Search methods, search systems, and search programs

Abstract

This document provides a search method, a search system, and a search program. [Solution] The search method includes one or more processors obtaining a user query entered via a user terminal to search for an item, obtaining an initial result which is the result of searching the database based on the user query, wherein the initial result includes multiple hit items which are items that matched the search, displaying the initial result on the user terminal as a search result for the user query, calculating a search score for each of the multiple hit items which indicates the degree of relevance to the user query, determining a score order which ranks the multiple hit items based on the search score, and transitioning the display mode of the search results displayed on the user terminal in the score order.

Description

Technical Field

[0001] The present disclosure relates to a search method, a search system, and a search program.

Background Art

[0002] An EC (Electronic Commerce) site generally has a search window for entering a search query. By entering a search query in the search window, a user can search for items such as products or content. For example, Patent Document 1 discloses an information providing device that calculates a score according to the relevance between an input search query and content. By using such an information providing device, search results ranked in score order can be displayed on an EC site.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] Ranking scores for search results can be calculated in various ways. Generally, when a method for calculating scores more precisely is adopted, the calculation time becomes longer. Therefore, the accuracy of search result ranking and the speed of search result display often trade off. If the display of search results is slow, users may feel stressed, and thus there is a risk that the abandonment rate from that EC site will increase. In providing a search system, a problem is how to achieve both the speed of search result display and the accuracy of ranking.

Means for Solving the Problems

[0005] A search method according to one aspect of the present disclosure includes: one or more processors obtaining a user query entered via a user terminal to search for an item; obtaining an initial result which is the result of searching a database based on the user query, wherein the initial result includes a plurality of hit items which are items that matched the search; displaying the initial result on the user terminal as a search result for the user query; calculating a search score for each of the plurality of hit items which indicates the degree of relevance to the user query; determining a score order which ranks the plurality of hit items based on the search score; and transitioning the display mode of the search results displayed on the user terminal in the score order.

[0006] A search system according to one aspect of the present disclosure comprises one or more processors, the one or more processors being configured to: acquire a user query entered via a user terminal to search for an item; acquire an initial result which is the result of searching a database based on the user query, wherein the initial result includes a plurality of hit items which are items that matched the search; display the initial result on the user terminal as a search result for the user query; calculate a search score for each of the plurality of hit items which indicates the degree of relevance to the user query; determine a score order which ranks the plurality of hit items based on the search score; and transition the display mode of the search results displayed on the user terminal in the score order.

[0007] A search program according to one aspect of the present disclosure involves causing one or more processors to perform the following actions: to obtain a user query entered via a user terminal to search for an item; to obtain an initial result which is the result of searching a database based on the user query, wherein the initial result includes a plurality of hit items which are items that matched the search; to display the initial result on the user terminal as a search result for the user query; and to transition the display of the search result shown on the user terminal in a score order ranked based on the search score, wherein the search score is calculated for each of the plurality of hit items to indicate the degree of relevance to the user query.

[0008] A search system according to one aspect of the present disclosure is a search system for performing an item search on an online platform, the search system comprising one or more processors and a memory for storing program code executed by the one or more processors, wherein the program code includes query code configured to cause at least one of the one or more processors to receive a user query entered via a user terminal to search for the item, search code configured to cause at least one of the one or more processors to search a database based on the user query, wherein the search result includes a plurality of hit items, and the plurality A search system comprising: a display code configured to display hit items in the results area of ​​the graphical user interface of the user terminal; a score code configured to cause at least one of the one or more processors to calculate a search score indicating the degree of relevance of each of the plurality of hit items to the user query; a ranking code configured to cause at least one of the one or more processors to determine the score order of the plurality of hit items based on the search score; and a transition code configured to cause at least one of the one or more processors to transition the order of the plurality of hit items in the results area of ​​the graphical user interface according to the score order. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1 is a schematic diagram of the search system according to the embodiment. [Figure 2] Figure 2 is an example of a search screen displayed on a user's terminal. [Figure 3] Figure 3 illustrates the transitions in how search results are displayed. [Figure 4] Figure 4 is an example of a search screen after the display mode has changed. [Figure 5]Figure 5 is a flowchart illustrating the search method according to this embodiment. [Modes for carrying out the invention]

[0010] Examples of the search system 11, search method, and search program will be described with reference to Figures 1 to 5. The present invention is not limited to these examples and is intended to include all modifications within the meaning and scope equivalent to the claims, as shown in the claims.

[0011] [Overview of the search system] The search system 11 shown in Figure 1 (hereinafter simply referred to as "System 11") may be operated in conjunction with an online platform such as an e-commerce site (hereinafter referred to as "EC site"). System 11 may be configured to use AI (Artificial Intelligence) to search for multiple products traded on the EC site. Products may be, but are not limited to, goods, services, or content. Hereinafter, products that are searched will be referred to as "items".

[0012] System 11 may include a service server 20 for providing search functionality. System 11 may also include a web server for providing an e-commerce site platform. In this example, the service server 20 also functions as the web server.

[0013] The service server 20 may be implemented by a computer having, for example, one or more processors 21, one or more memory 22, and a communication interface 23. For convenience, the following description will explain an example in which the service server 20 has one processor 21 and one memory 22. The processor 21 is also called a server processor.

[0014] Memory 22 may store a program 24 executed by the processor 21. The program 24 includes an application and an operating system. The processor 21 implements various functions by executing multiple program codes contained in the program 24. The communication IF 23 enables communication with other devices (e.g., user terminal 30) via the network.

[0015] System 11 may have one or more user terminals 30. Below, we describe an example where System 11 has one user terminal 30. The user is configured to communicate with the service server 20 via the network through the user terminal 30.

[0016] The user terminal 30 may be implemented by a computer equipped with one or more processors 31, one or more memory 32, and a communication interface 33. The user terminal 30 may also be a mobile terminal such as a smartphone. For convenience, the following description will explain an example in which the user terminal 30 is equipped with one processor 31 and one memory 32. The processor 31 is also called a terminal processor.

[0017] The communication IF 33 enables communication with other devices (e.g., the service server 20) via the network. The user terminal 30 may include an output device 34 and an input device 35, or these may be connected externally. The output device 34 may be, for example, a display. The user terminal 30 may also be equipped with a touch panel, which is an input / output device.

[0018] System 11 may include a data server 12 for storing various types of data. The data server 12 may be implemented, for example, by a computer having one or more processors, one or more memories, and a communication interface. The data server 12 may be configured to manage one or more databases 13.

[0019] The service server 20 accesses the data server 12 as needed to obtain, update, or search for data included in one or more databases 13. Alternatively, or in addition, some or all of the data included in one or more databases 13 (e.g., index data for searching) may be held in the memory 22 of the service server 20.

[0020] One or more databases 13 may include a plurality of databases for each of a plurality of EC sites. The plurality of EC sites may include, for example, a fashion item sales site, a daily necessities sales site, a flea market site, an auction site, a book or e-book sales site, or an online supermarket, but are not limited thereto.

[0021] One or more databases 13 may include a user database that stores user data related to a plurality of users of an EC site. The user database may include a plurality of user data sets for each of the plurality of users. Each user data set may include a user ID for identifying the user.

