JP7150090B2 - Product Attribute Extraction Method for Shopping Search - Google Patents

Description

TECHNICAL FIELD The present disclosure relates to a method and system for extracting product attributes for shopping search, and more specifically, extracting product attributes from a user query or search query input for shopping search to determine product search ranking. It relates to a method and system that can reflect

Recently, various services such as shopping, games, movies, and music are provided online in the Internet and mobile environment. In particular, online shopping is used by a large number of users because there is no time limit for purchasing products and it is possible to select and purchase products without directly visiting a product store.

Due to the convenience of online shopping, the number of users who use online shopping is increasing, and the number and types of products sold or handled online are increasing exponentially. Accordingly, there is a problem that it is difficult for the user to search for a product that matches his/her intention among various products sold online.

On the other hand, when a user connects to an online shopping system and inputs a query to search for a desired product, the product search results including product information related to the query and connection information regarding purchasers are displayed. A list can be provided. Since such an online shopping system simply compares a user query with product information in a product database to extract search results, it is difficult to derive product search results that match the intention of the user included in the query.

Korea Registered Patent No. 10-1017016

The present disclosure has been made to solve the above problems, and provides a product attribute extraction method, a computer program, and an apparatus (system) for shopping search.

In addition, the present disclosure provides a method and apparatus (system) capable of extracting more accurate product attributes by using a statistically-based classification model for product attribute extraction for shopping searches.

The present disclosure can be embodied in various forms including methods, apparatus (systems), and computer programs.

According to an embodiment of the present disclosure, a product attribute extraction method for shopping search performed by a computer system comprises: receiving a user query; extracting product search results for the user query from a product database; According to the classification model, calculating the relevance score of the user query and the plurality of product attributes, and adjusting the product search ranking according to the product search result based on the calculated relevance score.

According to another embodiment of the present disclosure, a product attribute extraction method for shopping search performed on a user terminal comprises: receiving a user query; The method includes the steps of: receiving relevance scores of product attributes; and adjusting product search rankings for user queries based on the calculated relevance scores.

According to yet another embodiment of the present disclosure, there is provided a computer program for executing the product attribute extraction method for shopping search described above on a computer.

According to yet another embodiment of the present disclosure, a learning method for product attribute extraction for shopping search performed by a computer system comprises a plurality of first user queries and a plurality of queries associated with the plurality of first user queries. generating a plurality of first learning data including the first product attributes of and, based on the plurality of first learning data, determining the relevance score of the one or more product attributes corresponding to the second user query training a statistically-based classification model to compute.

According to yet another embodiment of the present disclosure, there is provided a computer program for executing the above learning method for product attribute extraction for shopping search on a computer.

According to yet another embodiment of the present disclosure, a product attribute extraction system for shopping search includes a communication module for receiving a user query, a memory, and computer readable instructions coupled to the memory, comprising: at least one processor configured to execute At least one processor extracts product search results for user queries from a product database, calculates relevance scores of user queries and a plurality of product attributes using a statistically-based classification model, and calculates the relevance scores based on the calculated relevance scores. Adjust the product search ranking according to product search results.

According to various embodiments of the present disclosure, it is possible to output to the user terminal a search result reflecting the product attributes intended by the user based on the user query. In addition, according to various embodiments of the present disclosure, by displaying products having product attributes related to user queries at the top of search results, users' click rate and purchase rate for the products can be improved. can.

According to various embodiments of the present disclosure, improving the accuracy of the relevance of product attributes to user queries by using a statistically-based classification model to calculate relevance scores of multiple product attributes to user queries. can be made

According to various embodiments of the present disclosure, by dividing a user query into units of syllables and/or morphemes and calculating relevance scores for product attributes using a statistically-based classification model, it is possible to It is also possible to improve the accuracy of calculation of the associated score of product attributes for queries. In addition, the matching coverage of the learning data of the statistical-based classification model is maximized, and the accuracy of calculation of user queries and related scores of multiple product attributes can be further improved.

According to various embodiments of the present disclosure, the learning data used for learning a statistically-based classification model that calculates the relevance scores of multiple product attributes for a user query may be integrated search log data, shopping or product search log data, or the like. It can be generated from various search log data such as In addition, learning data can be automatically generated based on user queries, search results, and user actions in response to search results among search log data, so that learning data for the statistical-based classification model can be efficiently secured. .

According to various embodiments of the present disclosure, by displaying product attributes as product attribute codes, the same or similar product attributes are associated with different product types or user queries, thereby improving the accuracy of product attribute extraction results. It can prevent deterioration of performance. That is, by displaying product attributes as product attribute codes and learning a statistically-based classification model, it is possible to improve the accuracy of calculation of related points even for ambiguous product attributes and range-type product attributes.

Embodiments of the present disclosure will be described with reference to the following accompanying drawings. Here, like reference numbers indicate like elements, but are not so limited.
FIG. 5 is a diagram illustrating a screen displaying search results reflecting product attributes extracted from a user query by a product attribute extraction method for shopping searches according to an embodiment of the present disclosure; 1 is a schematic diagram illustrating a system in which a plurality of user terminals and a product attribute extraction server are communicatively coupled to provide a product attribute extraction service for shopping search according to an embodiment of the present disclosure; FIG. It is a block diagram showing the internal configuration of a user terminal and a product attribute extraction server according to an embodiment of the present disclosure. FIG. 4 is a diagram showing a configuration for calculating a product attribute related score for a user query using a statistically based classification model according to an embodiment of the present disclosure; FIG. 4 is a diagram showing a configuration for calculating a relevance score using a classifier including a preprocessor and a statistically-based classification model according to an embodiment of the present disclosure; FIG. 5 is a diagram illustrating an example of generating learning data of a statistically-based classification model for extracting product attributes from various search results according to an embodiment of the present disclosure; FIG. 4 is a diagram illustrating a method of generating learning data by representing product attributes with product attribute codes according to an embodiment of the present disclosure; 4 is a flowchart illustrating a product attribute extraction method for shopping search performed by a product attribute extraction server according to an embodiment of the present disclosure; 4 is a flowchart illustrating a product search ranking adjustment method for shopping search according to an embodiment of the present disclosure; 4 is a flowchart illustrating a product attribute extraction method for shopping search performed by a user terminal according to an embodiment of the present disclosure; 4 is a flowchart illustrating a statistically-based classification model learning method for product attribute extraction according to an embodiment of the present disclosure;

Hereinafter, specific contents for carrying out the present disclosure will be described in detail based on the accompanying drawings. however,
In the following description, specific descriptions of well-known functions and configurations are omitted when they may unnecessarily obscure the subject matter of the present disclosure.

In the accompanying drawings, the same or corresponding components are provided with the same reference numerals. Also, in the description of the embodiments below, duplicate descriptions of the same or corresponding components may be omitted. However, the omission of a description of an element should not be construed as omitting such element from being included in some embodiments.

Advantages and features of the disclosed embodiments, and the manner in which they are achieved, will become apparent with reference to the embodiments described below with reference to the accompanying drawings. The present disclosure may, however, be embodied in various different forms and should not be construed as limited to the embodiments set forth below. However, the examples are provided only so that this disclosure will be complete and will enable those of ordinary skill in the art to accurately recognize the categories of the invention.

