Intelligent recommendation method and system for beauty makeup industry and medium

Abstract

The invention provides an intelligent recommendation method and system for the beauty makeup industry and a medium. The intelligent recommendation method for the beauty makeup industry comprises the steps of obtaining user query data; based on user query data, obtaining a user intention through a double-path intention recognition module; in response to the situation that the intention recognition confidence coefficient is smaller than a preset threshold value, a map compensation mechanism is triggered, and the map compensation mechanism queries associated information in the dynamic beauty makeup knowledge map based on the feature entities extracted by the double-path intention recognition module to correct the user intention; based on the user intention, reply content is generated in combination with the dynamic beauty makeup knowledge graph, and the reply content comprises recommendation content which is determined through a knowledge pushing algorithm and comprises personalized recommendation products and / or use suggestions; compliance detection and risk early warning are carried out on user operation behaviors through a real-time monitoring mechanism based on a rule engine; and dynamically adjusting a recommendation strategy and system parameters according to user feedback data by using a closed-loop reinforcement learning optimization mechanism.

Description

Technical Field

[0001] This invention relates to the field of intelligent recommendation in the beauty industry, and more particularly to an intelligent recommendation method for the beauty industry, an intelligent recommendation system for the beauty industry, and a computer-readable storage medium. Background Technology

[0002] Currently, most recommendation systems widely used in the beauty industry are based on traditional static algorithms, relying on limited user tags (such as age, gender, and skin type) for product matching. These systems typically employ collaborative filtering or simple knowledge base queries, with fixed recommendation logic that cannot be dynamically adjusted. For example, many e-commerce platforms' recommendation systems only make recommendations based on users' browsing history and purchase records, completely ignoring potential allergy risks and changes in skin condition. This static recommendation model not only leads to low recommendation accuracy but may also trigger skin problems by recommending products containing allergenic ingredients. Statistics show that approximately 37% of users have experienced allergic symptoms due to incompatible recommended products.

[0003] In terms of user demand analysis, existing beauty platforms mostly use single intent recognition models, such as simple NLP algorithms based on keyword matching. These technologies have an accuracy rate of less than 70% when dealing with vague user descriptions (such as "my face is very dry after using the face cream") or complex symptom expressions (such as "redness, swelling, and stinging after three consecutive days of use"). Especially when dealing with users sensitive to ingredients, they cannot accurately identify their need to avoid specific ingredients (such as alcohol or fragrance), leading to a significant deviation between recommended results and the user's actual needs, resulting in a poor user experience and a high platform complaint rate.

[0004] In order to overcome the above-mentioned shortcomings of existing technologies, there is an urgent need in this field for an intelligent recommendation technology for the beauty industry that can comprehensively solve the deficiencies of traditional beauty recommendation systems in allergy prevention and intent recognition, significantly improve the accuracy of recommendations and the safety of recommended beauty products, and increase user satisfaction. Summary of the Invention

[0005] The following provides a brief overview of one or more aspects to offer a basic understanding of them. This overview is not an exhaustive summary of all conceived aspects, nor is it intended to identify key or decisive elements of all aspects, nor to define the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form to prepare for the more detailed descriptions that follow.

[0006] To overcome the aforementioned deficiencies in existing technologies, this invention provides an intelligent recommendation method for the beauty industry, an intelligent recommendation system for the beauty industry, and a computer-readable storage medium. These methods comprehensively address the shortcomings of traditional beauty recommendation systems in allergy prevention and intent recognition, significantly improve the accuracy of recommendations and the safety of recommended beauty products, and enhance user satisfaction.

[0007] Specifically, the intelligent recommendation method for the beauty industry provided by the first aspect of the present invention includes the following steps: acquiring user query data; acquiring user intent based on the user query data through a dual-path intent recognition module, the dual-path intent recognition module including a local feature extraction model for extracting keywords and a contextual semantic understanding model for understanding the usage scenario context; triggering a graph compensation mechanism in response to the intent recognition confidence being less than a preset threshold, the graph compensation mechanism querying related information in a dynamic beauty knowledge graph based on the feature entities extracted by the dual-path intent recognition module to correct the user intent, wherein the dynamic beauty knowledge graph includes a customer layer, a SOP process layer, and a business entity layer, and supports node parameterization and real-time updates; generating response content based on the user intent and in conjunction with the dynamic beauty knowledge graph, the response content including personalized recommended products and / or usage suggestions determined by a knowledge push algorithm; performing compliance detection and risk warning on user operation behavior through a real-time monitoring mechanism based on a rule engine; and dynamically adjusting the recommendation strategy and system parameters based on user feedback data using a closed-loop reinforcement learning optimization mechanism.