[0022] The plurality of users may include registered users of an EC site or purchasers at an EC site. The user data may include a purchase history of a user who is a purchaser at an EC site. The purchase history may include an item ID of an item purchased by the corresponding user and a seller ID of the seller who provided the item.

[0023] One or more databases 13 may include a seller database that stores seller data for a plurality of sellers. The seller database includes a plurality of seller data sets for each of the plurality of sellers. Each seller data set may include a seller ID for identifying the seller. The seller data may include item data for a plurality of items registered by each of the plurality of sellers.

[0024] One or more databases 13 may include an item database that stores item data relating to multiple items sold on an e-commerce site. The item database may include multiple item datasets, each corresponding to a different item. The item database may have multiple tables that individually correspond to multiple e-commerce sites, or it may be a relational database that links multiple databases corresponding to multiple e-commerce sites together.

[0025] Each item dataset may include an item ID to identify the item. The item database may be associated with at least one of the seller database or the user database, for example, using the item ID as a key.

[0026] Each item dataset may include multiple attribute fields that indicate the attributes of the corresponding item. These multiple attribute fields may include, but are not limited to, one or more categories, store name, item name, item image, item description, model number, stock quantity, quantity, price, color, size, material, brand, gender attribute (e.g., men's, women's, unisex), sales period, sales performance, user review comments, review score, or data update date.

[0027] One or more categories may include higher and lower categories used to classify all items handled on each e-commerce site into multiple levels. Furthermore, there may be one or more intermediate categories between the higher and lower categories.

[0028] For example, an item's category classification may include multiple higher categories and multiple subcategories within each higher category. The higher categories may include, but are not limited to, fashion, food, drinks, household goods, cosmetics, home appliances, sports, interior design, pets, games, or books. The subcategories may correspond to product names such as mobile phones, laptops, cameras, clothing, or accessories.

[0029] System 11 may include one or more machine learning models to assist in searching. Alternatively, the service server 20 may use one or more machine learning models not included in System 11. The one or more machine learning models may include pre-trained models trained on training data containing ground truth data, or they may include generative AI used for general purposes. The generative AI may be a language model that takes natural language as input. The language model may be a Vision-Language Model (VLM) that can take images as input in addition to language.

[0030] The language model may be a large-scale language model (LLM) or a small-scale language model (SLM). A large-scale language model is a language model trained using a large amount of text data. A small-scale language model is a language model that is smaller in scale than a large-scale language model (for example, with fewer parameters). Hereinafter, the machine learning model used in this disclosure will be referred to as LLM19.

[0031] LLM19 may be a general-purpose natural language processing (NLP) model that can adapt to various natural language processing tasks, such as information extraction, text summarization, text generation, or question-and-answer sessions, depending on the input prompt. The prompt may include instructions on what kind of output to produce.

[0032] The service server 20 is configured to input a prompt to the LLM 19 and then retrieve the output generated by the LLM 19 according to the instructions. Memory 22 may store one or more templates for generating one or more prompts. The service server 20 may generate prompts by combining templates with various data as appropriate.

[0033] [Search screen] Figures 1 to 4 illustrate the search screen 50 displayed on the e-commerce site. The service server 20 may be configured to display the search screen 50 on the user terminal 30's display in response to user operations. In this disclosure, displaying the search screen 50 on the user terminal 30's display may be simply referred to as "displaying on the user terminal 30".

[0034] The search screen 50 may include an input field 51 for the user to enter information (e.g., search criteria) for searching for items. The information entered by the user through the input field 51 is called a user query 54. The user query 54 may be natural language text, a photograph, an image, audio, or any combination thereof.

[0035] The search screen 50 may also be a chat screen for interacting with the AI. The AI ​​may be LLM19 or another machine learning model. In this example, LLM19 provides a response function to handle chats with the user.

[0036] The search screen 50 may include a login button 55 for the user to log in to the e-commerce site. The service server 20 may be configured to perform authentication processing for the user in response to the user's login operation.

[0037] When a user logs into the e-commerce site, the service server 20 gains access to user data, such as the registered user's registration data or their usage history on the e-commerce site. For example, if a registered user performs a search on the e-commerce site, the service server 20 may output search results corresponding to the user data.

[0038] The search screen 50 may include an explanatory text 52 that describes how to interact with the AI, or it may include one or more example questions 53 for interaction. When a user enters search criteria in the input field 51, the entered search criteria may be displayed on the chat screen as a user query 54. Meanwhile, the user query 54 is sent from the user terminal 30 to the service server 20, and the service server 20 performs a search based on the user query 54.

[0039] On the search screen 50, a first loading indicator 56 may be displayed to indicate that a search is in progress during the search waiting time T1 (see Figure 5) from the time the user terminal 30 or service server 20 obtains the user query 54 until the search results (initial results) are displayed. Displaying the first loading indicator 56 during a search is called the first pending display. The search waiting time T1 may also be the duration for which the first pending display is shown.

[0040] The first loading indicator 56 may be accompanied by an animation indicating that data is being processed. For example, the first loading indicator 56 may be a dot animation in which a dot blinks or moves, a spinner in which a shape rotates, a progress bar, or text indicating the processing content such as "Searching".

[0041] As shown in Figure 2, once the search is complete, the search screen 50 may display a results area 60 for displaying the search results. The search screen 50 may also include a response comment 57 that is displayed along with the results area 60. The response comment 57 may be generated by the service server 20 using LLM19.

[0042] The result area 60 may include a plurality of display slots 61 arranged in one direction (for example, horizontally or to the right). Each of the plurality of display slots 61 may display information for the corresponding hit item. Each display slot 61 may display one or more attribute values ​​corresponding to one or more attribute items, such as item image, item name, item description, or price, as information (item data) for the corresponding hit item. The item data displayed in each display slot 61 can be arbitrarily changed.

[0043] At the initial display stage when the results area 60 is displayed on the search screen 50, the results area 60 displays the initial results, which are the search results based on the user query 54. The initial results may include multiple hit items, which are items that matched the search. Multiple hit items may be included in the item list of the initial results.

[0044] The item list may be stored in the memory 22 of the service server 20. The item list may contain multiple item datasets (referred to as "initial datasets") for multiple hit items. The item list may be, but is not limited to, a structured document such as JSON (JavaScript Object Notation, JavaScript is a registered trademark) or XML (Extensible Markup Language) format.

[0045] After obtaining the initial results, the service server 20 may analyze the initial results so that the search results better reflect the user query 54. The search results that reflect the analysis results are called secondary results. The service server 20 and the user terminal 30 may cooperate to transition the display of the search results from the initial results (see Figure 2) to the secondary results (see Figure 4).

[0046] The transition in display mode may be a change in the display within the result area 60. The secondary result may be displayed in the result area 60 where the initial result was displayed, replacing the initial result. In the secondary result, as a result of the analysis, at least one of the item list included in the initial result and the order (ranking) of multiple hit items included in the item list may be changed. If the initial result and the secondary result are the same, the display mode does not need to change.

[0047] For analysis purposes, the service server 20 may calculate a search score for each of the multiple hit items included in the initial results, indicating its relevance to the user query 54. As a result of the analysis, the service server 20 may change the ranking of the item list to a score-based order, or modify the item list based on the search score. The search score may be calculated by a machine learning model (e.g., generative AI).

