E-commerce platform for providing matching-based services between supplier member and purchaser member

Abstract

Disclosed is an e-commerce platform that provides matching-based services between supplier members and purchaser members. The e-commerce platform, which mediates transactions between a supplier member and a purchaser member, comprises a processing server that performs: a matching task for calculating a matching score between a supplier member and a purchaser member on the basis of a similarity analysis between registered products registered by the supplier member and a product to be purchased by the purchaser member, and for querying the supplier member as to whether to accept establishment of a one-to-one relationship with the purchaser member on the basis of the calculated matching score; and a feedback-providing task of providing feedback reflecting purchase trends for registered products and detailed information thereof registered by the supplier member.

Description

A commerce platform that provides matching-based services between supplier members and buyer members

[0001] The present invention relates to a commerce platform that provides a matching-based service between a purchasing member and a buying member.

[0002] With the recent advancement of artificial intelligence (AI) technology, various tasks are being performed, including the generation of text or images, or the generation of different multimodals such as text and images.

[0003] Due to business trends such as the specialization, segmentation, and globalization of business areas, collaboration between different business partners is emerging as a key factor for business success. Consequently, there is a growing need for research on matching services that utilize artificial intelligence technology to facilitate the matching of different business partners.

[0004] An embodiment of the present invention includes a commerce platform that provides a matching-based service between a supplier member and a buyer member, wherein a matching service for establishing a one-to-one relationship between a supplier member and a buyer member is implemented based on a similarity analysis between a product registered by a supplier member and a product purchased by a buyer member, and a feedback provision service for providing feedback reflecting purchasing trends regarding a product registered by a supplier member and detailed information of the product.

[0005] In order to achieve the above objectives and other objectives, the commerce platform of the present invention, which provides a matching-based service between a purchasing member and a buying member, is

[0006] As a commerce platform that mediates commercial transactions between supplier members and buyer members,

[0007] A matching task for calculating a matching score between a supplier member and a buyer member from a similarity analysis between a product registered by a supplier member and a product purchased by a buyer member, and querying the supplier member regarding acceptance of establishing a one-to-one relationship between the supplier member and the buyer member based on the calculated matching score; and

[0008] It may include a processing server that performs a feedback provision task to provide feedback reflecting purchasing trends regarding registered products and detailed information of registered products registered by purchasing member.

[0009] According to the present invention, a commerce platform is provided that offers a matching-based service between a supplier member and a buyer member, wherein a matching service for establishing a one-to-one relationship between a supplier member and a buyer member is implemented based on a similarity analysis between a product registered by a supplier member and a product purchased by a buyer member, and a feedback provision service for providing feedback reflecting purchasing trends regarding a product registered by a supplier member and detailed information of the product registered by the supplier member.

[0010] FIG. 1 illustrates a diagram for explaining a commerce platform according to an embodiment of the present invention, and for explaining a matching task and a feedback provision task processed by a processing server.

[0011] FIGS. 2a and 2b are drawings for explaining a RAG implemented in one embodiment of the present invention, and FIGS. 2a and 2b respectively show different drawings for explaining processes performed in the preprocessing and service stages.

[0012] Figure 3 is a diagram illustrating that the similarity between embedding representations mapped to the same embedding space from an embedding model is high depending on the position, using attribute data combining various different attributes of a product as input. It is a diagram illustrating that the similarity between the input texts is mapped to embedding spaces at adjacent positions depending on the similarity of the input texts.

[0013] FIG. 4 shows a diagram schematically illustrating k-means clustering as an example of a clustering algorithm for calculating a matching score between a group of products associated with a purchasing member and another group of products associated with a purchasing member in one embodiment of the present invention, performing a similarity analysis between a group of products associated with a purchasing member and another group of products associated with a purchasing member according to the distance or angle between the centers of each cluster, and calculating a matching score between the purchasing member and the purchasing member associated with these products.

[0014] FIG. 5 illustrates a form of a screen that exemplarily shows the matching results between different purchasing members and buying members based on the matching score calculated between different purchasing members and buying members in one embodiment of the present invention.

[0015] FIG. 6 illustrates a form of a screen that queries a purchasing member regarding whether to establish a one-to-one relationship based on a matching score as an upstream task in one embodiment of the present invention.

[0016] FIGS. 7 to 9 illustrate screens that exemplarily show various configurations of a feedback screen provided to a purchasing member as a result of a task to provide feedback regarding purchasing trends in one embodiment of the present invention, and more specifically, drawings showing the configuration of a screen rendered in the shape of a straight line graph or rendered in the shape of a polyhedron for each of the different attributes of a purchasing member's registered product.

[0017] In order to achieve the above objectives and other objectives, the commerce platform of the present invention, which provides a matching-based service between a purchasing member and a buying member, is

[0018] As a commerce platform that mediates commercial transactions between supplier members and buyer members,

[0019] A matching task for calculating a matching score between a supplier member and a buyer member from a similarity analysis between a product registered by a supplier member and a product purchased by a buyer member, and querying the supplier member regarding acceptance of establishing a one-to-one relationship between the supplier member and the buyer member based on the calculated matching score; and

[0020] It may include a processing server that performs a feedback provision task to provide feedback reflecting purchasing trends regarding registered products and detailed information of registered products registered by purchasing member.