[0008] Preferably, in one embodiment of the present invention, the local feature extraction model is a TextCNN model, and the context semantic understanding model is a BiLSTM model.

[0009] Preferably, in one embodiment of the present invention, the step of constructing the dual-path intent recognition module includes: specifically optimizing the local feature extraction model and the contextual semantic understanding model based on product ingredient vocabulary and skin symptom description phrases in the beauty field.

[0010] Preferably, in one embodiment of the present invention, the customer layer includes user profile information, which includes stored user skin type and allergy history; the SOP process layer includes a standardized beauty operation process that includes testing, recommendation and warning nodes; the business entity layer includes an ingredient database, a skin type database and / or an allergy case database.

[0011] Preferably, in one embodiment of the present invention, the dynamic beauty knowledge graph supports dynamic binding updates, the dynamic binding updates including the steps of: updating the data of the business entity layer in real time; and updating the relevant nodes of the customer layer and the SOP process layer in response to the update of the business entity layer data.

[0012] Preferably, in one embodiment of the present invention, the step of determining recommended content including personalized product recommendations and / or usage suggestions through a knowledge push algorithm includes: based on the user intent, performing similarity matching in the dynamic beauty knowledge graph to determine the optimal SOP process node in the SOP process layer; calculating the correlation degree between each knowledge item in the dynamic beauty knowledge graph and the SOP process node according to the SOP process node, user historical interaction data, and entity relationships in the dynamic beauty knowledge graph; and pushing each knowledge item as recommended content in descending order of correlation degree.

[0013] Preferably, in one embodiment of the present invention, the step of determining recommended content including personalized product recommendations and / or usage suggestions through a knowledge push algorithm includes: dynamically adjusting the SOP process for recommending to users based on user profile information in the customer layer of the dynamic beauty knowledge graph and in combination with relevant knowledge in the business entity layer.

[0014] Preferably, in one embodiment of the present invention, the method further includes the step of: recommending to the user the SOP process node in the SOP process layer that has the highest matching degree with the application scenario based on the application scenario.

[0015] Preferably, in one embodiment of the present invention, the step of performing compliance detection and risk warning on user operation behavior through a real-time monitoring mechanism based on a rule engine includes: matching user operation behavior data with monitoring rules to detect whether the operation behavior data is abnormal; triggering a warning and pushing correction suggestions to the user in response to the detection of abnormal operation behavior data; and continuing to execute the next process in response to the detection of normal operation behavior data.

[0016] Preferably, in one embodiment of the present invention, the step of determining recommended content including personalized product recommendations and / or usage suggestions through a knowledge push algorithm includes: in response to the detection of anomalies in the operation behavior data, dynamically adjusting the subsequent SOP process for recommending to the user by combining relevant knowledge in the business entity layer of the dynamic beauty knowledge graph.

[0017] Preferably, in one embodiment of the present invention, the step of dynamically adjusting the recommendation strategy and system parameters based on user feedback data using a closed-loop reinforcement learning optimization mechanism includes: acquiring user feedback data, wherein the user feedback data includes user interaction data; and updating the weight parameters of the knowledge push algorithm using a reinforcement learning algorithm based on the user feedback data.

[0018] Furthermore, the intelligent recommendation system for the beauty industry provided according to the second aspect of the present invention includes a memory and a processor. The memory stores computer instructions. The processor is connected to the memory and configured to execute the computer instructions stored in the memory to implement the intelligent recommendation method for the beauty industry provided in any of the above embodiments.

[0019] Furthermore, the computer-readable storage medium provided according to the third aspect of the present invention stores computer instructions. When the computer instructions are executed by a processor, they implement the intelligent recommendation method for the beauty industry provided in any of the above embodiments. Attached Figure Description

[0020] The above-described features and advantages of the present invention will be better understood after reading the following detailed description of embodiments of the present disclosure in conjunction with the accompanying drawings. In the drawings, components are not necessarily drawn to scale, and components having similar related characteristics or features may have the same or similar reference numerals.

[0021] Figure 1 A schematic diagram of an intelligent recommendation system for the beauty industry provided according to some embodiments of the present invention is shown;

[0022] Figure 2 A schematic diagram of an intelligent recommendation system for the beauty industry provided according to some embodiments of the present invention is shown;

[0023] Figure 3 A flowchart of an intelligent recommendation method for the beauty industry, provided according to some embodiments of the present invention, is shown.