[0048] The service server 20 may display a second loading indicator 58 on the search screen 50 of the user terminal 30 to indicate that the initial results are being analyzed during the analysis waiting time T2 (see Figure 5) between displaying the initial results and transitioning to a different display mode. Displaying the second loading indicator 58 during the analysis is called a second pending display.

[0049] The second loading indicator 58 may be of the same type as the first loading indicator 56, or it may be of a different type. For example, the second loading indicator 58 may be a dot animation, a spinner, a progress bar, or text such as "Analyzing," but is not limited to these.

[0050] As shown in Figure 3, if the order of hit items changes due to the transition from the initial result to the secondary result, the transition in the display mode in the result area 60 may be accompanied by an animation in which the multiple display slots 61 to be sorted move in one direction (e.g., horizontally) within the result area 60. As a result of this movement, the multiple hit items included in the item list of the secondary result are arranged from left to right in descending order of search score. The order of hit items may also be changed on the user terminal 30 side, for example by implementing a FLIP animation using JavaScript.

[0051] The transition in display mode may include, for example, one or more of the following 1) to 3), but is not limited to these. 1) The display slot 61 for the hit item whose rank has increased according to the score order moves to the first direction (for example, left). 2) The display slot 61 for the hit item whose ranking has dropped due to the score order moves to the second direction (for example, to the right), which is the opposite direction to the first direction. 3) The display slot 61 for hit items whose rank is lower than Nth place (where N is a predetermined number that is a natural number greater than or equal to 2) based on the score order fades out in the second direction.

[0052] As shown in Figure 4, once the transition in display mode is complete, the search screen 50 may display a suggestion message 59 from the AI ​​to the user instead of the second loading indicator 58. The suggestion message 59 may include one or more keywords to narrow down the search results or change the search conditions. For example, if the search results include "ice cream cake," the one or more keywords may be "whole cake," "chocolate," or "character cake."

[0053] The search results may include N display slots 61 (initial slots) (where N is a predetermined number, such as 5, a natural number greater than or equal to 2) that are displayed by default in the result area 60, and one or more additional slots (additional display slots) that are displayed in addition according to additional display instructions. Each of the multiple initial slots and the one or more additional slots displays information about the corresponding hit item. The additional slots may have the same configuration as the initial slots. The additional display instructions may be, for example, operations on one or more additional display elements.

[0054] One or more additional display elements may include, but are not limited to, a "Show All" button, a "See More" button, a page turn button, a slider 62, or an expansion button 63. If the initial slots are displayed in one direction (for example, to the right), operating the slider 62 may display one or more additional slots further to the right of the initial slots. Alternatively, operating the expansion button 63 may display one or more additional slots below the initial slots (in a direction perpendicular to the one direction) as the second row, if the N initial slots arranged in one direction are considered the first row.

[0055] [Response generated using AI] The service server 20 may input a response prompt to the LLM 19 in order to enable chat responses when the search screen 50 is displayed to the user terminal 30. The response prompt may include a response instruction and one or more response rules. The response instruction may be written as, for example, "Please respond to users who are searching for items according to the following response rules," but is not limited to this. The one or more response rules may include, for example, one or more of the following 1) to 6), but is not limited to these.

[0056] 1) First, the system outputs the standard explanatory text: "AI will help you with your shopping! Please enter your question." 2) After outputting the explanatory text, generate three example questions. 3) If the user query contains a search request (for example, a product name, or a description such as "I want...", "I want to buy...", or "Search for..."), a search command will be output. 4) When the search completion command is entered, a response comment is generated. 5) Following the search results, generate a suggestion message to prompt the user for further input. 6) If a user query includes a question (for example, "What kinds are there?" or "Tell me about..."), generate an answer to the question as a response comment.

[0057] For example, LLM19 first outputs display data, namely the explanatory text 52 and the example question 53, according to the response instructions. If the output from LLM19 is display data, the service server 20 sends that display data to the user terminal 30. When the user terminal 30 receives the display data, it displays the display data (explanatory text 52 and example question 53) on the search screen 50.

[0058] When user terminal 30 obtains user query 54, it sends user query 54 to service server 20. Service server 20 inputs user query 54 received from user terminal 30 into LLM 19. In this way, service server 20 mediates the response between user terminal 30 and LLM 19. However, if LLM 19 outputs a search command in response to user query 54, service server 20 performs the search based on user query 54 without sending the output to user terminal 30. In other words, the search command is not data for display, but an instruction command to service server 20.

[0059] When the service server 20 receives a search command from LLM19, it may extract query data from the user query 54. The query data may include text written in natural language. The query data may also include extracted data from photographs, images, or audio.

[0060] The extracted data may include feature data such as color, shape, pattern, or logo extracted from a photograph or image using LLM19 or other machine learning models, or it may include text data such as brand names or logos extracted by image analysis. Alternatively, the service server 20 may compare the photograph or image with item images in the item database and include data for matching items (e.g., item name or attribute values) in the query data. The query data may also include text data converted from speech by speech recognition using LLM19 or other machine learning models by the service server 20.

[0061] The service server 20 may obtain the search query output by the generating AI, LLM19, by inputting an extraction prompt to LLM19. In particular, if the user query 54 is written as a sentence rather than one or more keywords, it is advisable to extract the search query from the user query 54 and use it for detection.

[0062] The extraction prompt may include query data extracted from user query 54 and extraction instructions for extracting a search query from the query data. If LLM19 is VLM, the photograph or image may be entered directly into LLM19.

[0063] Extracting search queries may include, but are not limited to, preprocessing, morphological analysis, and normalization of the text contained in user query 54. Preprocessing may include, but are not limited to, removing whitespace and symbols, removing unnecessary particles, unifying uppercase and lowercase letters, and unifying full-width and half-width characters. Morphological analysis generally includes splitting into words. Normalization may include, but are not limited to, unifying synonyms, converting abbreviations to full terms, and correcting typographical errors. Extraction instructions may include examples of such processing.

[0064] The service server 20 receives the search query extracted by LLM19 based on the extraction prompt. The search query may include one or more keywords extracted from the user query 54. For example, if the user query 54 is "ice cream cake for a party" or "I want to buy an ice cream cake for a party," the search query may include the keywords "for a party" and "ice cream cake" extracted from this user query 54. In this case, the service server 20 searches the item database by keyword matching using the search query "ice cream cake for a party."

[0065] The service server 20 obtains the initial result, which is the result of searching the item database based on the user query 54 (search query). This completes the search of the item database by the service server 20.

[0066] When the search is complete, the service server 20 may input a search completion command to the LLM 19. In response to the search completion command, the LLM 19 outputs a response comment 57 for the user query 54. When the service server 20 receives the response comment 57 from the LLM 19, it sends the response comment 57 and the search results (initial results) to the user terminal 30. As a result, the user terminal 30 displays the response comment 57 and the initial results on the search screen 50.

[0067] If LLM19 determines that user query 54 is a question and does not contain a search request, it may output the answer to the question as a response comment without outputting a search command. When service server 20 receives the answer to the question from LLM19, it sends this answer to user terminal 30. As a result, user terminal 30 displays the answer to the question on search screen 50.