[0021] For example, the above processing server,

[0022] In the above matching task, for each of the products registered by the supplier member and the products purchased by the buyer member that are the target for calculating the matching score, a similarity analysis is performed on the entire set of attribute data combining various different attributes of the products, and

[0023] In the above feedback provision task, for each of the trend products selected based on purchase preference or purchase ranking and the registered products of the purchasing member subject to the feedback, a similarity analysis regarding the individual attributes of the products can be performed.

[0024] For example, the above processing server,

[0025] In the above matching task, for each of the products registered by the supplier member and the products purchased by the buyer member that are the target for calculating the matching score, an embedding representation of the entire attribute data combining various different attributes of the products is calculated, and similarity analysis is performed between the calculated embedding representations based on distance or angle.

[0026] In the above feedback provision task, for each of the trend products selected based on purchase preference or purchase ranking and the registered products of the supplier member subject to the feedback, deviations for each attribute forming attribute data combining various different attributes are calculated, and feedback on purchase trends can be provided to the supplier member regarding the calculated deviations.

[0027] For example, the above processing server,

[0028] Based on the matching score calculated between the above-mentioned supplier member and buyer member, and the supplier member's response regarding the acceptance or rejection of the request to establish a one-to-one relationship with the supplier member, a one-to-one relationship is established between the above-mentioned supplier member and buyer member, and

[0029] For products set to private at the request of the aforementioned supplier member, images of the products shall not be disclosed to general buyer members, but may be disclosed to buyer members with whom a one-to-one relationship has been established with the relevant supplier member.

[0030] For example, the above processing server,

[0031] In the above feedback provision task, for each of the purchased member's registered products and trend products, the deviation of the attribute can be numerically calculated from the attribute quantified by the number of word expressions forming the attribute of the keyword or search term among each attribute forming attribute data in which different attributes are combined.

[0032] For example, the above processing server,

[0033] In the above matching task, for each of the products registered by the supplier member and the products purchased by the buyer member, among the respective attributes forming attribute data formed by combining different attributes, a dimensionally reduced latency expression or a context vector regarding the entire context of multiple word expressions forming the attributes of a keyword or search term is calculated, attribute data is generated that includes the calculated latency expression or context vector as a specified column element of the corresponding keyword or search term, and a similarity analysis can be performed on the attribute data generated for the products registered by the supplier member and the attribute data generated for the products purchased by the buyer member.

[0034] An e-commerce platform providing a matching service between a purchasing member and a buyer member according to one embodiment of the present invention will be described.

[0035] In one embodiment of the present invention, an e-commerce platform for mediating e-commerce can mediate a transaction of goods between a purchasing source classified as a wholesaler and a buyer classified as a retailer. For example, in one embodiment of the present invention, the buyer may be a business member corresponding to a retailer rather than an individual member.

[0036] In one embodiment of the present invention, the processing server, as the entity operating or managing the e-commerce platform, may provide a matching service as business partners between a supplier member and a buyer member. For example, the matching service between the supplier member and the buyer member may be implemented based on a similarity analysis between a registered product that the supplier member has registered on the processing server as a product they plan, and a purchased product that the buyer member has purchased from the processing server's commerce intermediary as a product they distribute.

[0037] In one embodiment of the present invention, a matching service (matching task) between a supplier member and a buyer member analyzes multiple product attributes as product factors capable of determining the tendency of the products registered by the supplier member and the products purchased by the buyer member, or the tendency of the products, and can calculate a matching score between different supplier members and buyer members, and can provide a recommendation list reflecting priority according to the matching score. For example, in one embodiment of the present invention, the tendency of the products is calculated by calculating attribute values ​​for various product attributes such as product category, product color tone, product material, and product size (such as one-dimensional length or three-dimensional volume), and calculates a matching score by combining the attribute values. If the calculated matching score is greater than or equal to a pre-set threshold matching score, the supplier member and the buyer member corresponding to the target of the matching score calculation can be determined as a matching pair that matches each other.

[0038] In various embodiments of the present invention, the matching service between a supplier member and a buyer member may be expressed in various ways. For example, it may be provided in the form of a service provided to a member by calculating a matching score between the applicant member who requested the matching service and other members, and providing the applicant member with a list reflecting priorities based on the calculated matching score. However, in one embodiment of the present invention, there is a high probability that the supplier member performing product planning or the scale of the wholesale business operated by the supplier member (e.g., including sales volume, size of internal organization such as employees, internal organizational capabilities, etc.) is larger than the scale of the retail business operated by the buyer member or the buyer member. Even if a matching partner is recommended to one party among different types of supplier members and buyer members where such asymmetry in scale exists, there is a high probability that the other party recommended as the matching partner will not initiate a business relationship with the first party as a business partner (e.g., due to a significant imbalance between the sales volume of one party and the sales volume of the other party), and accordingly, providing a matching service that includes the recommendation of a matching partner to one party may be less effective.

[0039] Based on the considerations above, in one embodiment of the present invention, rather than providing a matching service between a supplier member and a buyer member as one of the various applications provided to members on a commerce platform, priority may be granted to a supplier member whose business scale is predicted to be relatively large—for example, in terms of the size of the organization for product planning or the scale of sales—as a matching service or a downstream service that can be derived from the matching service, in accordance with the policy of the commerce platform regarding a one-to-one relationship established according to the policy of the commerce platform. For example, a supplier member and a buyer member may be selected as a matching pair from the matching service between a supplier member and a buyer member as described above, and a request may be transmitted to the supplier member among the selected matching pairs to inquire whether to accept the establishment of a one-to-one relationship with the recommended matching partner (matched buyer member). Depending on the acceptance or rejection response transmitted from the corresponding supplier member as a response to the request inquiring whether to accept transmitted from the processing server, a one-to-one relationship between the supplier member and the buyer member matched based on the matching service between the supplier member and the buyer member as described above It may be established (acceptance response) or a one-to-one relationship may not be established between them (rejection response).