[0024] Figure 4 A structural diagram of a dual-path intent recognition module provided according to some embodiments of the present invention is shown;

[0025] Figure 5 A schematic diagram of a dynamic beauty knowledge graph provided according to some embodiments of the present invention is shown;

[0026] Figure 6 A schematic diagram of a real-time monitoring mechanism provided according to some embodiments of the present invention is shown; and

[0027] Figure 7A schematic diagram of a closed-loop reinforcement learning optimization mechanism provided according to some embodiments of the present invention is shown.

[0028] Figure label:

[0029] 100: An intelligent recommendation system for the beauty industry;

[0030] 110: Memory;

[0031] 111: Computer-readable storage medium;

[0032] 120: Processor;

[0033] 200: An intelligent recommendation system for the beauty industry;

[0034] 210: Dynamic Beauty Knowledge Graph Module;

[0035] 220: Dual-channel intent recognition module;

[0036] 230: Real-time monitoring mechanism;

[0037] 240: Closed-loop reinforcement learning optimization module;

[0038] S310~S360: Steps;

[0039] 510: Customer layer;

[0040] 520: SOP process layer; and

[0041] 530: Business Entity Layer. Detailed Implementation

[0042] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. It should be noted that the aspects described below with reference to the accompanying drawings and specific embodiments are merely exemplary and should not be construed as limiting the scope of protection of the present invention in any way.

[0043] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0044] Furthermore, the terms "upper," "lower," "left," "right," "top," "bottom," "horizontal," and "vertical" used in the following description should be understood as the orientations shown in the relevant paragraphs and accompanying drawings. These relative terms are for illustrative purposes only and do not imply that the described apparatus must be manufactured or operated in a specific orientation, and therefore should not be construed as limiting the invention.

[0045] It is understood that although terms such as "first," "second," and "third" may be used herein to describe various components, regions, layers, and / or parts, these components, regions, layers, and / or parts should not be limited by these terms, and these terms are only used to distinguish different components, regions, layers, and / or parts. Therefore, the first components, regions, layers, and / or parts discussed below may be referred to as second components, regions, layers, and / or parts without departing from some embodiments of the present invention.

[0046] As mentioned above, existing static recommendation models not only have low accuracy but may also cause skin problems for users by recommending products containing allergenic ingredients. Regarding user needs analysis, most existing beauty platforms use single intent recognition models. These technologies have an accuracy rate of less than 70% when dealing with vague user descriptions or complex symptom expressions. Especially when dealing with users who are sensitive to ingredients, they cannot accurately identify their need to avoid specific ingredients (such as alcohol or fragrance), leading to a significant deviation between recommended results and the user's actual needs, resulting in a poor user experience and a high platform complaint rate.

[0047] To overcome the aforementioned deficiencies in existing technologies, this invention provides an intelligent recommendation method for the beauty industry, an intelligent recommendation system for the beauty industry, and a computer-readable storage medium. These methods comprehensively address the shortcomings of traditional beauty recommendation systems in allergy prevention and intent recognition, significantly improve the accuracy of recommendations and the safety of recommended beauty products, and enhance user satisfaction.

[0048] In some non-limiting embodiments, the intelligent recommendation method for the beauty industry provided in the first aspect of the present invention can be implemented via the intelligent recommendation system for the beauty industry provided in the second aspect of the present invention.

[0049] Please refer to Figure 1 , Figure 1 A schematic diagram of an intelligent recommendation system for the beauty industry, provided according to some embodiments of the present invention, is shown.

[0050] like Figure 1As shown, the intelligent recommendation system 100 for the beauty industry may be configured with a memory 110 and a processor 120. The memory 110 includes, but is not limited to, the computer-readable storage medium 111 described in the third aspect of the present invention, which stores computer instructions thereon. The processor 120 is connected to the memory 110 and is configured to execute the computer instructions stored in the memory 110 to implement the intelligent recommendation method for the beauty industry provided in the first aspect of the present invention.

[0051] In a preferred embodiment, the intelligent recommendation system for the beauty industry provided in the second aspect of the present invention may include multiple program modules. The program modules may be stored in memory. The program modules include, but are not limited to, an operating system, one or more applications, other program modules, and program data; each or some combination of these examples may include an implementation of a network environment. The program modules typically perform the functions and / or methods described in the embodiments of the present invention.

