KR-102961947-B1 - SERVER, SYSTEM, METHOD AND PROGRAM FOR PROVIDING SUBSCRIPTION RECOMMENDATION SERVICE USING USER DATA

Abstract

According to an embodiment of the present invention, a LangGraph-based cross-platform content recommendation system is provided for providing personalized content within a subscription-based content service by utilizing user behavior pattern data. The system comprises: a classification database that structures metadata including content genre, characters, length, structured summary, and campaign status, and stores the structured data along with the original text by vectorizing it through a classification model; an embedding database that vectorizes unstructured text data including structured summaries, reviews, and tags using an embedding technique, and stores the vector and the original text together; an initial state setting node that initializes the user state by collecting the user's viewing history, interaction logs, and feedback data; a search node that selects candidate content based on cosine similarity using the classification database; an evaluation node that performs a precise similarity evaluation using a Large Language Model (LLM) based on structured features and text embeddings for the candidate content; and a feedback reflection node that analyzes the user's behavior logs and natural language feedback to update weights, genre lists, and exclusion lists, and transmits them to the initial state setting node. It includes a cross-platform recommendation node that searches and evaluates content across multiple platforms to produce platform-specific recommendation results, wherein the cross-platform recommendation node is executed only when the number of recommendations, the number of completed content viewings, and the number of feedbacks are greater than or equal to preset threshold values, and a classification database and an embedding database are stored classified by platform, and search and evaluation using the search node and the evaluation node are performed independently for each platform, and each node references and updates a common state object to perform the search, evaluation, and feedback processes of recommendations within a consistent graph structure.

Inventors

• 이석준

Assignees

• 주식회사 피클플러스

Dates

Publication Date

20260507

Application Date

20251021

Claims (5)

1. In a LangGraph-based cross-platform content recommendation system for providing personalized content within a subscription content service by utilizing user behavior pattern data, A classification database that structures metadata including content genre, characters, length, structured summary, and campaign status, vectorizes the structured data through a classification model, and stores it together with the original text; An embedding database that vectorizes unstructured text data including structured summaries, reviews, and tags using embedding techniques, and stores the vectors and original text together; An initial state setting node that initializes the user state by collecting the user's viewing history, interaction logs, and feedback data; A search node that selects candidate content based on cosine similarity using the above classification database; An evaluation node that performs a precise similarity evaluation using a Large Language Model (LM) based on structured features and text embeddings for the above candidate content; A feedback reflection node that analyzes user behavior logs and natural language feedback to update weights, genre lists, and exclusion content lists, and transmits them to an initial state setting node; and It includes a cross-platform recommendation node that explores and evaluates content across multiple platforms to produce platform-specific recommendation results, and The above cross-platform recommendation node is executed only when the number of recommendations, the number of completed content viewings, and the number of feedbacks are above a preset threshold value, the classification database and the embedding database are stored classified by platform, and search and evaluation using the above search node and the above evaluation node are performed independently for each platform, and Each of the above nodes references and updates a common state object to perform the search, evaluation, and feedback processes of recommendations within a consistent graph structure, and The above evaluation node is, A structured feature evaluator that calculates similarity for quantitative attributes such as genre, character composition, and length; A summary similarity evaluator that calculates qualitative similarity by comparing structured summary information of an embedding database with an LLM chain; and It includes a bonus and penalty calculator that calculates adjustment values by reflecting whether it is a new release, promotion, oversaturated genre, or unexplored genre, and The final score for each content is calculated by the following mathematical formula, and g i is the genre similarity value, ch i is the character similarity value, d i is the length similarity value, σ i is the content similarity value, b i is the bonus/penalty correction value, and wg, wch, wd, and ws are the weighting coefficients of the respective similarity values, and The above feedback reflection node is, An interaction feedback processing module that analyzes behavioral data such as click-through rates, completion rates, and hover rates to adjust weights based on rules; and It includes a direct feedback processing module that interprets user natural language feedback through LLM to update weights and genre lists, and The two modules above sequentially share the same state object, but operate in a sequential overwrite structure where the result of the subsequent module overwrites the result of the previous step. Characterized by being configured so that the user's explicit intent is prioritized in the final recommendation parameters, Cross-platform content recommendation system.
2. delete
3. delete
4. In paragraph 1, The above evaluation refers to a state object to synthesize the output values of each of the above evaluation devices, and After sequentially recording the calculated genre similarity (gi), character similarity (chi), length similarity (di), content similarity (σi), and correction value (bi) for each candidate content into the state object, Calculate a weighted sum score by reflecting the weights of the above state objects, and Characterized by being configured to store the calculated result as the final recommendation list in the recommendation field of a state object, Cross-platform content recommendation system.
5. In paragraph 4, The above cross-platform recommendation node is, When activated, candidate content is searched in the classification database and embedding database for each platform using the user preference vector included in the state object, and The searched candidate content is passed to the evaluation stage to calculate similarity scores for genre, character, length, and summary, and then the final score is calculated by adding a platform bonus to the arithmetic mean of each score. Selecting content from platforms where the above final score exceeds a threshold as a recommendation target, sorting it in order of the above final score, and storing it in the cross-platform recommendation field of the state object, Cross-platform content recommendation system.