[0040] As such, in one embodiment of the present invention, priority regarding the establishment of a one-to-one relationship established according to the policy of a commerce platform may be processed in a manner where priority is granted to the supplier member, rather than being created based on mutual consent between the supplier member and the buyer member to whom the one-to-one relationship is matched. In one embodiment of the present invention, unlike a platform providing commerce mediation or social network services, the establishment of a one-to-one relationship may be processed based on the acceptance of one party (the supplier member) rather than based on mutual consent between the parties to whom the one-to-one relationship is established. The establishment of a one-to-one relationship is not processed based on the request of one party (the supplier member); that is, it does not depend on the request of one party (the supplier member) and the other party (the buyer member) or the acceptance of the other party (the buyer member). For example, in one embodiment of the present invention, the establishment of a one-to-one relationship may be processed based on the acceptance of one party (the supplier member). For example, in one embodiment of the present invention, a task for establishing a one-to-one relationship between a supplier member and a buyer member can be understood as an upstream task and a downstream task for a matching service between the supplier member and the buyer member, and the matching service between the supplier member and the buyer member corresponding to the upstream task is not initiated in response to a service request from the supplier member and the buyer member corresponding to the matching counterpart, and the processing server can extract attribute data of the registered product and the purchased product of the corresponding member from the product data registered on the processing server regarding the registered product from the supplier member and the purchased product from the buyer member, for example, in one embodiment of the present invention, the attribute data of the registered product and the purchased product can be expressed in the form of a one-dimensional vector arranged in one dimension with attributes regarding each product as factors or column elements, and an embedding representation can be generated from an embedding model that takes attribute data regarding each product as input.For example, in one embodiment of the present invention, attribute data regarding a group of registered products registered on a processing server and a group of purchased products registered on the processing server or a database connected to the processing server may be extracted, and an embedding representation regarding each product may be generated from an embedding model that takes the extracted attribute data as input, and the embedding representation regarding each product may be stored in a vector DB (vector database) as a vectorized representation generated from the embedding model. In one embodiment of the present invention, a database connected to the processing server may provide keyword search centered on keywords or key values, and in one embodiment of the present invention, the vector DB (vector database) may provide semantic search by inputting product attribute data into an embedding model for generating an embedding expression mapped to the same embedding space from an input expression, and may store the embedding expression generated from the embedding model, and may perform distance-based similarity analysis (Euclidean distance) between embedding expressions mapped to the same embedding space from embedding expressions regarding different products, or perform angle-based similarity analysis (cosine similarity) between embedding expressions mapped to the same embedding space.For example, in one embodiment of the present invention, attribute data expressed in the form of a one-dimensional vector having attributes of each product, including product category, product name (or keyword), color tone, price, material, and size (length), as attributes relating to different products, as factors or column elements, may comprehensively refer to embedding representations generated from an embedding model, for example, may comprehensively refer to embedding representations generated as low-dimensional dense representations to enable analysis of similarity between different representations from an embedding model, and may be used in a comprehensive sense including sparse representations, such as one-hot vectors, as a form prior to being input into an embedding model. For example, in one embodiment of the present invention, various data forms for expressing product attributes or sets of different product attributes may be comprehensively expressed as attribute data.

[0041] In a platform according to one embodiment of the present invention, a retrieval augmented generation (RAG) may be implemented in which a vector DB is constructed to store embedding expressions generated from an embedding model as product attribute data. For example, attribute data of purchased products, such as registered products and transaction history registered on a processing server, may be input into an embedding model to generate embedding expressions for each product, and preprocessing may be performed to store the generated embedding expressions in the vector DB. In a service stage, for example, in a service stage for calculating a matching score between a supplier member and a buyer member, attribute data (embedding expressions) of a group of products associated with a supplier member that is the subject of the matching score calculation may be extracted from the vector DB, and attribute data (embedding expressions) of a group of products associated with a buyer member may be extracted. Similarity between different groups of products and other groups of products may be calculated based on the mapped positions of the attribute data of a group of products associated with each supplier member and the attribute data of other groups of products associated with the buyer member within the same embedding space, and the supplier corresponding to the subject of the matching score calculation A matching score can be calculated between a member and a buyer member. For example, in one embodiment of the present invention, a matching score can be calculated between different supplier members and buyer members by calculating a matching score between a group of products linked to a supplier member (e.g., a group of products linked to the same supplier member that is subject to the calculation of the matching score, a group of registered products registered by the corresponding supplier member) and a group of other products linked to a buyer member (e.g., products in a different group linked to the same buyer member that is subject to the calculation of the matching score, a group of purchased products in a different group retrieved from the transaction history of the corresponding buyer member).