A brief description of terms used herein and a specific description of the disclosed embodiments are provided. The terms used in this specification have been selected as common terms that are currently widely used as much as possible while considering the function in this disclosure, but this is not the intention or precedent of an engineer engaged in the related field. , may change due to the emergence of new technologies. Also, certain terms may be arbitrarily chosen by the applicant, the meaning of which will be detailed in the description of the invention. Accordingly, terms used in this disclosure should be defined based on their meanings and overall context of this disclosure, rather than simply terminology.

In this specification, singular expressions may include pluralities and plural expressions may include the singular unless the context clearly dictates otherwise.

Throughout the specification, when a part "includes" an element, this does not exclude or may include other elements, unless specifically stated to the contrary. means you can.

Also, the term "module" or "unit" as used herein means a software or hardware component that performs a certain role. However, "module" or "unit" is not meant to be limited to software or hardware. A "module" or "unit" may be configured to reside on an addressable storage medium and configured to run on one or more processors. Thus, by way of example, a "module" or "part" may refer to components such as software components, object-oriented software components, class components, task components, as well as processes, functions, attributes, procedures, subroutines, It may include at least one of program code segments, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays or variables. Components and "modules" or "sections" may be combined with a smaller number of components and "modules" or "sections" or may be combined with additional components and "modules" or "sections". ” can be further separated into

According to one embodiment of the present disclosure, a "module" or "unit" may be embodied with a processor and memory. "Processor" should be interpreted broadly to include general purpose processors, central processing units (CPUs), microprocessors, digital signal processors (DSPs), controllers, microcontrollers, state machines, and the like. In some environments, a "processor" can also refer to an application specific semiconductor (ASIC), programmable logic device (PLD), field programmable gate array (FPGA), and the like. A "processor" may be, for example, a combination of a DSP and a microprocessor, a combination of multiple microprocessors, a combination of one or more microprocessors combined with a DSP core, or any other such configuration. It can also refer to any combination of processing devices. Also, "memory" should be interpreted broadly to include any electronic component capable of storing electronic information. "Memory" includes RAM (Random Access Memory), ROM (Read Only Memory), NVRAM (Non-Volatile Random Access Memory), PROM (Programmable Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM ( It can also refer to various types of processor-readable media such as electrically erasable programmable reda-only memory), flash memory, magnetic or optical data storage devices, registers, and the like. Memory is said to be in electronic communication with a processor when the processor can read information from, or record information in, the memory. Memory integrated with the processor is in electronic communication with the processor.

In the present disclosure, a "user query" can include, but is not limited to, text (eg, words, phrases, sentences, etc.) that can be entered and searched using a search engine. For example, a user query includes text that a user inputs through an input UI of a search program such as a web browser, search application, or shopping application, or text that a user can select from a query list provided by a search program by clicking on it. can contain In other examples, user queries are not limited to text, but can include text or information that can be extracted from multimedia content such as images, videos, sounds, voices, and the like.

In the present disclosure, the 'product attribute' may indicate various information about the product, including product name, product number, standard or size, shape, color, material, performance, price, and the like. For example, the product attributes may be provided to the user as a result of a shopping search or product search along with connection information regarding the purchaser of the product (eg, link information regarding a product purchase web page).

In the present disclosure, a “statistics-based classification model” is a classifier trained to classify new unlabeled information based on its relevance to known labeled data, or Classification models can be presented generically. For example, statistically based classification models include language model, SVM (support vector machine), maximum entropy model, CRF model (conditional random field model), FFNN (feed-forward neural network), RNN (recurrent neural network), GRU (gated recurrent unit), LSTM (long short-term memory), transformer model, BERT model (bidirectional encoder representations from transformers model), etc. It can be any one of a neural network model.

In this disclosure, "embedding" can refer to the technique of transforming a categorical or discrete variable into a continuous vector. Embedding can be used to reduce the dimensionality of information and derive meaning from categorical variables, such as user queries, by transforming them into continuous vectors. Also, if a user query is represented as a vector in an embedded vector space, neighborhood vectors, etc. on that vector space can share similar or related properties. For example, a method of embedding user queries in a vector space can be implemented by training an artificial neural network.

FIG. 1 is a diagram illustrating a screen 100 displaying search results 130 reflecting product attributes extracted from a user query 110 by a product attribute extraction method for shopping searches according to an embodiment of the present disclosure. As shown in the figure, when a user inputs a user query 110 into a search window of a shopping site, a search result 130 reflecting product attributes 134 and 138 extracted from the user query 110 can be displayed. The search result 130 includes images and names 132 and 136 indicating the searched product, product attributes 134 and 138, and the like. For example, when a user inputs "mini-notebook" as a search query, products 132 and 136 related to "mini", which is an expression related to the size or standard of the "notebook" computer, which is the product included in the query, are displayed. , the ranking of the search results may be adjusted so that they appear at the top of the search results 130 . That is, in the search result 130, among various notebook (or laptop) computer products, products 132 and 136 with screen sizes of '10 inches' and '11 inches' are preferentially displayed at the top of the search results 130. can be displayed in Therefore, through the search result 130, the user can easily and quickly check the information of the products 132, 136 related to the product attributes 134, 138 that can be extracted from the product search query entered by the user. In addition, the user can select or click any one of the products 132 and 136 as necessary to obtain more specific information about the selected product, or perform online purchases.

Further, in screen 100, above search results 130, a user interface 120 may be displayed that allows filtering or selection of product attributes. Accordingly, from the user interface 120, the user selects one or more of the product attributes associated with the user query 110, such as product category, manufacturer, screen size, etc., to obtain the products 132, 136 can be further adjusted. For example, when the user selects 'Company A' from among the manufacturer filtering items from the user interface 120, information such as products manufactured by 'Company A' may be displayed at the top of the search results 130 .

In one embodiment, when the user query 110 is input, the relevance or relevance probability of a plurality of product attributes to the user query 110 can be calculated by the product attribute extraction method. For example, the product attributes of a notebook computer may include color, case material, CPU performance, memory capacity, screen size and resolution. As another example, the product attributes of a home TV may include color, display type, size, resolution, internet connectivity, and the like. When the user query 110 is "mini-notebook", the product attribute extraction method includes not only multiple product attributes of notebook computers, but also multiple product attributes of different types of products such as household TVs, fashion miscellaneous goods, and lifestyle and health products. Relevance can also be calculated for product attributes. When the user query 110 is "mini-notebook", the relevance to the small screen size attribute (for example, "10 inches") is calculated to be the highest among the product attributes of the notebook computer. Relevance to different product attributes of computers and product attributes of different types of products can be calculated to be low.

On the other hand, a product search result 130 for the user query 110 is obtained from the product database. A product search based on the user query 110 can be executed by extracting product information highly relevant to the user query 110 from product information stored in the product database. In the product search results 130 extracted in this manner, product information highly relevant to the user query 110 may be displayed at a higher rank than product information less relevant. In addition, the product search ranking according to the product search results 130 can be adjusted based on the relevance (or related score) of product attributes to the user query 110 calculated as described above. According to this, when the user query 110 is “mini-notebook”, the product search results 130 may display product 132 and 136 information related to notebook computers with small screen sizes at the top.

In addition, according to the product attribute extraction method for the user query 110, the user query 110 is divided into syllable and/or morpheme units and analyzed, so that a plurality of product attributes related to the user query 110 are analyzed. Score can be calculated. In one embodiment, the method for extracting product attributes can calculate the relevance score of a plurality of product attributes with respect to the user query 110 by analyzing the user query 110 by dividing it into syllable units. In addition, the product attribute extraction method divides the user query 110 into morpheme units and analyzes them, so that the relevance score of a plurality of product attributes to the user query 110 can be calculated. The two relevance scores extracted as described above are used as the final relevance score by summing or combining.

