KR-102960683-B1 - METHOD AND ELECTRONIC DEVICE FOR CALCULATING PURCHASE LIKELIHOOD BASED ON USER CONTEXT INFORMATION AND ADAPTIVELY CHANGING RECOMMENDATION STRATEGIES

KR102960683B1KR 102960683 B1KR102960683 B1KR 102960683B1KR-102960683-B1

Abstract

A method and system for recommending products and distributing profits based on reliability are disclosed. A method of operation of an electronic device according to one embodiment of the present disclosure may include: an operation of training a neural network model based on purchase history information corresponding to each of a plurality of users; an operation of inputting context information of a first user into the trained neural network model to generate recommendation information including information related to a product purchased by an acquaintance of the first user; an operation of identifying at least one user among the plurality of users who contributed to the generation of the recommendation information when it is identified that the first user has purchased a first product based on the generated recommendation information; and an operation of identifying profits to be provided to the identified at least one user in relation to the first product.

Inventors

김연재

Assignees

주식회사 어바웃피싱

Dates

Publication Date: 20260507
Application Date: 20260330

Claims (1)

In the method of operating an electronic device, The operation of training a neural network model based on purchase history information corresponding to each of multiple users; The operation of inputting the context information of the first user into the learned neural network model to generate recommendation information including information related to a product purchased by the first user's acquaintance; If it is identified that the first user has purchased the first product based on the generated recommendation information, the operation of identifying at least one user among the plurality of users who contributed to the generation of the recommendation information; and The operation of identifying a profit to be provided in connection with the first product to at least one identified user; The operation of training the above neural network model is, The neural network model is trained based on the behavioral history information of the first user, the past context information of the first user, the social network information of the first user, and interaction information corresponding to the first user. The above method of operation is, The operation of displaying a UI (user interface) containing the above-mentioned generated recommendation information through a display; further comprising The above behavioral history information includes information on the past purchase history of the first user and information on the search behavior of the first user, and The above past context information includes past conversation information between the learned neural network model and the first user, and The above social network information includes purchase history information corresponding to each of the plurality of users, and relationship information between the first user and the plurality of users. The above interaction information includes relationship information between the learned neural network model and the first user, and The operation of generating the above recommendation information is, An operation to identify at least one acquaintance connected to the first user in the first step based on the social network information of the first user; An operation to identify a plurality of products associated with a product purchased by at least one acquaintance based on purchase history information corresponding to each of the at least one acquaintance identified above; For each of the above plurality of products, an operation of calculating a first reliability related to acquaintance reliability based on the social network information of the first user; For each of the above plurality of products, an operation of calculating a second reliability related to product suitability based on the behavioral history information of the first user; For each of the above plurality of products, an operation of calculating a third reliability related to the first user's purchase possibility based on the first user's context information; The operation of applying a first weight to the first reliability calculated above, applying a second weight to the second reliability calculated above, and applying a third weight to the third reliability calculated above to calculate a total reliability corresponding to each of the plurality of products; An operation to calculate a fourth reliability between the first user and the learned neural network model based on interaction information corresponding to the first user; If the above-mentioned fourth reliability is identified as being greater than or equal to the first value, an operation to perform a first update process of multiplying the first weight by a pre-set first ratio, multiplying the second weight by a pre-set second ratio, and multiplying the third weight by a pre-set third ratio; If the fourth confidence level is identified as being less than the first value, an operation to perform a second update process of multiplying the first weight by a preset fourth ratio, multiplying the second weight by a preset fifth ratio, and multiplying the third weight by a preset sixth ratio; and After the first update process or the second update process is performed, the operation of generating recommendation information related to the product with the relatively largest magnitude of the updated overall reliability; is included. The first ratio is a value greater than 1.5, the second ratio is a value less than 1, the third ratio is a value greater than 1 and less than 1.5, the fourth ratio is a value less than 1, the fifth ratio is a value greater than 1.5, and the sixth ratio is a value greater than 1 and less than 1.5. The relationship information between the first user and multiple users mentioned above is, It includes explicit trust relationship information, implicit trust relationship information, and network influence information, and The above explicit trust relationship information is, Includes whether the user is friends with the first user and whether the first user consents to recommending a product purchased by the friend, and The above implicit trust relationship information is, It includes information on the frequency of conversation and interaction patterns between the first user and other users, and The above network influence information is, A method of operation including information about a user's influence within a social network.