[0052] Please refer to Figure 2 , Figure 2 A schematic diagram of an intelligent recommendation system for the beauty industry, provided according to some embodiments of the present invention, is shown.

[0053] like Figure 2 As shown, the intelligent recommendation system 200 for the beauty industry may include a dynamic beauty knowledge graph module 210, a dual-path intent recognition module 220, a real-time monitoring mechanism 230, and a closed-loop reinforcement learning optimization module 240.

[0054] The dynamic beauty knowledge graph module 210 is used to construct a beauty knowledge graph and dynamically update each node in the knowledge graph based on real-time business data. The dynamic beauty knowledge graph module 210 can have a three-layer architecture, including a customer layer, an SOP (Standard Operating Procedure) process layer, and a business entity layer. Here, SOP refers to standardized beauty process nodes such as beauty product recommendations, usage guidance, and allergy treatment, including specific operational steps such as testing, recommendation, and alerts.

[0055] The dual-path intent recognition module 220 can identify user intent based on user query data. The dual-path intent recognition module 220 may include a local feature extraction model, a contextual semantic understanding model, and a graph compensation mechanism. When the user query is ambiguous, the graph compensation mechanism is triggered, combining with the dynamic beauty knowledge graph module 210 to achieve accurate identification of the user intent.

[0056] The real-time monitoring mechanism 230 can monitor the user's SOP process in real time and detect any anomalies based on the user's actions. If an anomaly is detected, an early warning mechanism is triggered in a timely manner, and corrective suggestions are pushed to the user. Violations detected by the real-time monitoring module 230 during monitoring can be used to feed back into the dynamic beauty knowledge graph module 210, updating the test node parameters in the graph and thus improving the recommendation process.

[0057] The closed-loop reinforcement learning framework 240 can dynamically adjust recommendation strategies and system parameters based on user feedback data, thereby sending more accurate recommended content to users.

[0058] The following will first describe the working principle of the aforementioned intelligent recommendation system for the beauty industry using examples of intelligent recommendation methods for the beauty industry. Those skilled in the art will understand that these examples of intelligent recommendation methods for the beauty industry are merely non-limiting implementations provided by this invention, intended to clearly demonstrate the main concepts of the invention and provide specific solutions convenient for public implementation, rather than limiting all functions or working methods of the intelligent recommendation system for the beauty industry. Similarly, the intelligent recommendation system for the beauty industry is also only one non-limiting implementation provided by this invention, and does not limit the executing entities and execution order of the steps in these intelligent recommendation methods for the beauty industry.

[0059] In some embodiments, the intelligent recommendation system for the beauty industry (hereinafter referred to as the intelligent recommendation system) provided by the present invention can be configured in or connected to shopping platforms, live streaming platforms or offline shopping guide platforms.

[0060] Please refer to Figure 3 , Figure 3 A flowchart of an intelligent recommendation method for the beauty industry, provided according to some embodiments of the present invention, is shown.

[0061] like Figure 3 As shown, the intelligent recommendation method 300 for the beauty industry may include step S310: obtaining user query data.

[0062] Intelligent recommendation systems can directly obtain user query data, or obtain user query data based on the platforms they connect to.

[0063] User queries can be direct and precise, such as "What are the effects of niacinamide?"; or they can be indirect and vague, such as "My face is very dry the next day after applying a face mask at night." User queries can be text-based input or data converted from voice input.

[0064] The intelligent recommendation method 300 for the beauty industry may include step S320: based on user query data, obtain user intent through a dual-path intent recognition module, the dual-path intent recognition module including a local feature extraction model for extracting keywords and a contextual semantic understanding model for understanding the usage scenario context.

[0065] Please refer to Figure 4 , Figure 4 A structural diagram of a dual-path intent recognition module provided according to some embodiments of the present invention is shown.

[0066] like Figure 4 As shown, in some embodiments, the dual-path intent recognition module recognizes user intent through the collaboration of two models.

[0067] Specifically, the dual-channel intent recognition module can first perform word segmentation on the user query data to obtain word vectors. Here, the dual-channel intent recognition module can use Word2Vec (a word vector conversion technology) to perform word segmentation on the user query data to obtain word vectors.