The product attribute extraction method for the user query 110 can calculate the related score of product attributes using a statistically based classification model. A statistically-based classification model can be trained to calculate the probabilities that a user query 110 is associated with multiple product attributes. In one embodiment, the statistically-based classification model is trained to take as input a user query 110 segmented into syllables and/or morphemes and calculate the probabilities (or association scores) associated with all possible product attributes. can. For example, when "mini notebook" is input as the user query 110, the probability value corresponding to the product attributes 134, 138 of "10 inches" or "11 inches" with a small screen size is "17 inches" with a large screen size. ” can be calculated to be larger than the probability value corresponding to the product attribute.

According to the method described above, the search rankings of the products 132 and 136 included in the search results 130 can be adjusted based on the related scores of the product attributes calculated by the statistically based classification model. The product search ranking (or product search ranking score) may be determined based on the lexical similarity of the product name or product attribute to the user query 110, user popularity of the product, and the like. At this time, a final search ranking score of the products 132 and 136 may be determined by adding or weighting the related scores of the product attributes to the search ranking scores of the products 132 and 136 . In one embodiment, the search rankings of the products 132 and 136 may be adjusted based on the highest relevance score among the relevance scores of the product attributes calculated for the user query 110 . Alternatively, the search rankings of the products 132 and 136 may be adjusted based on the top n (n is an integer greater than 1) related scores among the related scores of the product attributes calculated for the user query 110 .

In another embodiment, the product attribute extraction method for user query 110 is used to generate or update product attributes for product information from a product database. For example, the method for extracting product attributes is to calculate the associated scores of all possible product attributes based on the product name of the product information stored in the product database, and select the product attribute associated with the highest Can be stored or updated as attributes.

In the example shown in FIG. 1, product search results 130 or product search rankings determined by the product attribute extraction method for shopping search of the present disclosure are provided via a website or application for product search or shopping. However, it is not limited to this. Methods according to various embodiments of the present disclosure can also be provided via a search service of an integrated search site or a search application executed by a user terminal.

FIG. 2 illustrates a system 200 in which a plurality of user terminals 210_1, 210_2, 210_3 and a product attribute extraction server 230 are communicatively coupled to provide a product attribute extraction service for shopping search according to an embodiment of the present disclosure. It is a schematic diagram showing As shown in the figure, the system 200 includes a product attribute extraction server 230 that provides a product attribute extraction service for shopping searches, and a plurality of user terminals 210_1 and 210_2 connected to the product attribute extraction server 230 via a network 220. , 210_3 and a search log database 240 . According to one embodiment, the product attribute extraction server 230 is one or more computer-executable programs (e.g., downloadable applications) that can store, provide, and execute data for providing product attribute extraction services. It can include server devices and/or databases, and one or more distributed computer devices and/or distributed databases of cloud computing service infrastructure. The product attribute extraction service provided by the product attribute extraction server 230 can be provided to the user via a search application or web browser installed in each of the plurality of user terminals 210_1, 210_2, 210_3.

A plurality of user terminals 210_1, 210_2, and 210_3 can communicate with the product attribute extraction server 230 via the network 220. FIG. The network 220 can be configured to allow communication between multiple user terminals 210 and product attribute extraction servers 230 . The network 220 may be a wired network 220 such as Ethernet, PLC (Power Line Communication), telephone line communication device and RS-serial communication, mobile communication network, WLAN (Wireless LAN), etc., depending on the installation environment. , Wi-Fi, Bluetooth and ZigBee, etc., or a combination thereof. The communication method is not limited, and the user terminals 210_1 and 210_2 as well as the communication method using a communication network (eg, mobile communication network, wired Internet, wireless Internet, broadcasting network, satellite network, etc.) that can include the network 220. , 210_3 is also included. For example, the network 220 includes a PAN (personal area network), a LAN (local area network), a CAN (campus area network), a MAN (metropolitan area network), a WAN (wide area network), a BBN (broadband network), the Internet, and the like. Any one or more of the networks can be included. Network 220 may also include any one or more of network topologies including bus networks, star networks, ring networks, mesh networks, star-bus networks, tree or hierarchical networks, etc. It is not limited to this.

In FIG. 2, a mobile phone or smart phone 210_1, a tablet computer 210_2, and a laptop or desktop computer 210_3 are shown as examples of user terminals, but the user terminals 210_1, 210_2, and 210_3 are wired and/or wireless. It can be any computing device capable of communication and on which a search application, mobile browser application or web browser can be installed and executed. For example, the user terminal 210 includes a smart phone, a mobile phone, a navigation terminal, a desktop computer, a laptop computer, a digital broadcast terminal, a PDA (Personal Digital Assistants), a PMP (Portable Multimedia Player), a tablet computer, a game console (game console), wearable device, internet of things (IoT) device, virtual reality (VR) device, augmented reality (AR) device, and the like. In addition, although FIG. 2 shows that three user terminals 210_1, 210_2, and 210_3 communicate with the product attribute extraction server 230 via the network 220, the present invention is not limited to this, and different numbers of user terminals communicate with the network 220. It can also be configured to communicate with the product attribute extraction server 230 via.

According to one embodiment, the product attribute extraction server 230 can receive one or more user queries input from a plurality of user terminals 210_1, 210_2, and 210_3. According to one embodiment, the product attribute extraction server 230 can receive user queries entered into a search engine of a search site displayed by a web browser running on a plurality of desktop or laptop computers 210_3. According to another embodiment, the product attribute extraction server 230 extracts user queries entered into search engines of search sites displayed by mobile browser applications running on a plurality of mobile phones, smart phones and tablet computers 210_1, 210_2. can be received. According to another embodiment, the product attribute extraction server 230 can receive user queries entered into a search engine of an online shopping site that sells products.

In addition, the product attribute extraction server 230 can calculate the associated score of a plurality of product attributes with respect to the received user query. In one embodiment, the product attribute extraction server 230 may include a statistically-based classification model trained to calculate the relevance score of product attributes. The product attribute extraction server 230 divides the received user query into units of syllables and/or morphemes, and uses the divided user queries of units of syllables and/or morphemes as inputs for a statistically-based classification model that calculates a relevance score. can. Also, the product attribute extraction server 230 can receive learning data for learning a statistical-based classification model from the search log database 240 connected through the network 220 . For example, the learning data received from the search log database 240 includes a pair of user queries and product information (e.g., attribute information of specific products) selected by the user from product search results corresponding to the user queries. can be done.

The product attribute extraction server 230 can extract product attributes related to one or more top-level related scores among a plurality of product attributes, and search product information including the product attributes from a product database (not shown). The ranking of the retrieved product information can be readjusted based on the associated score of product attributes. In addition, the product attribute extraction server 230 can adjust the product search order based on the product attribute weighting value that determines the importance of the product attributes related to the top related scores and the relevant related scores. Here, the product attribute weight value can be changed according to the user's response to the product search result (for example, the number of times the user selects the top product search result).

