed entries for a product when only a portion of the model code is entered into the search window according to an embodiment of the present disclosure.

[0293] refer to Figure 14A The artificial intelligence device 100 can display a webpage 1400, including a search window 910, on a display 151.

[0294] Artificial intelligence device 100 can receive only the model code in search window 910. <fx>Part of the code is 1401. Each character or string constituting the model code can represent a detailed entry of the product, and the detailed entry corresponding to each letter or string can be pre-stored in memory 170. For example, F can represent the type of product, while X can represent the control panel.

[0295] When model code <fx>When part of the code 1401 is entered into the search window 910, the artificial intelligence device 100 can recognize part of the model code 1401 and display the tag list 1410.

[0296] The label list 1410 may include multiple detailed entries 1411 to 1416 for determining the detailed entry of the product corresponding to the partial code 1401 of the input model code.

[0297] The character corresponding to the first detail item 1411 is F, and the character corresponding to the second detail item 1412 is X. The AI ​​device 100 can generate the model code based on the input of selecting a capacity of 20 kg for the third detail item 1413. <fx20>And reflected in search window 910 <fx20>The artificial intelligence device 100 can display the code corresponding to the selected value by linking the code with the partial code 1401 based on the input used to select the value of each detailed entry.

[0298] Artificial intelligence device 100 can determine the model code based on the selected input for the fourth detailed entry 1413 to the sixth detailed entry 1416.

[0299] The artificial intelligence device 100 can display a product image 1420 on one side of the label list 1410 based on the selection of a detailed item. The product image 1420 can change to reflect the content of the selected detailed item based on the selection made for that item.

[0300] Meanwhile, if the search button 912 is not selected within a certain period of time after the tag list 1410 is displayed, the artificial intelligence device 100 can automatically run the chatbot application.

[0301] Therefore, according to embodiments of this disclosure, even if a user only enters a portion of the model code, he or she can easily search for products reflecting the desired attributes through the tag list 1410 and product image 1420.

[0302] refer to Figure 14B Artificial intelligence device 100 can identify the product type as a drum washing machine with Easy Circle function based on part of the model code 1401.

[0303] The artificial intelligence device 100 can run a chatbot application to guide the selection of detailed entries for the drum washing machine and display the chatbot application's execution screen 1430. The chatbot application's execution screen 1430 may include questions to determine the capacity of the drum washing machine.

[0304] Users can search for desired products by responding to questions.

[0305] Therefore, according to embodiments of this disclosure, even if a user only enters part of the model code, he or she can efficiently and conveniently search for products that reflect the desired attributes through a chatbot application.

[0306] Figure 15 This is a diagram illustrating a method of operating an artificial intelligence device according to an embodiment of the present disclosure.

[0307] The processor 180 of the artificial intelligence device 100 can obtain the model code (S1501) through user input.

[0308] Replace the description of step S1501 with Figure 3 The relevant explanation of step S301 in the process.

[0309] The processor 180 can receive a search request for the obtained model code (S1503).

[0310] The processor 180 can receive user input as a search request by selecting a search button located on one side of a search term input window included on a webpage.

[0311] As another example, processor 180 can obtain search requests via user voice commands.

[0312] The processor 180 can perform a search based on the received search request and determine whether the search for the model code has failed (S1505).

[0313] Processor 180 can execute in Figure 3 The operation of the processor 260 of the AI ​​server 200 is performed in step S305. For this purpose, the memory 170 can store multiple product model codes.

[0314] If the obtained model code is not stored in memory 170, the processor 180 can determine that the search for the model code has failed, and if the obtained model code is stored in memory 170, the processor 180 can determine that the search for the model code has succeeded.

[0315] If it is determined that the search for the obtained model code fails, the processor 180 may display on the display 151 a first product search result based on the similarity between the obtained model code and the previously stored model code (S1507).

[0316] If it is determined that the search for the obtained model code fails, the processor 180 may extract one or more model codes that are similar to the obtained model code from multiple model codes, and obtain product information identifying one or more products of the one or more extracted model codes as the first product search result.

[0317] Processor 180 can Figure 3 In step S307, the operation of processor 260 is performed. For this purpose, memory 170 can store multiple product model codes, multiple product embedding vectors matching each of the multiple product model codes, and multiple product information matching each of the multiple product model codes.