[0068] [How to search] The search method for this disclosure will be explained with reference to Figure 5. When a user enters information for a search into the input field 51 via the user terminal 30, in step S11, the user terminal 30 sends a user query 54 to the service server 20. When the service server 20 receives the user query 54, in step S21, it extracts a search query from the user query 54, for example using LLM 19.

[0069] After step S11, the user terminal 30 displays a first pending indicator in step S12. For example, the user terminal 30 displays a first loading indicator 56 on the search screen 50 as the first pending indicator.

[0070] Meanwhile, in step S22, following step S21, the service server 20 performs a search on the item database held by the data server 12 based on the extracted search query. In step S23, the service server 20 obtains initial results as search results. In step S24, the service server 20 sends the initial results, which are the search results, to the user terminal 30.

[0071] When the user terminal 30 receives search results from the service server 20, it stops the first pending display and, in step S13, displays the search results on the search screen 50. More specifically, the user terminal 30 displays the results area 60, which includes the initial results, on the search screen 50 instead of the first loading indicator 56.

[0072] Between steps S23 and S13, before displaying the initial results, the service server 20 or user terminal 30 may determine the initial display order, which is the order in which multiple hit items included in the initial results are arranged. The initial display order may be a simple sort based on one or more data items (e.g., price, rating score, or data update date), or it may be a rule-based ranking. The rule-based ranking may be based on the degree of relevance with the search query (e.g., exact match, partial match, mismatch), or it may be a combination of relevance and sorting. Relevance is an example of a degree of relevance.

[0073] The initial display order should be determined using a ranking method that can be decided more quickly than when determining the score order. Alternatively, the user terminal 30 may display the multiple hit items included in the initial results in no particular order.

[0074] The search waiting time T1 is the duration from when the user query 54 is obtained until the initial result is displayed on the service server 20. Alternatively, as shown in Figure 5, the search waiting time T1 may be defined as the duration from when the first pending display is started until the initial result is displayed on the user terminal 30.

[0075] After step S13, the user terminal 30 displays a second pending indicator in step S14. For example, the user terminal 30 displays a second loading indicator 58 on the search screen 50 as a second pending indicator.

[0076] Meanwhile, the service server 20 uses LLM19 or other generative AI to estimate the item type related to the search from the user query 54 in step S25 following step S24. The item type indicates the type or name of the item. The item type may match the category of the item (e.g., a subcategory), or it may be independently generated by LLM19. For example, if the user query 54 is "ice cream cake for a party," the service server 20 can estimate the item type related to the search as "ice cream cake."

[0077] In step S26, the service server 20 estimates the item type of each of the multiple hit items using LLM19 or other generative AI. For example, the service server 20 may estimate the item type based on at least a portion of the data contained in each of the multiple hit items' item datasets. For example, if the item name or description of a hit item includes "fruit ice cake," the service server 20 may estimate the item type of that hit item to be "ice cake."

[0078] In step S27, the service server 20 estimates item attributes from the user query 54 using LLM19 or other generative AI. Item attributes correspond to the attribute values ​​of search attributes. A search attribute is one of several attribute items that indicate the attributes of an item. The service server 20 may determine the search attributes based on the user query 54.

[0079] The search attribute may be an attribute item included in the item dataset, or it may be a unique attribute item not included in the item dataset. The search attribute may be selected by LLM19 from a pre-prepared list of attribute candidates. These attribute candidates may be generated in advance by LLM19 or another generative AI based on the item data.

[0080] For example, if user query 54 is "ice cream cake for a party," the service server 20 can infer the search term as "for a party." In this case, the search term can be said to be a usage attribute, a scene attribute, or an event attribute. These attribute items are not included as standard attribute items in the item dataset, but information about these attributes may be included in the item name or item description, for example.

[0081] In step S28, the service server 20 identifies the attribute value of the search item for each of the multiple hit items. For example, the service server 20 may identify the attribute value of the search item based on at least a portion of the data contained in each of the multiple hit items' item datasets (e.g., item name or item description).

[0082] For example, suppose the attribute fields in the item dataset do not include usage attributes, scene attributes, or event attributes. Even in this case, if the item name or item description of a hit item includes a description such as "for parties" or "for parties," the service server 20 may infer that the attribute value related to the search field of that hit item is "for parties."

[0083] The service server 20 may set multiple search items based on the user query 54. For example, if the user query 54 is "I want to buy a red women's T-shirt", the service server 20 may set "color", "item name", and "gender attribute" as search items. In this case, the attribute value of "color" estimated from the user query 54 is "red", the attribute value of "item name" is "T-shirt", and the attribute value of "gender attribute" is "women's".

[0084] In step S29, the service server 20 calculates a search score for each of the multiple hit items based on the results of steps S25 to S28. The search score may be calculated based on at least one of the following (A) or (B). (A) The degree of agreement between the item type estimated in step S25 and each item type estimated in step S26. (B) The degree of agreement between the item attributes estimated in step S27 and the attribute values ​​identified in step S28.

[0085] In step S30, the service server 20 may obtain the evaluation score for each of the multiple hit items from the item database or initial results. Furthermore, in step S31, the service server 20 may determine the score order of the multiple hit items based on the search score calculated in step S29 and the evaluation score obtained in step S30. For example, the score order may be determined such that the higher the search score, the higher the rank, and if the search scores are the same, the higher the evaluation score of the corresponding hit item.

[0086] The service server 20 may modify the initial results into secondary results by excluding one or more hit items whose search score falls below a threshold value from the item list (multiple hit items) included in the initial results. For example, the threshold value may be a search score of "2". In this case, hit items with a search score of "1" are excluded from the search results when the display mode changes. As a result, the modified search results, which are secondary results, will include multiple hit items (hereinafter also referred to as "selected items") whose search scores are equal to or greater than the threshold value (for example, a search score of "2" or higher).

[0087] As a result, the secondary results item list will include multiple selected items, and these multiple selected items will be ranked in order of score. Hit items excluded from the secondary results do not need to be assigned a score-based number. In this case, when the secondary results are ranked in order of score, items whose search score does not meet the threshold will be excluded from result area 60.

[0088] In step S32, the service server 20 sends the determined score order to the user terminal 30. Upon receiving the score order, the user terminal 30 transitions the display mode of the search results from the initial results to the secondary results in step S15. More specifically, the user terminal 30 removes items from the result area 60 that are not included in the score order from the hit items included in the search results, or changes the ranking of hit items in the result area 60 to score order.

[0089] After step S15, the user terminal 30 displays the suggested text 59 on the search screen 50 in place of the second loading indicator 58. This completes the display of search results. Thereafter, the user terminal 30 and the service server 20 execute the process shown in Figure 5 each time they obtain a new user query 54.

[0090] [Score ranking determined by generated AI] The service server 20 may use the generating AI LLM19 to perform at least part of the process for determining the score order. For example, after the search is complete and the initial results are obtained, the service server 20 may input an analysis prompt to LLM19 for analyzing the initial results.

[0091] The analysis prompt may include multiple analysis instructions for executing steps S25 to S29, respectively. Some of these analysis instructions may be combined into a single instruction. In addition to the multiple analysis instructions, the analysis prompt may include at least one of the following: a user query 54 used for the analysis (which may be query data or a search query) and an initial result.

[0092] The analysis instruction for step S25 could be written as, for example, "Analyze the user query to identify the product type related to the search (e.g., mobile phone, laptop, camera, clothing, accessories)." This analysis instruction is included in the analysis prompt along with user query 54.