FIG. 3 is a block diagram showing internal configurations of the user terminal 210 and the product attribute extraction server 230 according to an embodiment of the present disclosure. The user terminal 210 can represent any computer device capable of executing a web browser or an application and capable of wire/wireless communication. It may include a computer 210_2, a laptop or desktop computer 210_3, and so on. As shown in FIG. 3, user terminal 210 may include memory 312 , processor 314 , communication module 316 and input/output interface 318 . Similarly, item attribute extraction server 230 may include memory 332 , processor 334 , communication module 336 and input/output interface 338 . As shown in FIG. 3 , user terminal 210 and product attribute extraction server 230 can be configured to communicate information and/or data over network 220 using respective communication modules 316 , 336 . The input/output device 320 is also configured to input information and/or data to the user terminal 210 and output information and/or data generated from the user terminal 210 via the input/output interface 318. can.

Memory 312, 332 may include any non-transitory computer-readable medium. According to one embodiment, the memory 312, 332 is a permanent memory such as random access memory (RAM), read only memory (ROM), disk drive, solid state drive (SSD), and flash memory. A permanent mass storage device may be included. As another example, a permanent large-capacity storage device such as a ROM, an SSD, a flash memory, a disk drive, etc. may be used as a permanent storage device separate from the memories 312 and 332, and the user terminal 210 or the product attribute extraction server may be used. 230. In addition, the memories 312 and 332 store an operating system and at least one program code (for example, a code for executing a product attribute extraction method installed and driven by the user terminal 210 or the product attribute extraction server 230). obtain.

Such software components can be loaded from a computer-readable medium separate from the memories 312,332. Such separate computer-readable recording media can include recording media directly connectable to the user terminal 210 and the product attribute extraction server 230, such as floppy drives, discs, tapes, DVDs, etc. /CD-ROM drives and computer readable recording media such as memory cards. As another example, software components and the like can be loaded into memory 312, 332 via communication modules 316, 336, which are not computer-readable media. For example, at least one program is a computer program installed by a file provided via the network 220 by a developer or a file distribution system that distributes an application installation file (for example, a product attribute extraction service for shopping searches). It can be loaded into the memory 312, 332 based on the application serving).

The processors 314, 334 can be configured to process computer program instructions by performing basic arithmetic, logic and input/output operations. Instructions may be provided to processors 314 , 334 by memory 312 , 332 or communication modules 316 , 336 . For example, the processors 314,334 can be configured to execute instructions received by program code stored in a storage device, such as the memory 312,332.

Communication modules 316 and 336 can provide configurations and functions for mutual communication between user terminal 210 and product attribute extraction server 230 via network 220, and user terminal 210 and/or product attribute extraction server 230 can communicate with each other. configuration or functionality for communicating with other user terminals or other systems (eg, a separate search engine system, etc.). For example, the processor 314 of the user terminal 210 requests or data generated by program codes stored in a recording device such as the memory 312 (e.g., a search result output request considering product attributes for a user query) may be sent to the communication module. Control at 316 may be communicated to product attribute extraction server 230 via network 220 . Conversely, control signals or instructions provided by the control of processor 334 of product attribute extraction server 230 may be received by user terminal 210 via communication module 316 of user terminal 210 via communication module 336 and network 220 . For example, the user terminal 210 can receive, from the product attribute extraction server 230 through the communication module 316, a product search result corresponding to a user query, related scores of a plurality of product attributes, product search results reflecting the calculated related score, and the like. .

Input/output interface 318 may be a means for interfacing with input/output devices 320 . By way of example, input devices may include devices such as cameras including image sensors, keyboards, microphones, mice, etc., and output devices may include devices such as displays, speakers, haptic feedback devices, etc. . As another example, the input/output interface 318 may be a means for interfacing with a device such as a touch screen that has a single configuration and functions for performing input and output. For example, when the processor 314 of the user terminal 210 processes instructions of a computer program loaded in the memory 312, information and/or data provided by the product attribute extraction server 230 or other user terminals 210 may be used. Service screens and product search results can be displayed on the display via the input/output interface 318 . Although the input/output device 320 is not included in the user terminal 210 in FIG. In addition, the input/output interface 338 of the product attribute extraction server 230 is connected to the product attribute extraction server 230 or communicates with a device (not shown) for input or output that the product attribute extraction server 230 may include. It can be a means for interfacing. Although FIG. 3 shows the input/output interfaces 318 and 338 as elements configured separately from the processors 314 and 334, the present invention is not limited to this and the input/output interfaces 318 and 338 are configured to be included in the processors 314 and 334. You can also

The user terminal 210 and product attribute extraction server 230 can include more components than those shown in FIG. However, it is not necessary to explicitly show most prior art components. According to one embodiment, the user terminal 210 can be implemented to include at least a portion of the input/output device 320 described above. In addition, the user terminal 210 may further include other components such as a transceiver, a GPS (Global Positioning System) module, a camera, various sensors and a database. For example, if the user terminal 210 is a smart phone, it can generally include components that a smart phone has, such as an acceleration sensor, a gyro sensor, a camera module, various physical buttons, buttons using a touch panel, input/output ports, and Various components such as a vibrator for vibration may be further included in the user terminal 210 .

According to one embodiment, the processor 314 of the user terminal 210 can be configured to run a search application, mobile browser application, or web browser capable of connecting to a product search or shopping site. At this time, program code associated with the application or web browser can be loaded into the memory 312 of the user terminal 210 .

The processor 314 of the user terminal 210 receives information and/or data provided from the input/output device 320 via the input/output interface 318, and receives information and/or data from the product attribute extraction server 230 via the communication module 316. can be received and the received information and/or data can be processed and stored in memory 312 . Such information and/or data can also be provided to the product attribute extraction server 230 via the communication module 316 .

According to one embodiment, processor 314 can receive input or selected text, images, actions, etc. through an input/output device 320 such as a touch screen or keyboard coupled to input/output interface 318 . The received text, images and/or actions may be stored in memory 312 or provided to item attribute extraction server 230 via communication module 316 and network 220 . For example, the processor 314 can receive one or more user queries and/or product attributes corresponding to the user queries via an input/output device such as a touch screen, keyboard, etc., and process the received user queries and/or product attributes. can be provided to the product attribute extraction server 230 via the communication module 316 and the network 220 .

According to one embodiment, the processor 314 can provide the user query received by the input device or a search result request for the user query to the product attribute extraction server 230 via the network 220 and the communication module 316 . Upon receiving the user query, the product attribute extraction server 230 can calculate the related scores of a plurality of product attributes from the user query. In addition, the product attribute extraction server 230 may provide information on product search results or product search rankings reflecting the calculated product attribute related score.

According to another embodiment, the processor 314 can provide the user query received by the input device or a search result request for the user query to the product attribute extraction server 230 via the network 220 and the communication module 316 . In addition, the processor 314 can receive product search results based on user queries from the product attribute extraction server 230 or product database via the network 220 and the communication module 316 . The processor 334 may calculate the relevance score of a plurality of product attributes to the user query according to the user query or a search result request for the user query, and apply the calculated relevance score to the product search result to adjust the product search ranking. .

According to another embodiment, the product attribute extraction server 230 can receive user queries and user-selected product attributes from the user terminal 210 via the network 220 and the communication module 316 . Alternatively, the product attribute extraction server 230 can receive data pairs of user queries and user-selected product attributes from the search log database (240 in FIG. 2) via the network 220 . The product attribute extraction server 230 generates learning data from the received data pairs of the user query and product attributes, and uses the generated learning data to calculate the user query and product attribute relevance score. can be learned.