[0042] In one embodiment of the present invention, a retrieval augmented generation (RAG) may be implemented in the processing server or a storage device connected to the processing server to store embedding representations for each product so as to map attribute data of different products onto the same embedding space, thereby enabling rapid calculation of a matching score in the service stage. In one embodiment of the present invention, in the service stage for calculating the matching score (e.g., an upstream task for sending a request to a purchasing member regarding acceptance of the establishment of a one-to-one relationship), the products (identification data regarding products) linked to the purchasing member and the buyer member, respectively, which are the targets for calculating the matching score, may be used as inputs to extract (retrieve) embedding representations (attribute data) regarding each product, and embedding representations regarding each product may be extracted from a vector DB, and a similarity or matching score may be calculated between a group of products linked to the purchasing member extracted from the vector DB and another group of products linked to the buyer member. For example, in a RAG implemented in one embodiment of the present invention, preprocessing including loading content, splitting (or chunking) of content, embedding of content, and storage in a vector DB, and real-time processing such as embedding of a query (products of purchasing member and buyer member, identification data regarding products) and searching of the vector DB through a retriever can be performed.

[0043] In one embodiment of the present invention, a group of products registered by a purchasing member corresponding to the calculation target of the matching score can be clustered into a first cluster by mapping within a first category in the embedding space, and another group of products retrieved from the transaction history of a buyer member corresponding to the calculation target of the matching score can be clustered into a second cluster by mapping within a second category in the embedding space.For example, in one embodiment of the present invention, a supplier member associated with a group of products clustered into a first cluster can be predicted to conduct wholesale business, such as product planning, production, or import, within the generally identical or similar group of products in the first cluster, and in one embodiment of the present invention, a buyer member associated with another group of products clustered into a second cluster can be predicted to conduct retail business, such as product distribution, within the generally identical or similar group of products in the second cluster, and the centroid of the first cluster regarding the group of products associated with the supplier member is calculated, and the centroid of the second cluster regarding the other group of products associated with the buyer member is calculated, and the similarity between the supplier member associated with the group of products and the buyer member associated with the other group of products can be calculated according to the distance between each centroid of the first cluster and the centroid of the first cluster, for example, based on the distance between the centroid of the first cluster regarding the group of products and the centroid of the second cluster regarding the other group of products calculated in an embedding space or Based on the angle between the center of the first cluster regarding a group of products and the center of the second cluster regarding a different group of products, the similarity between the first cluster regarding a group of products linked to a supplier member and the second cluster regarding a different group of products linked to a buyer member can be calculated, and the similarity between the center of the first cluster regarding a group of products linked to a supplier member and the center of the second cluster regarding a different group of products linked to a buyer member can be used as the similarity between the tendency of the group of products linked to a supplier member and the tendency of the other group of products linked to a buyer member or the similarity between the corresponding supplier member and buyer member, and the similarity between the calculated supplier member and buyer member can be calculated as a score.

[0044] In one embodiment of the present invention, the creation of a first cluster and a second cluster linked to a purchasing member and a buyer member, respectively, which are the targets for calculating the matching score, and the calculation of the centers of these first and second clusters can be predicted from a clustering algorithm such as k-means clustering. For example, in the k-means clustering, a plurality of data are clustered into any k clusters (corresponding to hyperparameters), and the cluster centers are calculated for each of the k clusters. The clustering can be terminated at a final stage where the cluster centers converge and the cluster centers do not change in any further iterations.

[0045] More specifically, k-means clustering applicable in one embodiment of the present invention can cluster multiple data through the following process.

[0046] 1. Select k random cluster centroids

[0047] 2. Calculate the distance between the cluster centroids and each data point, and include each data point in the cluster with the minimum distance.

[0048] 3. Calculate / update the centroids of the clusters newly generated as a result of 2.

[0049] 4. Repeat processes 2 and 3 until the cluster centroid no longer changes.

[0050] In various embodiments of the present invention, a group of products associated with a supplier member is formed as a first cluster, and another group of products associated with a buyer member is formed as another second cluster. A similarity or matching score between a supplier member and a buyer member can be calculated based on the distance between the center of the first cluster and the center of the second cluster, or based on the angle between the center of the first cluster and the center of the second cluster, and a separate clustering algorithm such as k-means clustering may not be applied. For example, such separate clustering may not be performed in a distribution where the embedding representation or individual data regarding the embedding representation of a group of products associated with a supplier member in the embedding space, or the embedding representation or individual data regarding the embedding representation of another group of products associated with a buyer member in the embedding space, are densely distributed relative to each other and are generally not dispersed. However, in a distribution where the embedding representation or individual data regarding the embedding representation of a group of products associated with a purchasing member in the embedding space, or the embedding representation or individual data regarding the embedding representation of another group of products associated with a buying member in the embedding space, are not densely clustered with each other but are dispersed over a wide range, a clustering algorithm such as k-means clustering may be applied, and this may occur in instances where a purchasing member engages in wholesale business of products with different tendencies or a buying member engages in retail business of products with different tendencies.In one embodiment of the present invention, when the distribution of products associated with one member among the purchasing member and the buying member, which are the subjects of the calculation of the matching score, is distributed over a wide space in the embedding space, they can be clustered into multiple clusters as described above, and accordingly, in such instances, the matching between the purchasing member and the buying member can be implemented for each cluster. For example, regarding a first cluster specified among multiple clusters regarding a group of products associated with a purchasing member, it can be matched with a second cluster regarding another group of products of the first buying member, and regarding a second cluster specified among multiple clusters regarding a group of products associated with a purchasing member, it can be matched with a second cluster regarding another group of products of the second buying member. At this time, the processing server can recommend the first buying member in a one-to-one relationship for the products of the first cluster to the corresponding purchasing member, and can recommend the second buying member in a one-to-one relationship for the products of the second cluster.