[0093] The analysis instructions for step S26 could be written as, for example, "Analyze the product data for each hit item to identify the main product type (e.g., mobile phones, laptops, cameras, clothing, accessories)." These analysis instructions are included in the analysis prompt along with the initial results.

[0094] The analysis instruction for executing step S27 can be written, for example, as follows: "Identify the main product attributes specified in the user query. Examples of product attributes include color or pattern (e.g., red, blue, floral), size or dimensions (e.g., S, M, L, free size, 25 cm), brand, material (e.g., leather, cotton, metal), and gender (e.g., men's, women's, unisex)." This analysis instruction is included in the analysis prompt along with user query 54.

[0095] The analysis instructions for performing step S28 can be written as follows: "Analyze the product data for each hit item to identify key product attributes. Examples of product attributes include color or pattern (e.g., red, blue, floral), size or dimensions (e.g., S, M, L, free size, 25 cm), brand, material (e.g., leather, cotton, metal), and gender (e.g., men's, women's, unisex)." These analysis instructions are included in the analysis prompt along with the initial results.

[0096] To execute step S29, the service server 20 may input a calculation prompt to the generating AI LLM19, causing LLM19 to calculate a search score. The calculation prompt may include a calculation instruction and the results obtained in steps S25 to S28. The calculation instruction may be, for example, "For each item in the item list, analyze its relationship with the user query and assign a score from 1 to 3 based on the following criteria." Score 3: Perfect match, all item types and product attributes match. Score 2: Item type matches, but some or all product attributes do not match; partial match. It can be written as follows: "Score 1: Item type does not match, mismatch."

[0097] When a calculation prompt is entered, LLM19 outputs a search score of 1 to 3 for each of the multiple hit items. The search score is not limited to a ranking of 1 to 3. For example, points may be assigned such as 10 points for an exact match and 5 points for a partial match, or weights may be assigned to the item type and one or more attribute values ​​according to their importance. Furthermore, when calculating the search score, points may be added or subtracted based on other data items such as popularity, price, shipping cost, and stock status.

[0098] [Effects of this disclosure] When a user enters search criteria in the input field 51, initial results are displayed as search results in the results area 60 after a search waiting time T1 (for example, 8 seconds) has elapsed. Subsequently, after an analysis waiting time T2 (for example, 4 seconds) has elapsed, the display mode in the results area 60 transitions from initial results to secondary results. During this transition, hit items are sorted in order of score based on the search score, and hit items with low search scores are excluded.

[0099] Secondary results more accurately reflect the user's search intent by analyzing the initial results in greater depth. However, this type of analysis presents challenges, such as increased computational load on the service server 20 and longer waiting times for the final search results to be displayed due to the analysis waiting time T2. These challenges are particularly pronounced when optimizing rankings using machine learning, such as generative AI.

[0100] For example, when ranking popular items, the processing time increases in the following order. 3) and 4) are also called AI rankings. 1) Simple sorting 2) Rule-based rankings 3) Ranking using machine learning to perform feature calculations 4) Ranking using generative AI with language models

[0101] The accuracy of rankings is generally a trade-off with processing time. Longer waiting times for search results cause user stress, leading to a higher bounce rate from the site. Therefore, displaying the first loading indicator 56 during the search waiting time T1 can reduce user stress.

[0102] Furthermore, displaying initial results as provisional search results before analysis can reduce the impression that "search results are slow to display." Displaying the second loading indicator 58 during the analysis waiting time T2 also indicates that processing is underway, which contributes to reducing user stress. This is because users feel distrustful and stressed if the search screen 50 freezes after the search results (initial results) are displayed, or if the search results are subsequently changed.

[0103] Displaying ranking changes with animations makes it visually clear how search results have changed. This allows users to infer that a second pending result was due to this change, or to understand how the search results have evolved. This understanding contributes to reducing user stress.

[0104] If secondary results are displayed from the start, sorted by score, without displaying initial results, a significant waiting time (e.g., 12 seconds) occurs, making users feel that the search is slow. However, this disclosure reduces user stress while ensuring the accuracy of search results by sequentially transitioning between the search screens 50. This improves the user experience (UX) for item searches.

[0105] [Effects of this disclosure] According to this disclosure, the following effects can be achieved. (1) After temporarily displaying initial results as search results, the user's search intention By implementing a two-stage display that transitions to a secondary result reflecting the previous information, it is possible to achieve both speed in displaying search results and accuracy in ranking.

[0106] (2) By comparing the item attributes based on the user query 54 with the attribute values ​​of each hit item, it is possible to determine whether or not the search results include items that reflect the user's search intent.

[0107] (3) By using a generating AI that produces output in response to prompts, it is possible to flexibly calculate a search score regardless of the content and type of the user query 54, or the type of item or the content of the item data.

[0108] (4) By comparing the item type estimated from the user query with the item type of each hit item, it is possible to determine whether or not the search results include items that reflect the user's search intent.

[0109] (5) By including the evaluation score as a factor in determining the score order, the search results can reflect the user's search intent while simultaneously presenting users with highly-rated items. By suggesting better items to users in this way, user satisfaction can be increased.

[0110] (6) By analyzing the initial results, items with low search scores can be removed from the initial results list, thereby improving the search results to better reflect the user's search intent.

[0111] (7) When transitioning between search results, the movement of display slot 61 is displayed with animation, allowing the user to visually see how the ranking of the search results has changed.

[0112] (8) When transitioning search results, the direction of movement for increasing and decreasing the score order can be set to be opposite to indicate that the movement is for changing the ranking. In addition, the item list can be changed naturally by fading out the display slot 61 of the item that is excluded from the ranking along with the movement of the display slot 61 for changing the ranking.

[0113] (9) By displaying the first loading indicator 56 during the search waiting time T1 and the second loading indicator 58 during the analysis waiting time T2, the user can visually see that some processing is being carried out even after the search. By displaying the progress of the processing in this visible way, it is possible to show the user that the processing is proceeding normally. As a result, the stress that the user feels during waiting times can be reduced.

[0114] (10) By sorting or ranking the initial results using a rule-based method, the initial display order can be determined in a shorter processing time than AI ranking using a machine learning model (e.g., generative AI). Subsequently, by re-ranking the results by score, which is the AI ​​ranking method, it is possible to balance processing time and the accuracy of the search results.

[0115] (11) After the service server 20 sends the initial results to the user terminal 30, it sends the score order to the user terminal 30. In this case, the amount of data is less than if the service server 20 were to send all of the secondary results, including multiple selected items, to the user terminal 30. Therefore, the communication load on the user terminal 30 can be reduced, thereby shortening the analysis waiting time T2.

[0116] (12) The service server 20 performs a search using the search query extracted from the user query 54. Therefore, the user can enter a description of what they want in text or image form into the input field 51 without having to think of search keywords. In addition, the service server 20 can use a generation AI to obtain a search query that has been preprocessed, analyzed, or has intent estimated from the user query 54.

[0117] This embodiment can be implemented with the following modifications. This embodiment and the following modifications can be combined with each other to the extent that they do not contradict each other technically.