FIG. 4 is a diagram illustrating a configuration for calculating product attribute related scores 434, 444, and 454 for a user query 410 using a statistically-based classification model 420 according to an embodiment of the present disclosure. The statistical-based classification model 420 can be generated, stored, or executed by the product attribute extraction server 230 or user terminal 210 shown in FIGS. As shown in FIG. 4 , the statistical-based classification model 420 can output association scores 434 , 444 , 454 for a plurality of product attributes 430 , 440 , 450 with a user query 410 as input. Here, the multiple product attributes 430 , 440 , 450 can be represented by multiple corresponding product attribute codes 432 , 442 , 452 . At this time, the product attribute codes 432, 442, 452 can be defined to correspond to ambiguous product attributes or range-type product attributes. For example, among the product attributes, "10 inches" (430) indicating the screen size of a notebook and "10 inches" indicating the screen size of a home TV can indicate mutually different product attributes. It can be represented by different product codes. Also, the association scores 434 , 444 , 454 may be probability values that the user query 410 is associated with each of the plurality of product attributes 430 , 440 , 450 . For example, if the user query 410 is "mini-notebook", the probability value associated with the product attribute of "10 inches" (430) with a small screen size is "0.7" and " The probability value associated with the product attribute of "12 inches" (440) is "0.17", and the probability value associated with the product attribute of "15 inches" (450) with a large screen size is "0.01". obtain.

Relevance scores 434 , 444 , 454 for each product attribute 430 , 440 , 450 can be calculated by the statistically based classification model 420 . The statistical-based classification model 420 can learn to calculate probability values associated with multiple product attributes from the input user query 410 . In this case, the product attribute extraction server or user terminal can receive learning data for learning the statistical-based classification model 420 from the search log database (eg, 240 in FIG. 2). For example, the learning data received from the search log database may include a pair of the user query 410 and product information (eg, attribute information of a specific product) selected by the user from the product search results corresponding to the user query 410. . Statistical-based classification model 420 may also be a machine learning or artificial neural network model for determining the relevance of multiple product attributes to user query 410 .

FIG. 4 shows that the association scores 434, 444, 454 of the three product attributes 430, 440, 450 for the user query 410 are calculated using one statistically-based classification model 420, but is limited to this. not something. According to various implementation methods, the statistical-based classification model 420 may be configured to calculate the relevance scores of less than three or more than four product attributes for the user query 410 . Also, depending on how the user query 410 is pre-processed or analyzed, more than one statistically-based classification model can be used. In one embodiment, the user query 410 is split into multiple smaller units of data (e.g., morpheme and/or syllable units of data), and the split data are input to corresponding statistically-based classification models. can be In this case, the final relevance score can be calculated by summing or combining the relevance scores calculated from the plurality of statistical-based classification models.

FIG. 5 is a diagram illustrating a configuration for calculating relevance scores 534, 544, 554 using a classifier 520 including a preprocessor 510 and statistically-based classification models 522, 524 according to an embodiment of the present disclosure. The preprocessor 510 and classifier 520 can be generated, stored, or executed by the product attribute extraction server 230 or the user terminal 210 shown in FIGS. As shown in FIG. 5, the product attribute extraction server or user terminal includes a preprocessor 510 that analyzes a user query 410 to generate input data for statistically based classification models 522 and 524, and and a classifier 520 that uses the input data to calculate relevance scores for multiple product attributes.

The preprocessor 510 may include a syllable unit segmentation unit 512 that segments the user query 410 into syllable units, and a morpheme unit segmentation unit 514 that divides the user query 410 into morpheme units. In one embodiment, the syllable unit divider 512 and the morpheme unit divider 514 can transform the text of the user query 410 into syllable embedding vectors and morpheme embedding vectors, respectively. Each of the syllable unit segmentation unit 512 and the morpheme unit segmentation unit 514 can be implemented by an artificial neural network trained to embed the user query 410 in a continuous vector space.

In addition, the classifier 520 includes a first statistically-based classification model 522 that receives the user query 410 divided into syllable units, and a second statistical-based classification model 522 that receives the user query 410 divided into morpheme units. A classification model 524 and a combiner 526 that sums or combines the relevance scores calculated from each of the statistically-based classification models 522, 524 to calculate final relevance scores 534, 544, 554 may be included.

In one embodiment, the product attribute extraction server or user terminal inputs the first statistically-based classification model 522 that receives the user query 410 divided into syllable units and the user query 410 divided into morpheme units. Relevant scores 534, 544, and 554 can be calculated using the second statistically-based classification model 524 as

Statistical-based classification models 522 and 524 included in the classifier 520 receive the user query 410 divided into units of syllables and morphemes, and obtain association scores 534, 544, and 554 for a plurality of product attributes 530, 540, and 550. can be calculated. Each of the statistical-based classification models 522 and 524 uses the learning data extracted from the search log database to calculate the association scores 534, 544 and 554 of the plurality of product attributes 530, 540 and 550 for the user query 410. can learn The learning data extracted from the search log database may include pairs of user queries and product information (for example, attribute information of specific products) selected by the user from product search results corresponding to the user queries.

The statistically-based classification models 522 and 524 do not use the text (eg, sentences, phrases, words, etc.) that make up the user query 410 as they are, but divide them into units of syllables or morphemes. , or the user query 410 does not appear in the training data and is unfamiliar. For example, if the user query 410 is 'mini-notebook', 'mi', 'di', 'no', 'to', and 'book' obtained by dividing 'mini-notebook' into syllable units are classified into statistical-based classifications. If the user query 410 contains misspellings such as "mini nato book", it is divided into syllable units "mi", "ni", "na", and "to". , and 'book' are the same or similar to the relevance scores calculated using the statistical-based classification model 522 as an input. As another example, when the user query 410 is 'mini-notebook', 'mini' and 'notebook' obtained by dividing 'mini-notebook' into units of morphemes are calculated as inputs of the statistical-based classification model 524. If the order of morphemes in the user query 410 is changed, such as 'notebook mini', the relation score and the 'notebook' and 'mini' divided into morpheme units are calculated as inputs to the statistical-based classification model 522. The associated score is the same or similar.

The relevance scores calculated by the statistically-based classification models 522 and 524 can be summed or combined by a combiner 526 . As a result, final relevance scores 534, 544, 554 can be calculated that are more accurate than using a single statistically-based classification model. With the above configuration, it is possible to maximize the matching coverage of learning data used by the statistical-based classification models 522 and 524, and improve the accuracy of calculating user queries and related scores of a plurality of product attributes.

FIG. 6 is a diagram illustrating an example of generating learning data of a statistical-based classification model for product attribute extraction from various search results according to an embodiment of the present disclosure. Training data used for training the statistical-based classification model can be extracted from the search log database 650 . The search log database 650 stores, for example, user queries input by the user while using the shopping search service (e.g. sentences, phrases, words, etc. containing the name of the product to be searched), shopping search results (e.g. , search results including product names, product attributes, etc.), and user responses to shopping search results (e.g., whether or not a user clicks on a specific product name, product attributes, etc.) can be saved. The learning data of the statistical-based classification model may include pairs of user queries extracted from the search log database 650 and product attributes associated with the user queries.

As shown in the figure, for example, when a user inputs a user query 110 into a search window of a shopping site or an integrated search site, search results 130 related to the input user query 110 can be output. The search result screen 100 includes a search window in which a user query 110 is input, a user interface 120 for filtering product attributes, search products 132 and 136, and product attributes 134 and 138 associated with the search products 132 and 136. etc., including search results 130 .