[0051] In one embodiment of the present invention, attribute data of a group of products and another group of products linked to a purchasing member and a buyer member, respectively, which are the targets for calculating the matching score, may take the form of a one-dimensional vector in which attributes regarding each product are used as factors or column elements. For example, when generating attribute data in which individual attributes such as product category, product name, keyword (or search term), color tone, price, material (substance), and size (length) are used as factors or column elements as described above, for example, the product name and keyword (or search term) may include multiple word expressions. Such multiple word expressions may be input into an encoder capable of generating lower-dimensional latency expressions to extract latency expressions regarding the multiple word expressions, and attribute data in which such lower-dimensional latency expressions are used as column elements of attribute data regarding each product may be generated.

[0052] In one embodiment of the present invention, a RAG may be applied to calculate similarity based on the distance or angle between embedding expressions mapped to adjacent embedding spaces according to semantic search or semantic similarity; in the preprocessing step, attribute data related to each product may be used as input to an embedding model, and embedding expressions (vectorized expressions) may be stored in a vector DB; and in the service step for calculating a matching score between a supplier member and a buyer member, embedding expressions may be extracted from the vector DB as attribute data of the corresponding product, and a matching score between a supplier member and a buyer member may be calculated from a distance-based similarity analysis or an angle-based similarity analysis between the extracted embedding expressions. In various embodiments of the present invention, when calculating the matching score between the supplier member and the buyer member, the preprocessing step as described above is not implemented; instead, in the service step, that is, in the step for calculating the matching score, each of the products in a group associated with the supplier member corresponding to the target of the matching score calculation can be input into an embedding model and an embedding representation for each product can be generated, and each of the products in another group associated with the buyer member can be input into an embedding model and an embedding representation for each product can be generated. In this way, from the embedding representations mapped on the same embedding space, as previously described, similarity or the matching score between the supplier member and the buyer member can be calculated based on the distance between the center of the first cluster regarding each group of products and the center of the second cluster regarding other groups of products, or based on an angle.

[0053] In one embodiment of the present invention, based on the calculated matching score, the establishment of a one-to-one relationship may be recommended according to the priority of a buyer member with a relatively higher matching score calculated between a supplier member with a calculated matching score and a buyer member with a relatively higher matching score. For example, in one embodiment of the present invention, a processing server operating or managing a commerce platform may transmit a request to inquire about acceptance regarding the establishment of a one-to-one relationship with a buyer member with a relatively higher matching score calculated with the corresponding supplier member. Depending on whether the corresponding supplier member accepts the request, a one-to-one relationship may be established between the supplier member and the buyer member for whom a one-to-one relationship was recommended (the buyer member with a relatively higher matching score with the corresponding supplier member). For example, in one embodiment of the present invention, a one-to-one relationship may be established as a different relationship of closeness according to the policy of the commerce platform, and may be referred to by different expressions such as first cousin, neighbor, or friend.

[0054] In one embodiment of the present invention, the establishment of a one-to-one relationship may be initiated by a request from a processing server to query a purchasing member regarding acceptance of the establishment of a one-to-one relationship based on a matching score calculated by the processing server, without relying on mutual consent of the parties to the establishment of the one-to-one relationship or on a request from one party and acceptance by the other party. Depending on the purchasing member's response (acceptance response) to the processing server's request, the one-to-one relationship may be established, or depending on the purchasing member's response (rejection response), the one-to-one relationship may not be established. For example, the processing server may recommend multiple purchasing members based on the matching score calculated for a purchasing member, and may request individual acceptance or rejection responses regarding the establishment of a one-to-one relationship with multiple purchasing members from the corresponding purchasing members.

[0055] In one embodiment of the present invention, a processing server may establish a one-to-one relationship defined according to the policy of a commerce platform. Regarding a product that has been set to private by the corresponding supplier member to prevent imitation, the processing server may keep the product information (e.g., product image) private from general buyer members, but disclose the product information (e.g., product image) to buyer members with whom a one-to-one relationship has been established between the supplier member and the buyer member with whom the one-to-one relationship has been established. For example, regarding the one-to-one relationship in one embodiment of the present invention, the processing server may disclose the image of the private product, which has been set to private by the supplier member, to buyer members with whom the one-to-one relationship has been established with the supplier member. In one embodiment of the present invention, the processing server may fundamentally block the disclosure of the image of the private product, for example, not disclose it to general members, but disclose the image of the product to buyer members with whom the one-to-one relationship has been established with the corresponding supplier member. For example, for products belonging to small accessories such as children's clothing or jewelry, images may be kept private for general members to prevent imitation, while being disclosed to buyer members with whom a one-to-one relationship has been established. Through such one-to-one relationships, purchasing members can increase sales by blocking imitation of products with a relatively high risk of imitation while simultaneously disclosing product images to buyer members with whom a one-to-one relationship has been established.