[0118] [Example of changing the alignment direction of hit items] Multiple display slots 61 may be arranged vertically in the results area 60. Alternatively, when the search screen 50 is displayed on a personal computer, the multiple display slots 61 may be arranged horizontally, while when it is displayed on a mobile device such as a smartphone, they may be arranged vertically.

[0119] [Example 1 of changing the method for calculating score ranking] When calculating a detection score based on (B), the calculation prompt may include the item attribute, the attribute values ​​of each of the multiple hit items, and a calculation instruction. The calculation instruction may be written as, for example, "Calculate the degree of match between the item attribute and the attribute values ​​of each of the multiple hit items," but is not limited to this.

[0120] When calculating the detection score based on both (A) and (B), the calculation prompt further includes the item type estimated in step S25 and the item type of each of the multiple hit items estimated in step S26. In this case, the calculation instruction is, for example, "[1] The degree of agreement between the item attribute and the attribute values ​​of each of the multiple hit items, and, [2] Calculate the search score based on the degree of match between the item type related to the search and the item type of each of the multiple hit items. The search score is: If both [1] and [2] are exactly the same, then 3. [1] If only this matches exactly, then 2. You could write it like this: "[1] If it does not match, set it to 1," but this is not the only way.

[0121] [Example 2 of changing the method for calculating score ranking] The score order may be calculated based on the following [1] to [3]. [1] The degree of agreement between the item's attribute and the attribute values ​​of each of the multiple hit items. [2] The degree of match between the item type related to the search and the item type of each of the multiple hit items. [3] Evaluation score for each hit item.

[0122] In this case, the search score may be determined as follows: If both [1] and [2] are exactly the same, and the evaluation score is equal to or greater than the threshold value (for example, if the maximum evaluation score is 5, then the evaluation score is 4.5), then the score is 4. If both [1] and [2] are a perfect match, and the evaluation score is below the threshold (for example, if the maximum evaluation score is 5, then the evaluation score is 4.5), then the score is 3. • If only [1] is a perfect match, the score is 2. • If [1] does not match, the score is 1.

[0123] [Examples of changes to the search score calculation method] The search score may be calculated using a trained machine learning model. For example, a machine learning model may be generated that learns the relationship between search queries and item data using past search history as training data. Alternatively, the LLM19 may be made to output the search score by inputting a prompt to the LLM19 that includes a user query 54, an item list, and a calculation instruction to calculate a search score based on the user query 54 and the item list.

[0124] [Example of changes to the search screen] The search screen 50 does not have to be a chat screen; it may also be a page on an e-commerce site with only an input field 51. The search screen 50 may be included in one of several e-commerce sites. Alternatively, the search screen 50 may provide a cross-search function that searches products from multiple e-commerce sites simultaneously. The input field 51 may be a keyword input field for entering only search keywords, not text. In this case, multiple keyword input fields may be provided for entering multiple keywords. Alternatively, only the display mode of the results area 60 may be adopted in a conventional search screen. Furthermore, the results area 60 is not limited to the display included in the user terminal 30; it may also be an external or network-connected display device to the user terminal 30. In any case, it is sufficient that the results area 60 of the graphical user interface, which is visible to the user, first displays multiple hit items as initial results, and then the order of the multiple hit items within the results area 60 of the graphical user interface transitions (changes) in order of score.

[0125] [Example 1 of changing the display mode transition] The transition from the initial result to the secondary result is not limited to the movement animation of the display slot 61. For example, to indicate that the initial result is a temporary display, the initial result may be displayed in a light color, or the display slot 61 may be accompanied by a slow, alternating light and dark animation. Furthermore, a message such as "Reviewing search results" may be displayed while the initial result is being displayed. In such cases, where the initial result itself indicates that it is a provisional result, the second loading indicator 58 may be omitted.

[0126] [Example 2 of changing the display mode transition] Transitions in the display mode may include changes in the size or color of the display slot 61, or changes in the content displayed in the display slot 61. For example, the size of the display slot 61 for higher-ranked scorers may be made larger than that of the display slot 61 for lower-ranked scorers, or the color of the display slot 61 for higher-ranked scorers may be changed to a more conspicuous color than that of the display slot 61 for lower-ranked scorers. In addition to or instead of this, the number of data items included in the display slot 61 for higher-ranked scorers may be greater than that of the display slot 61 for lower-ranked scorers, or a number or mark indicating the rank may be attached to the higher-ranked scorers (e.g., 1st to 3rd place).

[0127] [Example of changing search method] When a user logs in from the search screen 50 and performs a search, the user data of the logged-in user may be reflected in the search results. For example, the search score may be calculated by considering the logged-in user's click history, browsing history, or purchase history on the e-commerce site. Alternatively, or in addition to the above, the search score may also be calculated by considering the logged-in user's address, workplace, or location information of the user terminal 30.

[0128] [Example of changing search target] The search system disclosed herein is not limited to being linked to e-commerce sites, but may also be used for online searches of various types of information, such as books, literature, articles, music, videos, knowledge, or FAQs. The searched information may be used in information provision services such as job postings, real estate information, or transportation information, or in reservation services such as accommodation, facilities, or tickets.

[0129] [Example of changes to the generated AI] System 11 uses one LLM 19 for multiple purposes, but instead, multiple machine learning models (e.g., LLMs) can be used depending on the purpose. For example, multiple LLMs may include a chat LLM for responding to chats, an extraction LLM for extracting search queries, an analysis LLM for analyzing initial results, and a scoring LLM for calculating search scores. Multiple LLMs may have different configurations from each other, or some or all of them may be the same model.

[0130] [Examples of changes to flowcharts and configuration diagrams] The flowcharts and configuration diagrams of this disclosure illustrate the architecture, functionality, and operation of the apparatus, system, method, and program according to embodiments of this disclosure. Each step included in these flowcharts and each component included in the configuration diagrams may correspond to a part of a program containing one or more instructions for realizing a logical functional unit. In other embodiments, some of the illustrated steps may be omitted, other steps may be included, the order of the steps may be different, or some steps may be executed simultaneously. Furthermore, a flowchart described as a series of actions may be divided into several parts and executed, or multiple flowcharts may be executed sequentially or in relation to each other. Also in other embodiments, some of the illustrated components may be omitted, other components may be included, or the arrangement of components may be changed. Furthermore, the functions realized by these steps and components may be realized by hardware, software, or a combination of hardware and software.

[0131] [Example of changes to the search system] The various processes performed by the service server 20 and the data server 12 may be performed by one or more servers that execute those processes. For example, the multiple processes performed by the service server 20 may be distributed among a web server for displaying the search screen 50 on the user terminal 30, a search server for performing searches, and an AI server for input and output to the LLM 19. Alternatively, the service server 20 may maintain the item database and perform searches without going through the data server 12. Thus, the system 11 may be implemented as a single server (e.g., a computer), or it may be distributed among multiple devices (e.g., computers) or subsystems that collaborate to execute programs.

[0132] The statement "one or more processors execute the program code and perform the specified operation" means that if system 11 has multiple servers, at least one processor among those servers will execute the corresponding program code or instruction. If system 11 is implemented by a single server, it means that one or more processors on that server will execute the corresponding program code or instruction. In other words, "one or more processors" may be multiple server processors on different servers, or one or more server processors on any one server. Similarly, if the user terminal 30 executes part of the program code or instruction, "one or more processors" may include the processor of the user terminal 30 (terminal processor). The program code or instruction executed by one or more processors may be stored in the memory of the computer that has the processors that execute it.