In one embodiment, when a user who has entered a user query 110 clicks a screen size attribute 610 included in the user interface 120, learning data 640 for the user query 110 and clicked screen size attribute 610 pair is retrieved. It can be stored in the log database 650 . The processor of the product attribute extraction server can extract learning data including pairs of user queries 110 and product attributes from the search log database 650 storing such user queries 110 and click log information. For example, when a user enters "mini-notebook" as the user query 110 and clicks "12 inches" among product attributes via the user interface 120 to generate search log data, "mini-notebook- 12 inches” of training data 640 can be generated.

As another example, when a user who has entered a user query 110 clicks on products 132 and 136 output as search results 130, product attributes 134 and 138 associated with the user query 110 and the products 132 and 136 are displayed. Paired learning data 620 , 630 may be stored in search log database 650 . For example, if a user clicks on items 132 and 136 in the search results 130 after entering "mini notebook" as the user query 110, the item attribute associated with the clicked item 132 is "10 inches" ( 134) and the user query 110 pair, and learning data 630 for the product attribute associated with the clicked item 136, "11 inches" (138) and the user query 110 pair. Additionally or alternatively, if the user who entered the user query 110 clicks on an item attribute 134, 138 associated with the item, learning data 620, 630 of the clicked item attribute 134, 138 and user query 110 pair is generated. can be

FIG. 7 is a diagram illustrating how product attributes 714, 724, 734 are represented by product attribute codes 716, 726, 736 to generate learning data 710, 720, 730 according to one embodiment of the present disclosure. As shown in the figure, learning data used for learning the statistical-based classification model includes user queries 712, 722, and 732, corresponding product attributes 714, 724, and 734, and each product attribute 714 , 724, 734, and product attribute codes 716, 726, 736. In one embodiment, the user queries 712, 722, 732 and the product attributes 714, 724, 734 are generated by entering the user queries 712, 722, 732 in the search window of the shopping site or integrated search site as described with reference to FIG. When the input user clicks on the product or product attributes displayed in the search results, it can be extracted from the generated search log database. Also, each of the product attributes 714, 724, 734 can be represented by corresponding product attribute codes 716, 726, 736. FIG. In this way, by expressing the product attributes contained in the learning data with product attribute codes, when different types of products have the same or similar product attributes (i.e. ambiguous product attributes and range-type product attributes) , the statistically-based classification model can also calculate an accurate relevance score.

For example, when a user enters "mini-notebook" (712) in the search window of a shopping site and clicks the product attribute of "10 inches" (714), the entered "mini-notebook" (712) user Learning data 710 can be generated that includes the query, the item attribute of "10" (714) that was clicked, and the item attribute code "M10834726" (716) of "10" (714). Also, when the user enters 'mini-notebook' (722) in the search window of the shopping site and clicks the product attribute of '12 inches' (724), the user of the entered 'mini-notebook' (722) Learning data 720 can be generated that includes the query, the product attribute of "12" (724) that was clicked, and the product attribute code of "12" (724) "M10706689" (726).

In another embodiment, when a user who inputs a user query clicks two or more product attributes in a search result, learning data including both the input user query and the clicked product attributes can be generated. For example, when the user inputs 'mini-notebook' (732) in the search window of the shopping site and clicks on each of the product attributes of '13 inches and 12 inches' (734), the input 'mini-notebook' ( 732) and clicked product attributes of "13 inches and 12 inches" (734) and corresponding product attribute codes "M10835015 and M10706689" (736) can be generated.

FIG. 8 is a flowchart illustrating a product attribute extraction method 800 for shopping searches performed by a product attribute extraction server according to one embodiment of the present disclosure. In one embodiment, the product attribute extraction method 800 for shopping search can be initiated by receiving a user query (S810). In one embodiment, a user query can be received from a search window of a shopping site or an integrated search site displayed on a user terminal and forwarded to a processor of a product attribute extraction server.

Based on the received user query, a product search result corresponding to the user query is extracted from the product database (S820). In one embodiment, the processor of the product attribute extraction server can extract product search results for user queries from a product database. At this time, the product search result may include product search ranking information determined based on user queries, lexical similarity of product names stored in the product database, and user popularity of products. For example, if the user query is "mini-notebook", among the products whose product name includes at least a part of "mini-notebook", products with a high purchase rate by users or products with a high click-through rate by users can be extracted as product search results with high rank.

The user query can be divided into units of syllables and/or morphemes (S830). In one embodiment, the processor of the product attribute extraction server can divide the user query into units of syllables and/or morphemes and input them into the statistically-based classification model. For example, a user query can be converted into a syllable embedding vector and/or a morpheme embedding vector and input to a statistically based classification model.

In addition, the statistical-based classification model can be used to calculate the associated score of multiple product attributes with respect to the user query (S840). In one embodiment, the processor of the product attribute extraction server can use a statistically-based classification model to calculate the probability that the user query is associated with the product attributes indicated by the product attribute code. In this way, by displaying product attributes with product attribute codes, the statistics-based classification model can calculate accurate relevance scores corresponding to ambiguous product attributes and range-type product attributes. When a user query is divided into units of syllables or morphemes, the statistical-based classification model can calculate a probability value that the user query divided into units of syllables or morphemes is associated with a plurality of product attributes.

In another embodiment, the processor of the product attribute extraction server can divide the user query into units of syllables and morphemes, respectively. The processor uses a first statistically based classification model to calculate a first probability that the user query divided into syllable units is associated with a plurality of product attributes, and uses a second statistically based classification model to: A second probability that the morpheme-divided user query is associated with the product attribute can be calculated. The processor can then sum or combine the first and second probabilities to calculate a final item attribute relevance score. As described above, by dividing the user query into syllable and/or morpheme units and calculating the relevance score for the product attributes by the statistically-based classification model, it is possible to identify the product even if the user query is mistyped or not included in the training data. It is possible to improve the accuracy of calculation of attribute relevance scores.

Based on the calculated relevance score, the product search ranking according to the product search result for the user query can be adjusted (S850). In one embodiment, the processor of the product attribute extraction server may adjust the product search ranking by reflecting the calculated one or more related scores in the product search result. As a result, products including product attributes with high related scores can be displayed at the top of the search results. By displaying products having product attributes related to the user query at the top of the search results, it is possible to improve the click rate and purchase rate of users for the products.

FIG. 9 is a flowchart illustrating a product search ranking adjustment method 900 for shopping searches according to one embodiment of the present disclosure. In one embodiment, the product search ranking adjustment method 900 may begin by calculating the relevance score of a plurality of product attributes to a user query (S910). A method of calculating the relevance score of a plurality of product attributes with respect to a user query may be performed in the same manner as the method described with reference to FIG.

Next, a first product attribute associated with a first related score, which is the highest related score, among the calculated related scores can be extracted (S920). In one embodiment, the product attribute extraction server or the processor of the user terminal can extract the first product attribute associated with the first related score, which is the highest related score among the calculated related scores. Also, based on the extracted first product attribute, the first product including the first product attribute can be searched from the product database (S930).

Alternatively, among the calculated relevance scores, a first product attribute group (i.e., , one or more first product attributes) can be extracted. After that, a product including at least one product attribute included in the first product attribute group can be retrieved from the product database.