[0318] If the processor 180 determines that the search for the obtained model code is successful, the processor 180 can display a second product search result on the display 151 based on the functional similarity of the product with the searched model code (S1509).

[0319] Processor 180 can Figure 3 In step S313, the operation of processor 260 is performed. For this purpose, memory 170 can store the product's knowledge graph and knowledge graph embedding model.

[0320] An artificial intelligence device 100 according to an embodiment of the present disclosure may include a display 151; and at least one processor 180 configured to: receive a model code identifying a product via user input; if a search for the model code fails, obtain a first product search result based on the similarity between the model code and a pre-stored product model code; and display the obtained first product search result on the display.

[0321] The first product search result may include multiple product entries displayed in a first area, and if one of the multiple product entries is selected, at least one processor may display search categorization results rearranged based on the category or attribute of the selected product entry in a second area.

[0322] The first product search result includes multiple product entries, and these entries can be displayed in different areas based on their attributes.

[0323] If the search for the model code is successful, at least one processor 180 can display a second product search result on display 151 based on the functional similarity of the product corresponding to the model code.

[0324] The second product search results may include basic product search results for products that match the successfully searched model code, as well as similar product search results for one or more similar products corresponding to one or more product model codes that are similar to the successfully searched model code.

[0325] Similar product search results can include feature summary information to explain the differences between the features of the product matching the model code and the features of one or more similar products.

[0326] The similar product search results include multiple similar product entries.

[0327] At least one processor 180 can display multiple attribute buttons on a display for classifying multiple similar product entries based on the product's attributes.

[0328] If a partial code of the model code is entered, at least one processor 180 can display a list of labels on the display including detailed entries for products identified based on the partial code, and display a code corresponding to the selected value by linking the code with the partial code according to the input for selecting the value of each detailed entry.

[0329] At least one processor 180 can select a product image on a display, the product image corresponding to the code reflected when the value of each detailed entry is selected.

[0330] If a partial model code is entered and the product type is confirmed based on that partial model code, at least one processor can display an execution screen of the chatbot application on the display, and display a query for detailed entries to determine the product on the chatbot application execution screen. For example, according to an embodiment, in response to user input including a partial model code, a product type based on the partial model code and an execution screen of the chatbot application can be displayed. For example, if "FX" is a partial model code entered by the user, suggested or updated complete model codes can be displayed or populated in search window 910 and / or the chatbot can be activated.

[0331] The functions of the elements disclosed in this invention can be implemented using circuits or processing circuits including general-purpose processors, application-specific processors, integrated circuits, application-specific integrated circuits (ASICs), existing circuits, and / or combinations thereof. A processor can be defined as a processing circuit or circuit that includes transistors and other circuitry.

[0332] In this invention, a circuit, unit, or device can be hardware designed or programmed to perform a specified function. The hardware can be the hardware disclosed herein or other known hardware programmed or configured to perform the specified function. If the hardware is a processor, which can be considered a type of circuit, then the circuit, device, or unit is a combination of hardware and software, and the software can constitute the hardware and / or processor.

[0333] The above disclosure can be implemented as computer-readable code on a program recording medium. Computer-readable non-transitory media include all types of recording devices that store data readable by a computer system. Examples of computer-readable media include hard disk drives (HDDs), solid-state drives (SSDs), silicon disk drives (SDDs), ROMs, RAM, CD-ROMs, magnetic tapes, floppy disks, and optical data storage devices. Additionally, the computer may include a processor 180 for an artificial intelligence device.

[0334] This disclosure includes non-restrictive examples of the following provisions:

[0335] Clause 1. An artificial intelligence device, said artificial intelligence device comprising:

[0336] Displays; and

[0337] At least one processor, said at least one processor being configured to:

[0338] Receive user input based on the model code that identifies the product;

[0339] In response to a search based on the user input failing to match one of a plurality of pre-stored model codes, a first product search is obtained based on the similarity between the user input and one of the plurality of pre-stored model codes.

[0340] Search results; and

[0341] The search results for the first product are displayed.