[0056] In one embodiment of the present invention, a matching score calculated between a supplier member and a buyer member may be calculated based on the similarity between the products registered by the supplier member and the products purchased by the buyer member (e.g., average similarity between a number of registered products and a number of purchased products, similarity based on the distance or angle between the center of a first cluster of a number of registered products and a second cluster of a number of purchased products). Herein, the similarity analysis between a number of registered products and a number of purchased products may be implemented based on attribute data corresponding to the attribute information of each product, and the attribute data regarding each product may be based on the detailed information of the products registered by the supplier member. For example, the detailed information of the products registered by the supplier member may include attribute information regarding the registered products, such as product category, product name, keyword or search term, color tone, price, material, and size.

[0057] In a commerce platform according to one embodiment of the present invention, feedback regarding detailed product information entered by a corresponding supplier member can be provided to a supplier member who registers a product. For example, feedback can be provided regarding the product itself or the detailed information of the registered product by the supplier member by reflecting the preferences of buyers that change over time or current trends. For example, in one embodiment of the present invention, a comparative analysis between a registered product and a trend product with the highest current buyer preference can be provided based on the similarity between the registered product registered by the corresponding supplier member and the trend product that shows the highest preference within the same product category, provided that the registered product belongs to the same product category. In one embodiment of the present invention, buyer preference may refer to the purchase ranking in which purchases are made by the buyer. For example, among products belonging to the same product category as the product registered by the supplier member, the product with the highest purchase ranking may be selected as the trend product, and feedback may be provided to enhance competitiveness regarding the registered product itself or detailed information about the registered product based on a comparative analysis between the product registered by the corresponding supplier member and the trend product according to buyer preference.

[0058] For example, a comparative analysis between the registered product of a relevant supplier member and a trend product can calculate the deviation of the attribute value between the attribute value of the supplier member's registered product and the attribute value of the trend product for each product attribute, and for example, the deviation between the attribute value of the supplier member's registered product and the trend product can be expressed from the distribution of the corresponding attribute value of all products, using the entire product category to which the relevant product belongs as the population, for example, and the deviation between the supplier member's registered product and the trend product can be expressed as the relative ratio of the deviation between the supplier member's registered product and the trend product to the total distribution of attribute values ​​(total span) on the distribution of the corresponding attribute value of the entire population. As such, the deviation between the registered products of the corresponding supplier member and the trend products relative to the distribution of the entire population can be expressed in attributes such as product name, keyword or search term, product price, color tone, product material, and size among the product details of the registered products obtained from the supplier member. For example, in one embodiment of the present invention, separate scaling may not be required for the product price or product size. However, scaling for the product material may be performed based on a similarity analysis between the product materials or product materials, for example, through a similarity analysis between product materials.In one embodiment of the present invention, as described above, products associated with a supplier member and a buyer member can be mapped into an embedding space according to the attribute data of each product, and similarity between different products can be calculated based on the distance between embedding representations mapped into the embedding space or the angle between embedding representations mapped into the embedding space, and similarly, similarity analysis can be performed based on distance or angle between embedding representations extracted from an embedding model that takes the material (substance) of the product as an attribute of the product as input, and the deviation between the material (substance) of the supplier member and the trend product can be estimated based on the score calculated from the similarity analysis.

[0059] In one embodiment of the present invention, feedback for each attribute can be provided by extracting respective attribute values ​​from the attribute data of a product registered by a purchasing member and the attribute data of a trending product, and calculating the deviation for each attribute value. In one embodiment of the present invention, in the task of calculating a matching score between a supplier member and a buyer member by calculating a similarity between a product registered by a supplier member and a product purchased by a buyer member as described above, an embedding expression regarding all attributes, not individual attributes, is calculated through an embedding model from the entire one-dimensional array regarding each individual attribute, that is, the attribute data of the product linked to the supplier member and the buyer member. However, in the task of providing feedback on purchasing trends from a similarity analysis between a product registered by a supplier member and a trend product, an embedding expression for each attribute can be calculated without calculating an entire embedding expression for the attribute data of each product as described above. However, since individual attribute values ​​regarding price or size, which are already expressed as numerical values, are expressed as scaled numerical values ​​and the deviation between them can be calculated through numerical calculation, there is no need to extract a separate embedding expression. However, for materials (materials) that are not expressed as numerical attribute values, it may be necessary to calculate a separate embedding expression.

[0060] In this way, in a task for comparing and analyzing product attributes in one embodiment of the present invention, differential processing can be implemented according to the purpose assigned to each task. For example, in a task for calculating similarity between a product registered by a supplier member and a product purchased by a buyer member for matching between a supplier member and a buyer member, an embedding representation for the entire attribute data of each product can be calculated. In contrast, in a task for calculating similarity between a product registered by a supplier member and a trend product for providing feedback on purchasing trends for a supplier member, an embedding representation for the entire attribute data of each product can be calculated, or an embedding representation can be calculated at the individual attribute unit forming the attribute data of each product, or attribute values ​​can be compared and analyzed at the individual attribute unit. For example, in one embodiment of the present invention, the task for matching between a supplier member and a buyer member is to calculate the overall similarity between the supplier member's registered product and the buyer member's purchased product to calculate a matching score between the supplier member and the buyer member who are the matching targets. Therefore, it is necessary to analyze the overall similarity between the supplier member's registered product and the buyer member's purchased product. However, in order to provide feedback on purchasing trends for the supplier member, it is required to analyze the similarity of individual product attributes. Therefore, rather than calculating a score that combines the various attributes of each product into one, it is necessary to compare and analyze the various attributes of each product at the individual attribute level. For example, in order to provide feedback regarding the supplier member's registered product or feedback regarding detailed information on the registered product, feedback regarding specific attributes of the supplier member's registered product is provided. In order to stimulate the supplier member's interest in specific attributes, feedback to the supplier member can be implemented at the individual attribute level.