[0133] [Example of memory modification] The memory included in the apparatus of this disclosure is a computer-readable storage medium, including non-transitory computer-readable medium. The memory may include, but is not limited to, ROM, hard disks, storage, removable media, flash memory, memory sticks, optical media, magneto-optical media, and CD-ROMs.

[0134] [Example of processor change] One or more processors included in the apparatus of this disclosure may include, but are not limited to, a CPU (central processing unit), a GPU (graphics processing unit), an APU (accelerated processing unit), an NPU (Neural network Processing Unit), a microprocessor, a microcontroller, a DSP (digital signal processor), an FPGA (field programmable gate array), a CPLD (Complex Programmable Logic Device), an application-specific integrated circuit (ASIC), a general-purpose processor, or any combination thereof designed to perform the functions described herein.

[0135] [Example of network changes] Communication between multiple devices or systems may be carried out via one or more communication networks in accordance with well-known communication protocols. The communication network may be, but is not limited to, an intranet, the internet, a local area network, a wide area network, a wireless network, a wired network, a virtual network, a software-defined network, or any other type of network, or a combination thereof.

[0136] [Example of changing the communication interface] A communication interface (IF) enables one device to communicate with other devices via a communication network. The communication interface may be, but is not limited to, a LAN (Local Area Network), Wi-Fi (registered trademark), Bluetooth (registered trademark), NFC (Near Field Communication), or other wireless communication interfaces.

[0137] The embodiments and modifications described above are listed below. [1] One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, The process involves determining the score order by ranking the aforementioned multiple hit items based on the search scores, The display mode of the search results shown on the user terminal is to be changed in order of the score, Search methods, including those mentioned.

[0138] [2] The aforementioned 1 or more processors The method involves estimating item attributes from the user query, wherein the item attributes correspond to the attribute values ​​of the search attributes, and the search attributes are one of several attribute items that represent the attributes of the item. For each of the aforementioned multiple hit items, the attribute value of the search attribute is identified, Includes, The search score is calculated based on the degree of correlation between the item attribute and the attribute values ​​of each of the multiple hit items. The search method described in [1] above.

[0139] [3] The aforementioned search score is calculated by the generating AI by inputting a calculation prompt into the generating AI. The calculation prompt includes the item attribute, the attribute value of each of the multiple hit items, and a calculation instruction to calculate the search score based on the degree of correlation between the item attribute and the attribute value of each of the multiple hit items. The search method described in [1] or [2] above.

[0140] [4] The aforementioned 1 or more processors Estimating the item type related to the search from the user query, To estimate the item type of each of the aforementioned multiple hit items, Includes, The search score is calculated based on the degree of relevance between the item type related to the search and each of the item types of the multiple hit items. The search method described in any of the above [1] to [3].

[0141] [5] The database includes multiple datasets, each corresponding to a set of items, and each dataset includes an evaluation score for the corresponding item. The aforementioned score order is: The higher the aforementioned search score, the higher the result, and If the search scores are the same, the higher the evaluation score of the corresponding hit item, the higher the result will be. The search method described in any of the above [1] to [4].

[0142] [6] The aforementioned 1 or more processors The initial results are modified into secondary results by excluding one or more hit items whose search score falls below a certain threshold. When transitioning the display mode, the search results include excluding one or more hit items that fall below the threshold value, including, The search method described in any of the above [1] to [5].

[0143] [7] The search results are displayed in a results area which includes multiple display slots arranged in one direction. Each of the aforementioned display slots displays information about the corresponding hit item. The transition of the display mode is accompanied by an animation in which the plurality of display slots, which are sorted in the order of the scores, move in one direction within the result area. A search method described in any of the above [1] to [6].

[0144] [8] When the number of the plurality of display slots is N, the transition of the display mode is: The display slots for hit items whose ranking has increased according to the aforementioned score order move in the first direction, The display slot for a hit item whose rank has dropped according to the aforementioned score order moves to the second direction, which is the opposite direction to the first direction. The display slots for hit items whose rank is lower than Nth place according to the score order fade out in the second direction, The search method described in any of the above [1] to [7].

[0145] [9] The aforementioned 1 or more processors During the search waiting time from obtaining the user query to displaying the initial results, a first loading indicator is displayed on the user terminal. During the analysis waiting period from the display of the initial results to the transition of the display mode, a second loading indicator is displayed on the user terminal. including, The search method described in any of the above [1] to [8].

[0146]

[10] The one or more processors include determining an initial display order, which is the order in which the multiple hit items included in the initial result are arranged, before displaying the initial result. The initial display order is determined by sorting by one or more data items or by a rule-based method. The aforementioned ranking is determined based on the search score calculated by the machine learning model. The search method described in any of the above [1] to [9].

[0147]

[11] The aforementioned 1 or more processors The process involves inputting an extraction prompt to a generating AI, wherein the extraction prompt includes the user query and an extraction instruction for extracting a search query for the search from the user query. The system receives the search queries extracted by the AI ​​generating the data, Includes, The initial result is the result of searching the database based on the search query. The search method described in any of the above [1] to

[10] .

[0148]

[12] It comprises one or more processors, and the one or more processors To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, The process involves determining the score order by ranking the aforementioned multiple hit items based on the search scores, The display mode of the search results shown on the user terminal is to be changed in order of the score, A search system configured to perform the following:

[0149]

[13] A service server configured to perform the aforementioned search, The user terminal and, The one or more processors mentioned above include one or more server processors provided by the service server and one or more terminal processors provided by the user terminal. The service server transmits the score order to the user terminal, The user terminal transitions the display mode of the initial results in the order of the scores, including, The search system described in

[12] above.

[0150]

[14] One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, The display of the search results shown on the user terminal is transitioned in a score order ranked based on the search score, wherein the search score is calculated to indicate the degree of relevance of each of the multiple hit items to the user query. A search program to execute [the command / action].

[0151]

[15] A search system for performing an item search on an online platform, wherein the search system One or more processors, A memory for storing program code executed by the one or more processors, The program code is provided with, A query code is configured to cause at least one of the one or more processors to receive a user query entered via a user terminal in order to search for the item, A search code configured to cause at least one of the one or more processors to search the database based on the user query, wherein the search result includes multiple hit items, A display code configured to cause the plurality of hit items to be displayed in the results area of ​​the graphical user interface of the user terminal is provided for at least one of the one or more processors. A score code configured to cause at least one of the one or more processors to calculate a search score indicating the degree of relevance to the user query for each of the multiple hit items, At least one of the one or more processors is configured with a ranking code that determines the score order of the multiple hit items based on the search score, At least one of the one or more processors is configured to have a transition code that changes the order of the multiple hit items in the result area of ​​the graphical user interface in the order of the scores, A search system that includes this. [Explanation of Symbols]

[0152] 11...Search system, 12...Data server, 13...Database, 19...LLM, 20...Service server, 21...Processor, 22...Memory, 23...Communication interface, 24...Program, 30...User terminal, 31...Processor, 32...Memory, 33...Communication interface, 34...Output device, 35...Input device, 50...Search screen, 51...Input field, 52...Description, 54...User query, 55...Login button, 56...First loading indicator, 57...Response comment, 58...Second loading indicator, 59...Suggestion text, 60...Result area, 61...Display slot, 62...Slider, 63...Expand button, T1...Search wait time, T2...Analysis wait time.