[0342] Clause 2. The artificial intelligence device according to Clause 1, wherein the first product search result comprises multiple product entries displayed on a first area of ​​the display, and

[0343] The at least one processor is further configured to:

[0344] In response to receiving a selection of one of the plurality of product entries, search categorization results are displayed on a second area of ​​the display, the search categorization results being rearranged based on the category of the selected product or the attributes of the selected product entry corresponding to the selection.

[0345] Clause 3. The artificial intelligence device as described in Clause 1, wherein the first product search result includes multiple product entries displayed differently based on corresponding attributes.

[0346] Clause 4. The artificial intelligence device according to Clause 1, wherein the at least one processor is further configured to:

[0347] In response to a successful match between the user's input search and one of the plurality of pre-stored model codes, a second product search result is displayed based on the functional similarity of the product corresponding to the one of the plurality of pre-stored model codes.

[0348] Clause 5. The artificial intelligence device according to Clause 4, wherein the second product search result includes basic product search results and similar product search results, the basic product search results being for products matching the model code in the user input, and the similar product search results being for one or more similar products, the one or more similar products corresponding to one or more product model codes similar to the model code in the user input or corresponding to the functional similarity.

[0349] Clause 6. The artificial intelligence device as described in Clause 5, wherein the similar product search results include a summary of functions regarding the differences between the functions of the product matching the model code in the user input and the functions of the one or more similar products.

[0350] Clause 7. The artificial intelligence device as described in Clause 4, wherein the similar product search results include multiple similar product entries, and

[0351] The at least one processor is further configured to:

[0352] Displays multiple attribute buttons for classifying the multiple similar product entries based on attributes.

[0353] Clause 8. The artificial intelligence device according to Clause 1, wherein the at least one processor is further configured to:

[0354] In response to the user input including a portion of the model code, a list of tags is displayed, comprising detailed entries for products identified based on the portion of the model code; and

[0355] In response to receiving a selected value corresponding to one of the detailed entries, an update code corresponding to the selected value is displayed.

[0356] Clause 9. The artificial intelligence device according to Clause 8, wherein the at least one processor is further configured to:

[0357] Displays the product image corresponding to the selected value.

[0358] Clause 10. The artificial intelligence device according to Clause 1, wherein the at least one processor is further configured to:

[0359] In response to the user input including a portion of the model code, the execution screen of the chatbot application is displayed; and

[0360] The execution screen of the chatbot application displays a query for identifying detailed entries of a product.

[0361] Clause 11. A method for controlling an artificial intelligence device, the method comprising the steps of:

[0362] Receive user input based on the model code that identifies the product;

[0363] In response to a search based on the user input failing to match one of a plurality of pre-stored model codes, a first product search result is obtained based on the similarity between the user input and one of the plurality of pre-stored model codes; and

[0364] The search results for the first product are displayed.

[0365] Clause 12. The method described in Clause 11 further comprises the following steps:

[0366] The first product search result, comprising multiple product entries, is displayed on a first area of ​​the display of the artificial intelligence device; and

[0367] In response to receiving a selection of one of the plurality of product entries, search categorization results are displayed on a second area of ​​the display, the search categorization results being rearranged based on the category of the selected product or the attribute of the selected product entry corresponding to the selection.

[0368] Clause 13. The method according to Clause 11, wherein the first product search result includes multiple product entries, and the multiple product entries are displayed differently based on corresponding attributes.

[0369] Clause 14. The method according to Clause 11 further includes the following steps:

[0370] In response to a successful match between the user's input search and one of the plurality of pre-stored model codes, a second product search result is displayed based on the functional similarity of the product corresponding to the one of the plurality of pre-stored model codes.

[0371] Clause 15. The method according to Clause 14, wherein the second product search result includes a basic product search result and a similar product search result, the basic product search result being for products matching the model code in the user input, and the similar product search result being for one or more similar products, the one or more similar products corresponding to one or more product model codes similar to the model code in the user input or corresponding to the functional similarity.

[0372] Clause 16. The method according to Clause 15, wherein the similar product search results include a summary of features regarding the differences between the features of the product matching the model code in the user input and the features of the one or more similar products.