A product attribute weight that determines the importance of the first product attribute can be determined (S940). The product attribute weight value may indicate the degree of influence of the extracted product attribute on adjustment of the search ranking of related products. In one embodiment, the product attribute weight value may be a default value for each product attribute. Alternatively, the item attribute weights can vary dynamically depending on the user's response to the item search results. Specifically, the product attribute extraction server or the processor of the user terminal outputs a search result screen displaying the first product adjusted based on the predetermined product attribute weighting value and the related score, and outputs the product search The number of user clicks on the first product can be received on the results screen. If the number of clicks for the first item is greater than or equal to the predetermined threshold, the processor can change the item attribute weights.

The search ranking of the first product may be adjusted based on the first relevance score and the product attribute weight (S950). As a result of adjusting the product search ranking, the first product related to the extracted first product attribute is higher in the product search results than other products not related to the first extracted product attribute. can be displayed in Thereby, based on the user query, it is possible to output to the user terminal a search result reflecting the product attributes intended by the user.

FIG. 10 is a flowchart illustrating a product attribute extraction method 1000 for shopping search performed by a user terminal according to one embodiment of the present disclosure. In one embodiment, the product attribute extraction method 1000 for shopping search performed by a user terminal can be initiated by a processor of the user terminal receiving a user query (S1010). A user query can be received from a search window of a shopping site or an integrated search site displayed on the user terminal.

The processor of the user terminal can receive the relevance scores of a plurality of product attributes calculated by the statistically based classification model based on the user query (S1020). At this time, the statistical-based classification model can be learned in advance to calculate the relevance score of a plurality of product attributes with respect to the user query. In addition, in order to calculate the relevance score with high accuracy, the statistical-based classification model receives a user query divided into units of syllables and/or morphemes and uses two or more statistics for calculating the relevance score of product attributes. A base classification model can be included.

The processor of the user terminal may adjust the product search order for the user query based on the calculated relevance score (S1030). Specifically, the processor can extract a first product attribute associated with a first association score, which is the highest association score, among the calculated association scores. The processor can then receive the first item including the first item attribute retrieved from the item database, and adjust the item search ranking of the retrieved first item with the first relevance score. At this time, the search ranking of the first product may be adjusted based on the product attribute weighting value that determines the importance of the first product attribute and the first relevance score.

FIG. 11 is a flowchart illustrating a method 1100 of training a statistically-based classification model for product attribute extraction according to one embodiment of the present disclosure. As shown, the learning method 1100 includes a processor generating first learning data based on a plurality of first user queries and a plurality of first product attributes associated with the plurality of first user queries. It can start with step (S1110). The processor can extract a plurality of first product attributes from a search log database associated with the plurality of first user queries. The search log database may include product attribute information selected or clicked by the user from user queries and product search results extracted thereby. For example, the processor can extract product attributes associated with a plurality of user queries from at least one of a federated search log database, a shopping or product search log database. The processor can also transform the plurality of first product attributes into corresponding plurality of product attribute codes so that the statistically-based classification model can accommodate ambiguous product attributes or range-type product attributes. With the above configuration, the learning data is generated by the user's input by the user and the user's action on the product search results extracted by the user query (e.g., product attribute filtering, product or product attribute selection or click, etc.). Since it is obtained from the search log database, it is possible to efficiently secure training data for the statistically based classification model.

The processor can learn a statistically-based classification model to calculate the relevance score of one or more product attributes corresponding to the second user query based on the plurality of first learning data (S1120). In one embodiment, the processor can divide the plurality of first user queries into units of morphemes to learn a statistically based classification model. Specifically, the processor can learn a statistically-based classification model to calculate relevance scores of one or more product attributes corresponding to user queries based on a plurality of first user queries divided into units of morphemes. . Additionally or alternatively, the processor can divide the plurality of first user queries into units of syllables to learn a statistically based classification model. The processor can learn a statistically-based classification model to calculate relevance scores of one or more product attributes corresponding to user queries based on a plurality of first user queries divided into syllable units.

In another embodiment, the processor can divide the plurality of first user queries into units of syllables and morphemes to learn a statistically based classification model. Specifically, the processor can divide the plurality of first user queries into units of morphemes and syllables. Then, the processor learns the first statistical-based classification model to calculate the relevance score of one or more product attributes corresponding to the user query based on the plurality of first user queries divided into units of morphemes. can. In addition, the processor learns the second statistical-based classification model to calculate the relevance score of one or more product attributes corresponding to the user query based on the plurality of first user queries divided into units of syllables. can. As described above, the processor can more accurately calculate the relevance score of the user query and product attributes by using two or more statistically-based classification models according to the method of dividing the user query.

The product attribute extraction method and product attribute extraction learning method for shopping search described above can also be embodied as computer-readable codes on a computer-readable recording medium. The recording medium may be of continuous storage or temporary storage for execution or download of the computer-executable program. In addition, the medium can be a variety of recording means or storage means in the form of a single or multiple pieces of hardware combined, but is not limited to a medium directly connected to a computer system, and is distributed over a network. can be Examples of media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks. medium), ROM, RAM, flash memory, etc., which are configured to store program instructions. Examples of other media include recording media and storage media managed by application stores that distribute applications, sites that supply and distribute various software, and servers.

The methods, acts, or techniques of the disclosure may be embodied by various means. For example, such techniques may be implemented in hardware, firmware, software, or any combination thereof. A person of ordinary skill in the art should understand that the various exemplary logical blocks, modules, circuits and algorithm steps described in the present disclosure can be embodied in electronic hardware, computer software, or a combination of both. is. To clearly illustrate such mutual alternatives of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality. Whether such functionality is implemented as hardware or software will vary depending on the particular application and design requirements imposed on the overall system. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementations should not be interpreted as departing from the scope of the present disclosure.

In a hardware implementation, the processing unit utilized to perform the techniques may be one or more of ASICs, DSPs, digital signal processing devices (DSPDs), programmable logic devices (PLDs). , field programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, electronic devices, and other electronic units designed to perform the functions described in this disclosure , a computer, or a combination thereof.

Thus, the various exemplary logic blocks, modules and circuits described by this disclosure may be general purpose processors, DSPs, ASICs, FPGAs or other programmable logic devices, discrete gate or transistor logic, discrete hardware components, or , may be embodied or performed in any combination, including those designed to perform the functions described herein. A general-purpose processor can be a microprocessor, but, in the alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. A processor may also be embodied in a combination of computing devices, such as a DSP and a microprocessor, multiple microprocessors, one or more microprocessors associated with a DSP core, or any other combination of configurations. .

In the implementation of firmware and/or software, the technique is random access memory (RAM), read-only memory (ROM), non-volatile random access memory (NVRAM), programmable read-only memory (PROM), erasable programmable memory (EPROM). It can be embodied by instructions stored on a computer-readable medium such as a read-only memory), electrically erasable PROM (EEPROM), flash memory, compact disc (CD), magnetic or optical data storage device, and the like. The instructions may be executable by one or more processors and may cause the processors to perform certain aspects of the functions described in this disclosure.

When implemented in software, the techniques can be stored on or transferred over a computer-readable medium as one or more instructions or code. Computer-readable media includes both computer storage media and communication media that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a computer. By way of non-limiting example, such computer readable media may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any desired program code. It can also include any other computer-accessible medium that can be used to transport or store instructions or data structures in a form. Also, any connection may be included as a computer-readable medium.