Description

Server, system, method, and program for providing subscription recommendation service using user data The present invention relates to a server, system, method, and program for providing a subscription recommendation service utilizing user data. Unless otherwise indicated in this specification, the contents described in this section are not prior art for the claims of this application, and are not to be recognized as prior art simply because they are included in this section. Recently, personalized recommendations have become a key competitive advantage for OTT and subscription content platforms. However, existing recommendation systems are limited to data within a single platform, leading to a significant drop in personalization efficiency when users move to other platforms or use various content services simultaneously. Furthermore, existing collaborative filtering or single-embedding-based recommendation methods cause data sparsity, cold start issues, or oversaturation centered on the same genre. Furthermore, differing data schemas and policies across platforms make consistent recommendations difficult in a cross-platform environment, and fixed recommendation thresholds hinder the reflection of users' exploration needs. Existing systems also faced limitations in ensuring diversity because they failed to account for differences in content catalogs across platforms, even when utilizing accumulated user preference data. Therefore, an advanced recommendation structure is required that can reuse users' accumulated preference vectors while comprehensively considering multiple streaming platforms. The matters described in the background technology above are intended to aid in understanding the background of the invention and may include matters that are not disclosed prior art. FIG. 1 is a schematic diagram of a system according to one embodiment of the present invention. FIG. 2 is a block diagram illustrating the configuration and data flow of a LangGraph-based personalized content recommendation system according to one embodiment of the present invention. FIG. 3 is a diagram illustrating a Langgraph state transition flow according to an embodiment of the system according to FIG. 2. FIG. 4 is a diagram illustrating a Langgraph state transition flow according to another embodiment of the system according to FIG. 2. FIG. 5 is a block diagram illustrating the configuration and data flow of a LangGraph-based content search system according to one embodiment of the present invention. FIG. 6 is a diagram illustrating a Langgraph state transition flow according to one embodiment of the system according to FIG. 5. Figure 7 is a diagram illustrating the hardware configuration of a service provider server according to Figure 1 in an exemplary manner. The present invention is susceptible to various modifications and may have various embodiments; specific embodiments are illustrated in the drawings and described in detail in the detailed description. However, this is not intended to limit the invention to specific embodiments, and it should be understood that the invention includes all modifications, equivalents, and substitutions that fall within the spirit and scope of the invention. Similar reference numerals have been used for similar components in the description of each drawing. Terms such as first, second, A, B, etc., may be used to describe various components, but said components should not be limited by said terms. These terms are used solely for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be named the second component, and similarly, the second component may be named the first component. The term "and/or" includes a combination of a plurality of related described items or any of a plurality of related described items. When it is stated that one component is "connected" or "connected" to another component, it should be understood that while it may be directly connected or connected to that other component, there may also be other components in between. On the other hand, when it is stated that one component is "directly connected" or "directly connected" to another component, it should be understood that there are no other components in between. The terms used in this application are used merely to describe specific embodiments and are not intended to limit the invention. The singular expression includes the plural expression unless the context clearly indicates otherwise. In this application, terms such as "comprising" or "having" are intended to specify the presence of the features, numbers, steps, actions, components, parts, or combinations thereof described in the specification, and should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. The "similarity" used in this application may be a known