Description

Method and electronic device for calculating purchase likelihood based on user context information and adaptively changing recommendation strategies The present disclosure relates to a method and system for recommending products and distributing profits based on reliability. In the recent e-commerce market, efficient product recommendations and marketing management between small business owners, SMEs, and various types of consumers (individual buyers, reviewers, influencers, etc.) are becoming increasingly important. In this context, an "e-commerce company" refers to a business operator selling products online, while a "user" refers to an individual who purchases products or participates in recommendations. Various information, such as purchase history, is utilized in the recommendation process between e-commerce companies and users, and the reliability and real-time nature of this information, as well as the transparency of revenue distribution, can determine the competitiveness of a recommendation platform. Conventional product recommendation services operated primarily through collaborative or content-based filtering, requiring e-commerce companies to post advertisements and users to go through complex procedures such as clicking, purchasing, and reviewing after viewing them. In this process, both e-commerce companies and users faced various problems, including false or exaggerated advertising, opaque basis for recommendations, non-disclosure of revenue sharing standards, and ambiguity in measuring marketing effectiveness. In particular, conventional influencer marketing services operate with a structure where revenue is concentrated among a small number of influencers. This structure can lead to general users being excluded from revenue generation opportunities, and the overall credibility of the platform suffers from revenue distribution that does not accurately reflect users' actual contributions. Furthermore, the lack of systematic management systems—such as tracking purchase conversions and calculating contributions after a recommendation—has resulted in issues with the efficiency and reliability of marketing operations. Furthermore, existing recommendation systems suffered from the cold start problem caused by the scarcity of user-item interaction data, limitations in reflecting the complexity of real human relationships by relying solely on simple similarity calculations, and a failure to adequately consider trust relationships or social influence among users. In particular, AI model-based recommendation systems provided only unidirectional recommendations from the system to the user, and due to limitations in natural interaction and the formation of trust between the user and the AI model, the authenticity and level of personalization of recommendations were restricted. The fundamental cause of these problems lies in the absence of a mechanism in existing AI recommendation systems to quantify and dynamically reflect trust relationships between users. Conventional collaborative filtering methods failed to fundamentally resolve the "cold start" problem, where recommendation quality for new users or products drops significantly due to the sparsity of user-item interaction data. Furthermore, relying solely on simple ratings or purchase history similarity, they failed to adequately reflect the complex trust relationships and social influence among actual users. Additionally, existing AI models were unable to learn trust through real-time user interaction, resulting in limitations in dynamically reflecting changes in users' trust levels or personal preferences. Consequently, recommendation systems could not accurately grasp users' current situations and needs, nor could they systematically integrate the experiences or opinions of trusted acquaintances into the recommendation process, inevitably limiting the persuasiveness of their recommendations and conversion rates. Accordingly, there is a need for a trust-based recommendation system and an automated revenue distribution method that is reasonably acceptable to all participants using the recommendation service (e-commerce companies, general users, and AI Friends owners), as well as a method for selecting and recommending products with a high matching probability based on trust relationships between users and between users and personalized AI models, and for real-time revenue distribution based on contribution when a user makes a purchase. FIG. 1 is a block diagram showing the configuration of an electronic device according to one embodiment. FIG. 2 is a flowchart illustrating a method of operation of an electronic device according to one embodiment. FIG. 3 is a flowchart illustrating a method for generating recommendation information according to one embodiment. FIG. 4 is a flowchart illustrating a method for generating recommendation information according to one embodiment. FIG. 5 is a flowchart illustrating a method for identifying at least one user