Claims

1. One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, This involves analyzing the initial results, and this analysis is performed by For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, This includes determining a score order for the aforementioned multiple hit items based on the search score, After the analysis waiting time for the aforementioned analysis has elapsed, the display of the search results shown on the user terminal will be changed to the order of the scores without any user intervention. Search methods, including those mentioned.

2. One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, wherein the search results are displayed in a results area including a plurality of display slots arranged in one direction, and each of the plurality of display slots displays information of the corresponding hit item. For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, The process involves determining the score order by ranking the aforementioned multiple hit items based on the search scores, The display mode of the search results shown on the user terminal is transitioned in order of the score, and the transition of the display mode is accompanied by an animation in which the plurality of display slots, which are sorted in order of the score, move in one direction within the result area. Search methods, including those mentioned.

3. When the number of the plurality of display slots is N, the transition of the display mode is: The display slots for hit items whose ranking has increased according to the aforementioned score order move in the first direction, The display slot for a hit item whose rank has dropped according to the aforementioned score order moves to the second direction, which is the opposite direction to the first direction. The display slots for hit items whose rank is lower than Nth place according to the score order fade out in the second direction, The search method described in claim 2.

4. One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, During the search waiting time from obtaining the user query to displaying the initial results, a first loading indicator is displayed on the user terminal. For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, The process involves determining the score order by ranking the aforementioned multiple hit items based on the search scores, The display mode of the search results shown on the user terminal is to be changed in order of the score, During the analysis waiting period from the display of the initial results to the transition of the display mode, a second loading indicator is displayed on the user terminal. Search methods, including those mentioned.

5. The one or more processors described above The method involves estimating item attributes from the user query, wherein the item attributes correspond to the attribute values ​​of the search attributes, and the search attributes are one of several attribute items that represent the attributes of the item. For each of the aforementioned multiple hit items, the attribute value of the search attribute is identified, Includes, The search score is calculated based on the degree of correlation between the item attribute and the attribute values ​​of each of the multiple hit items. A search method according to any one of claims 1 to 4.

6. The aforementioned search score is calculated by the generating AI by inputting a calculation prompt into the generating AI. The calculation prompt includes the item attribute, the attribute value of each of the multiple hit items, and a calculation instruction to calculate the search score based on the degree of correlation between the item attribute and the attribute value of each of the multiple hit items. The search method described in claim 5.

7. The one or more processors described above Estimating the item type related to the search from the user query, To estimate the item type of each of the aforementioned multiple hit items, Includes, The search score is calculated based on the degree of relevance between the item type related to the search and each of the item types of the multiple hit items. A search method according to any one of claims 1 to 4.

8. The database includes multiple datasets, each corresponding to a set of items, and each dataset includes an evaluation score for the corresponding item. The aforementioned score order is: The higher the aforementioned search score, the higher the result, and If the search scores are the same, the higher the evaluation score of the corresponding hit item, the higher the result will be. A search method according to any one of claims 1 to 4.

9. The one or more processors described above The initial results are modified into secondary results by excluding one or more hit items whose search score falls below a certain threshold. When transitioning the display mode, the search results include excluding one or more hit items that fall below the threshold value, including, A search method according to any one of claims 1 to 4.

10. The one or more processors include determining an initial display order, which is the order in which the multiple hit items included in the initial result are arranged, before displaying the initial result. The initial display order is determined by sorting by one or more data items or by a rule-based method. The aforementioned ranking is determined based on the search score calculated by the machine learning model. A search method according to any one of claims 1 to 4.

11. The one or more processors described above The process involves inputting an extraction prompt to a generating AI, wherein the extraction prompt includes the user query and an extraction instruction for extracting a search query for the search from the user query. The system receives the search queries extracted by the AI ​​generation system, Includes, The initial result is the result of searching the database based on the search query. A search method according to any one of claims 1 to 4.

12. The system comprises one or more processors, and the one or more processors To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, This involves analyzing the initial results, and this analysis is performed by For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, This includes determining a score order for the aforementioned multiple hit items based on the search score, After the analysis waiting time for the aforementioned analysis has elapsed, the display of the search results shown on the user terminal will be changed to the order of the scores without any user intervention. A search system configured to perform the following:

13. The system comprises one or more processors, and the one or more processors To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, wherein the search results are displayed in a results area including a plurality of display slots arranged in one direction, and each of the plurality of display slots displays information of the corresponding hit item. For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, The process involves determining the score order by ranking the aforementioned multiple hit items based on the search scores, The display mode of the search results shown on the user terminal is transitioned in order of the score, and the transition of the display mode is accompanied by an animation in which the plurality of display slots, which are sorted in order of the score, move in one direction within the result area. A search system configured to perform the following:

14. The system comprises one or more processors, and the one or more processors To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, During the search waiting time from obtaining the user query to displaying the initial results, a first loading indicator is displayed on the user terminal. For each of the aforementioned multiple hit items, a search score indicating the degree of relevance to the user query is calculated, The process involves determining the score order by ranking the aforementioned multiple hit items based on the search scores, The display mode of the search results shown on the user terminal is to be changed in order of the score, During the analysis waiting period from the display of the initial results to the transition of the display mode, a second loading indicator is displayed on the user terminal. A search system configured to perform the following:

15. A service server configured to perform the aforementioned search, The user terminal and, The one or more processors include one or more server processors provided by the service server and one or more terminal processors provided by the user terminal. The service server transmits the score order to the user terminal, The user terminal transitions the display mode of the initial results in the order of the scores, including, A search system according to any one of claims 12 to 14.

16. One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, After the analysis waiting time for the analysis of the initial results has elapsed, the display of the search results shown on the user terminal is transitioned in a score order ranked based on the search score, without any user operation, wherein the analysis includes the calculation of the search score, and the search score is calculated to indicate the degree of relevance of each of the multiple hit items to the user query. A search program to execute [the command / action].

17. One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, wherein the search results are displayed in a results area including a plurality of display slots arranged in one direction, and each of the plurality of display slots displays information of the corresponding hit item. The display mode of the search results shown on the user terminal is transitioned in a score order ranked based on the search score, wherein the search score is calculated for each of the multiple hit items to indicate the degree of relevance to the user query, and the transition of the display mode is accompanied by an animation in which the multiple display slots, sorted in the score order, move in one direction within the results area. A search program to execute [the command / action].

18. One or more processors, To retrieve user queries entered via the user terminal to search for items, Obtaining initial results, which are the results of searching the database based on the user query, wherein the initial results include multiple hit items, which are items that matched the search. The initial results are displayed on the user terminal as search results for the user query, During the search waiting time from obtaining the user query to displaying the initial results, a first loading indicator is displayed on the user terminal. The display of the search results shown on the user terminal is transitioned in a score order ranked based on the search score, wherein the search score is calculated to indicate the degree of relevance of each of the multiple hit items to the user query. During the analysis waiting period from the display of the initial results to the transition of the display mode, a second loading indicator is displayed on the user terminal. A search program to execute [the command / action].

Images