[0061] In one embodiment of the present invention, in a task for matching between a supplier member and a buyer member, the product category is considered as a primary attribute of the product, for example, as a primary attribute for determining similarity between a product registered by a supplier member and a product purchased by a buyer member, or for calculating a matching score between a supplier member and a buyer member; however, in a task for analyzing similarity between a product registered by a supplier member and a trend product, or for providing feedback on purchasing trends to a supplier member, the trend product is selected as the product with the highest buyer preference or purchase ranking within the product category to which the supplier's registered product belongs, so there is no need to separately consider the product category as an attribute, and since the product registered by a supplier member and the trend product can belong to the same product category, the attribute value of the product category between them can be predicted to be identical.

[0062] In one embodiment of the present invention, an attribute of a keyword entered by a supplier member may be entered as an attribute of a product. For example, in one embodiment of the present invention, the attribute of the keyword may refer to an attribute entered as detailed product information for product registration so that a corresponding product can be searched in response to a query matching the keyword as a buyer's query. For example, in one embodiment of the present invention, in the similarity analysis between products for matching between a supplier member and a buyer member, a low-dimensional latency representation or context vector to express the context of the entire text sequence may be extracted, similar to processing the attribute of the product name, and attribute data may be constructed using the extracted latency representation or context vector as a single argument or column element along with the value of another attribute. For example, in one embodiment of the present invention, for a text sequence that may include multiple word expressions, for example, in which multiple word expressions may be expressed in a form separated by special characters such as spacers or commas, an embedding expression that expresses the context regarding the entire input text sequence may be extracted from an input embedding in which each word expression is input sequentially. For example, attributes that may be expressed in the form of a combination of multiple word expressions as described above may be examples such as product names and keywords. For such a combination of multiple word expressions, a context vector or a low-dimensionally reduced latency expression that expresses the entire context of the combined entire text sequence may be generated (for example, generated from an encoder in an architecture of a transformer including an encoder and a decoder), and the generated context vector or latency expression may be used as one attribute value, and attribute data may be expressed in a form that combines various other attribute values.

[0063] As such, unlike the task for matching supplier members and buyer members, the task for providing feedback to supplier members performs a comparative analysis between the attribute values ​​of each attribute combined in the attribute data for each product. Accordingly, it is necessary to provide feedback regarding the attribute values ​​of specific attributes to each supplier member; for example, it is necessary to calculate the deviation from the attribute values ​​of specific attributes based on a comparative analysis with trend products. For example, the supplier member receiving the feedback needs to receive feedback regarding specific attributes of the products registered by the supplier member. For instance, in the comparative analysis with trend products, rather than comparing the overall attributes of the registered products one has registered with the overall attributes of the trend products (for example, a comparison of the entire product attribute is implemented in the task for matching supplier members and buyer members through a comprehensive comparison of the products registered by the supplier member and the buyer member), a specific attribute among the attributes of the products registered by the supplier member is compared with the individual attributes of the trend products. This is done in a comparative analysis with trend products that are currently ranked high in purchase rankings. Given that it is desirable to provide feedback regarding whether competitiveness is relatively low, it may be desirable to compare and analyze individual attributes between products registered by supplier members and products purchased by buyer members.

[0064] In one embodiment of the present invention, regarding the calculation of deviations by individual attributes between a purchased member's registered product and a trend product, the deviation of individual attributes can be calculated in the form of a difference for 1) attribute values ​​of quantified attributes (price, size, etc.), and 2) for attributes expressed as a combination of multiple word expressions (product name, keyword), a one-dimensional vector value can be calculated, for example, by calculating a low-dimensional latency expression for each of the multiple word expressions forming the attribute or a context vector representing the entire context of the multiple word expressions, and the deviation between these attributes can be calculated from a distance-based or angle-based similarity calculation between such latency expressions or context vectors regarding the entire context. However, in one embodiment of the present invention, among attributes encompassing multiple word expressions, the keyword attribute may exceptionally not be processed in the manner described in 2) above.For example, in a task for matching between a supplier member and a buyer member, in the analysis of the overall attribute similarity between products associated with the supplier member and products associated with the buyer member, for instance, for attributes containing multiple word expressions (product name, keyword, etc.), processing as described in 2) above may be performed, and the similarity regarding the entire attribute data containing the attribute value of the corresponding attribute in a designated column position is calculated using a context vector or latent expression implying multiple word expressions. However, in a task for providing feedback to a supplier member, unlike this, for the product name attribute among attributes containing multiple word expressions (product name, keyword), a latent vector or context vector is calculated as in 2), and distance-based or angle-based similarity between them is calculated; meanwhile, for the keyword attribute among the attributes containing multiple word expressions, the deviation between the supplier member's registered products and trend products may be calculated based on the number of keywords. Thus, rather than performing similarity analysis based on semantics that extract latent expressions or context vectors as in the processing of 2) for the keyword attribute among attributes containing multiple word expressions, the word expressions forming the keyword attribute By quantifying the data based on the number of items, it is possible to compare and analyze different products, that is, products registered by purchasing members and trend products, from the difference of the quantified attributes as in 1).