[0373] Clause 17. The method according to Clause 14, wherein the similar product search results include multiple similar product entries, and

[0374] The method further includes the following steps:

[0375] Displays multiple attribute buttons for classifying the multiple similar product entries based on attributes.

[0376] Clause 18. The method described in Clause 11 further comprises the following steps:

[0377] In response to the user input including a portion of the model code, a list of tags is displayed, comprising detailed entries for products identified based on the portion of the model code; and

[0378] In response to receiving a selected value corresponding to one of the detailed entries, an update code corresponding to the selected value and a product image corresponding to the code are displayed.

[0379] Clause 19. The method described in Clause 11 further comprises the following steps:

[0380] In response to the user input including a portion of the model code, the execution screen of the chatbot application is displayed; and

[0381] The execution screen of the chatbot application displays a query for identifying detailed entries of a product.

[0382] Clause 20. A non-transitory computer-readable medium storing computer-executable instructions that, when executed by a processor, cause the processor to perform the following operations:

[0383] Receive user input based on the model code that identifies the product;

[0384] In response to a search based on the user input failing to match one of a plurality of pre-stored model codes, a first product search result is obtained based on the similarity between the user input and one of the plurality of pre-stored model codes; and

[0385] The search results for the first product are displayed.

[0386] Cross-references to related applications

[0387] This application claims the benefit and priority of Korean Patent Application No. 10-2024-0188525, filed in Korea on December 17, 2024, the entire contents of which are incorporated herein by reference. < / fx> < / fx>

Claims

1. An artificial intelligence device, the artificial intelligence device comprising: monitor; as well as At least one processor, said at least one processor being configured to: Receive user input based on the product model code; In response to a search for the user input failing to match one of a plurality of pre-stored model codes, a first product search result is obtained based on the similarity between the user input and one of the plurality of pre-stored model codes; and The search results for the first product are displayed.

2. The artificial intelligence device according to claim 1, wherein, The first product search result includes multiple product entries displayed on a first area of ​​the display, and The at least one processor is further configured to: In response to receiving a selection of one of the plurality of product entries, search categorization results are displayed on a second area of ​​the display, the search categorization results being rearranged based on the category of the selected product or the attributes of the selected product entry corresponding to the selection.

3. The artificial intelligence device according to claim 1, wherein, The first product search result includes multiple product entries displayed differently based on corresponding attributes.

4. The artificial intelligence device according to claim 1, wherein, The at least one processor is further configured to: In response to a successful match between the user's input search and one of the plurality of pre-stored model codes, a second product search result is displayed based on the functional similarity of the product corresponding to the one of the plurality of pre-stored model codes.

5. The artificial intelligence device according to claim 4, wherein, The second product search results include basic product search results and similar product search results. The basic product search results are for products that match the model code in the user input, and the similar product search results are for one or more similar products. The one or more similar products correspond to one or more product model codes that are similar to the model code in the user input or correspond to the functional similarity.

6. The artificial intelligence device according to claim 5, wherein, The similar product search results include a summary of features regarding the differences between the features of the product matching the model code in the user input and the features of one or more similar products.

7. The artificial intelligence device according to claim 4, wherein, The similar product search results include multiple similar product entries, and The at least one processor is further configured to: Displays multiple attribute buttons for classifying the multiple similar product entries based on attributes.

8. The artificial intelligence device according to claim 1, wherein, The at least one processor is further configured to: In response to the user input including a portion of the model code, a list of tags is displayed, including detailed entries for products identified based on the portion of the model code; as well as In response to receiving a selected value corresponding to one of the detailed entries, an update code corresponding to the selected value is displayed.

9. A method for controlling an artificial intelligence device, the method comprising the following steps: Receive user input based on the product model code; In response to a search for the user input failing to match one of a plurality of pre-stored model codes, a first product search result is obtained based on the similarity between the user input and one of the plurality of pre-stored model codes; and The search results for the first product are displayed.

10. A non-transitory computer-readable medium storing computer-executable instructions that, when executed by a processor, cause the processor to perform the following operations: Receive user input based on the product model code; In response to a search for the user input failing to match one of a plurality of pre-stored model codes, a first product search result is obtained based on the similarity between the user input and one of the plurality of pre-stored model codes; and The search results for the first product are displayed.

Images