For example, software may be transferred from a website, server or other remote source using coaxial cable, fiber optic cable, lead wire, digital subscriber line (DSL), or wireless technologies such as infrared, radio and microwave. Coaxial cable, fiber optic cable, lead wire, digital subscriber line, or wireless technologies such as infrared, radio and microwave, where applicable, are included within the definition of medium. As used herein, disk and disc include CDs, laser discs, optical discs, DVDs (digital versatile discs), floppy discs and Blu-ray discs, where disk generally refers to a magnetic A disc reproduces data optically using a laser, whereas a disc reproduces data optically. Combinations and the like should also be included within the scope of computer-readable media and the like.

A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium may be coupled to the processor, such that the processor reads information from, and writes information to, the storage medium. Alternatively, the storage medium may be integrated into the processor. The processor and storage medium can also reside within an ASIC. The ASIC may also reside in a user terminal. Alternatively, the processor and storage medium may exist as separate components in the user terminal.

Although the foregoing embodiments are described as utilizing aspects of the presently disclosed subject matter in one or more stand-alone computer systems, the disclosure is not so limited, and may be used in networked or distributed computing environments, such as. It can be embodied by any computing environment. Further, aspects of the subject matter of this disclosure may be embodied in multiple processing chips or devices, and storage may be similarly affected across multiple devices. Such devices may also include PCs, network servers and handheld devices.

In the present specification, the present disclosure has been described with some embodiments, but various modifications and changes can be made without departing from the present disclosure that can be understood by a person of ordinary skill in the technical field to which the disclosed invention belongs. It is possible. Also, such variations and modifications should be understood to fall within the scope of the claims appended hereto.

100 Screen 110 User Query 120 User Interface 130 Search Result 132, 136 Product 134, 138 Product Attribute 210 User Terminal 220 Network 230 Product Attribute Extraction Server 240 Search Log Database 420 Statistical Base Classification Model 510 Preprocessor 520 Classifier

Claims (15)

In a product attribute extraction method for shopping search performed by a computer system,
receiving a user query;
extracting product search results for the user query from a product database;
calculating a relevance score of the user query and a plurality of product attributes according to a statistically-based classification model; and
adjusting the product search ranking according to the product search results based on the calculated relevance score ;
The step of calculating relevance scores of the user query and the plurality of product attributes by the statistically-based classification model,
dividing the user query into units of syllables and morphemes;
calculating a first probability that the syllable-divided user query is associated with the plurality of product attributes according to a first statistically-based classification model;
calculating a second probability that the user query divided into morphemes is associated with the plurality of product attributes by a second statistically-based classification model; and
Based on the first probability and the second probability, calculating as the related score of the plurality of product attributes,
Product attribute extraction method for shopping search. The step of calculating relevance scores of the user query and the plurality of product attributes by the statistically-based classification model,
2. The product attribute extraction method for shopping search according to claim 1, comprising calculating a probability that the user query is associated with the plurality of product attributes each represented by a product attribute code, according to the statistically-based classification model. . The step of calculating relevance scores of the user query and the plurality of product attributes by the statistically-based classification model,
dividing the user query into at least one of syllable and morpheme units; and
3. The method according to claim 1, further comprising calculating a probability that the user query divided into at least one of the syllable and the morpheme unit is associated with the plurality of product attributes by the statistically-based classification model. Product attribute extraction method for shopping search. The step of adjusting the product search ranking according to the product search results based on the calculated relevance score,
a step of extracting a first product attribute associated with a first association score, which is the highest association score among the calculated association scores;
searching the product database for a first product that includes the extracted first product attribute; and
4. The product attribute extraction method for shopping search according to any one of claims 1 to 3 , comprising adjusting the product search order of the searched first product according to the first relevance score. The step of adjusting the product search ranking according to the product search results based on the calculated relevance score,
a step of extracting a first product attribute group related to a first group of related scores having a related score greater than or equal to a predetermined threshold among the calculated related scores;
searching from the product database for products including at least one product attribute included in the extracted first product attribute group; and
4. The product attribute extraction method for shopping search according to any one of claims 1 to 3 , comprising a step of adjusting the product search order of the searched product according to the first related score group. The step of adjusting the product search ranking according to the product search results based on the calculated relevance score,
5. The shopping search of claim 4 , further comprising adjusting a search ranking of the first product based on a product attribute weighting value for determining importance of the first product attribute and the first relevance score. Product attribute extraction method for a step of outputting a search result screen on which the first product is displayed by adjusting the search ranking;
receiving the number of clicks of the first product on the output search results screen; and
7. The product attribute extraction method for shopping search as claimed in claim 6 , further comprising changing the product attribute weighting value when the number of clicks of the first product is greater than or equal to a predetermined threshold. A computer program for executing the product attribute extraction method for shopping search according to any one of claims 1 to 7 on a computer. In a learning method of product attribute extraction for shopping search performed by a computer system,
generating a plurality of first learning data including a plurality of first user queries and a plurality of first product attributes associated with the plurality of first user queries; and
learning a statistically-based classification model to calculate relevance scores for one or more product attributes corresponding to a second user query based on the plurality of first learning data ;
The step of learning the statistically-based classification model includes:
dividing the plurality of first user queries into units of morphemes;
dividing the plurality of first user queries into units of syllables;
learning a first statistical-based classification model to calculate relevance scores of one or more product attributes corresponding to user queries based on the plurality of first user queries divided into morpheme units; as well as,
learning a second statistically-based classification model to calculate relevance scores of one or more product attributes corresponding to the user queries based on the plurality of first user queries divided into syllable units; include,
How to learn product attribute extraction for shopping search. The step of generating the plurality of first learning data includes:
10. The shopping search of claim 9 , comprising extracting the plurality of first product attributes from at least one of federated search click information, shopping click information, or a product database associated with the plurality of first user queries. How to learn product attribute extraction for The step of generating the plurality of first learning data includes:
11. The learning method for product attribute extraction for shopping search according to claim 9 or 10 , comprising the step of converting said plurality of first product attributes into a plurality of first product attribute codes that can correspond to ambiguous product attributes. . The step of learning the statistically-based classification model includes:
dividing the plurality of first user queries into units of morphemes; and
Based on the plurality of first user queries divided into units of the morpheme, learning the statistical-based classification model to calculate the relevance score of one or more product attributes corresponding to the user query; The learning method for product attribute extraction for shopping search according to any one of claims 9 to 11 . The step of learning the statistically-based classification model includes:
dividing the plurality of first user queries into units of syllables; and
Based on the plurality of first user queries divided into syllable units, learning the statistically-based classification model to calculate the relevance score of one or more product attributes corresponding to the user query; The learning method for product attribute extraction for shopping search according to any one of claims 9 to 11 . A computer program for executing the learning method for product attribute extraction for shopping search according to any one of claims 9 to 13 on a computer. In the product attribute extraction system for shopping search,
a communication module for receiving user queries;
memory; and
at least one processor coupled to the memory and configured to execute computer readable instructions contained in the memory;
The at least one processor
Extracting product search results for the user query from a product database, calculating the relation scores of the user query and a plurality of product attributes using a statistically-based classification model, and calculating the product search results based on the calculated relation scores Adjust the product search ranking by
According to the statistical-based classification model, calculating the user query and the related scores of the plurality of product attributes includes:
dividing the user query into units of syllables and morphemes;
calculating a first probability that the user query divided into syllables is associated with the plurality of product attributes by a first statistically based classification model;
calculating a second probability that the user query divided into morpheme units is associated with the plurality of product attributes by a second statistically-based classification model;
Based on the first probability and the second probability, by calculating as the related score of the plurality of product attributes,
Product attribute extraction system for shopping search.

Images