[0065] In one embodiment of the present invention, the products registered by the purchasing member and the trend products belong to the same product category but may be classified as different products. Accordingly, providing feedback to assign identity to the keywords or search terms themselves in the classification between different products may not be desirable in terms of product differentiation. However, for example, even if they are different products, depending on the number of word expressions included in the attributes of the keywords or search terms, they may be included in the list that is replied to or returned in response to a query from a buyer, or they may not be included in the list that is replied to or returned in response to a query from a buyer. In this way, in one embodiment of the present invention, when the word expressions included in the keyword or search term attribute are sparse as attributes of a registered product of a purchasing member compared to a trend product that includes multiple word expressions included in the keyword or search term attribute, in order to encourage the purchasing member's interest in the product, for example, in the case where the number of word expressions included in the keyword or search term attribute in the detailed information of a registered product registered by a purchasing member is relatively sparse and the frequency of exposure included in the reply to a buyer's query or in the list of returned products may be relatively low, for example, the frequency of exposure can be increased by taking an action to increase the number of word expressions included in multiple search terms or keywords, for example, among attributes that include multiple word expressions, for the keyword or search term attribute, a numerical attribute value can be calculated from data quantified according to the number of word expressions forming the corresponding keyword or search term, and a deviation can be calculated from the difference of the numerical attribute value.In one embodiment of the present invention, for the comparative analysis of attributes for each product, an embedding expression regarding the entire attribute data or each individual attribute can be extracted for semantic search, and a vector DB for storing such embedding expressions can be operated; however, for product search for replying to or returning a query input by a buyer member, a search of a database centered on key values ​​can be performed, and accordingly, product search based on whether a search term or keyword matches can be implemented, as described above, in the comparison of attributes of a search term or keyword, a comparison or difference between attributes quantified by the number of word expressions forming the search term or keyword can be performed. In this way, in one embodiment of the present invention, a list of searched products is generated by inputting a query of keywords or search terms for product search by a buyer member. In one embodiment of the present invention, considering product search based on key values ​​of keywords or search terms, a numerical deviation can be calculated from the difference of quantified attributes according to the number of word expressions forming the keywords or search terms of each product in the attributes of keywords or search terms of registered products and trend products of a supplier member.

[0066] Although the present invention has been described with reference to embodiments illustrated in the attached drawings, this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom.

[0067] The present invention can be used in industrial fields where a commerce platform can be applied.

Claims

1. As a commerce platform that mediates commercial transactions between supplier members and buyer members, A matching task for calculating a matching score between a supplier member and a buyer member from a similarity analysis between a product registered by a supplier member and a product purchased by a buyer member, and querying the supplier member regarding acceptance of establishing a one-to-one relationship between the supplier member and the buyer member based on the calculated matching score; and A commerce platform including a processing server that performs a feedback provision task to provide feedback reflecting purchasing trends regarding registered products and detailed information of registered products registered by purchasing member.

2. In Paragraph 1, The above processing server is, In the above matching task, for each of the products registered by the supplier member and the products purchased by the buyer member that are the target for calculating the matching score, a similarity analysis is performed on the entire set of attribute data combining various different attributes of the products, and A commerce platform characterized by performing a similarity analysis regarding the individual attributes of each product for each of the registered products of the purchasing member corresponding to the feedback provision target and the trend products selected according to purchase preference or purchase ranking in the above feedback provision task.

3. In Paragraph 1, The above processing server is, In the above matching task, for each of the products registered by the supplier member and the products purchased by the buyer member that are the target for calculating the matching score, an embedding representation of the entire attribute data combining various different attributes of the products is calculated, and similarity analysis is performed between the calculated embedding representations based on distance or angle. A commerce platform characterized by, in the above feedback provision task, calculating deviations for each attribute forming attribute data combining various different attributes for each of the trend products selected according to purchase preference or purchase ranking and the registered products of the purchasing member corresponding to the feedback provision target, and providing feedback on purchasing trends to the purchasing member regarding the calculated deviations.

4. In Paragraph 1, The above processing server is, Based on the matching score calculated between the above-mentioned supplier member and buyer member, and the supplier member's response regarding the acceptance or rejection of the request to establish a one-to-one relationship with the supplier member, a one-to-one relationship is established between the above-mentioned supplier member and buyer member, and A commerce platform characterized by not disclosing product images to general buyer members for products set to private at the request of the above-mentioned supplier member, but disclosing product images to buyer members with whom a one-to-one relationship has been established with the relevant supplier member.

5. In Paragraph 1, The above processing server is, A commerce platform characterized by, in the above-mentioned feedback provision task, for each of the registered products and trend products of a supplier member, numerically calculating the deviation of an attribute from an attribute quantified by the number of word expressions forming the attribute of a keyword or search term among each attribute forming attribute data in which different attributes are combined.

6. In Paragraph 1, The above processing server is, A commerce platform characterized by, in the above matching task, for each of the products registered by a supplier member and the products purchased by a buyer member, calculating a dimensionally reduced latency expression or a context vector regarding the entire context of a plurality of word expressions that form the attributes of a keyword or search term among each attribute forming attribute data formed by combining different attributes, generating attribute data that includes the calculated latency expression or context vector as a specified column element of the corresponding keyword or search term, and performing a similarity analysis on the attribute data generated for the products registered by the supplier member and the attribute data generated for the products purchased by the buyer member.

Images