[0068] Subsequently, the dual-path intent recognition module can extract local and contextual features from the user's query data using a local feature extraction model and a contextual semantic understanding model, respectively. The local feature extraction model captures key beauty phrases such as ingredients and symptoms, while the contextual semantic understanding model fully understands descriptions such as usage scenarios. For example, when a user inputs "whitening essence containing niacinamide," the local feature extraction model can accurately extract ingredient-related words such as "niacinamide" and "whitening essence." Or, when a user describes "I used face cream at night, and my face is very dry in the morning," the local feature extraction model can extract ingredient and symptom-related words such as "face cream" and "dry face," while the contextual semantic understanding model can analyze the user's dissatisfaction with the face cream's moisturizing effect and the fact that it was used at night.

[0069] Preferably, the local feature extraction model can be a TextCNN (Text Convolutional Neural Network) model, which extracts key information through convolutional kernels. The contextual semantic understanding model can be a Bidirectional Long Short-Term Memory (BiLTSM) model, which can capture the dependencies between different parts of a time series.

[0070] Even better, when training the dual-path intent recognition model, the intelligent recommendation system can specifically optimize the local feature extraction model and the contextual semantic understanding model for product ingredient terms such as "niacinamide" and "salicylic acid" and skin symptom description phrases such as "redness, swelling and stinging", thereby better adapting to feature extraction in the beauty industry.

[0071] Then, the extracted local features and contextual features are concatenated and attention-weighted calculations are performed. Based on the weighted features, the intent recognition confidence score is calculated. When the intent recognition confidence score is greater than or equal to a preset threshold, the user intent obtained by the dual-path intent recognition module can be considered relatively accurate, and the user intent is output.

[0072] When the confidence level of intent recognition is less than a preset threshold, it indicates that the dual-model approach cannot accurately identify the user's specific intent. In this case, the intelligent recommendation system can trigger the graph compensation mechanism in the dual-path intent recognition module to obtain the corrected user intent.

[0073] Please continue to refer to this. Figure 3 The intelligent recommendation method 300 for the beauty industry may include step S330: in response to the intent recognition confidence being less than a preset threshold, a graph compensation mechanism is triggered. The graph compensation mechanism queries related information in the dynamic beauty knowledge graph based on the feature entities extracted by the dual-path intent recognition module to correct the user intent.

[0074] In some embodiments, the graph compensation mechanism can be configured in the dual-path intent recognition module, and the user intent can be corrected based on the feature entities extracted by the dual models of the dual-path intent recognition module, combined with the dynamic beauty knowledge graph.

[0075] For example, in some embodiments, after a user inputs "My skin feels a little uncomfortable after using this serum," the dual-path intent recognition module cannot accurately identify the user's specific intent; that is, the intent recognition confidence level is less than a preset threshold. The intelligent recommendation system can trigger a graph compensation mechanism. Based on the feature entity "serum" extracted by the dual models of the dual-path intent recognition module, it queries related information in the dynamic beauty knowledge graph, such as querying the ingredient list of "serum." Based on the query results, the intelligent recommendation system can match and select the user intent with the highest confidence level to cover the original output of the dual-path intent recognition module, thereby determining the corrected user intent.

[0076] Through the graph compensation mechanism, the intelligent recommendation system can extract entities by querying the dynamic beauty knowledge graph to determine the associated intent when the confidence of intent recognition is insufficient, thus ensuring accurate responses even when users input low-confidence query data.

[0077] Here, before implementing the intelligent recommendation method for the beauty industry provided by the present invention, a process for constructing a dynamic beauty knowledge graph is also included.

[0078] A dynamic beauty knowledge graph can have a three-layer architecture, including a customer layer, a SOP process layer, and a business entity layer. Furthermore, a dynamic beauty knowledge graph can support node parameterization and real-time updates.

[0079] Please refer to Figure 5 , Figure 5 A schematic diagram of a dynamic beauty knowledge graph provided according to some embodiments of the present invention is shown.

[0080] like Figure 5 As shown, the customer layer 510 can be used to store user profile information, including skin type (sensitive skin, acne-prone skin, dry skin, oily skin, etc.), allergy history, age, gender, and other basic characteristics. Figure 5 In the embodiment shown, the dynamic beauty knowledge graph constructs user profile nodes in the customer layer 510, classifying users with similar characteristics, such as dividing users with sensitive skin who are allergic to alcohol into a subgraph.

[0081] Here, the intelligent recommendation system can form user profile information through user registration information and user's historical purchase records. For example, based on the user's registration information, it can filter out users with sensitive skin and mark their allergy to alcohol.

[0082] SOP process layer 520 may include standardized beauty operation procedures that include testing, recommendation, and alert nodes. An intelligent recommendation system can define standardized beauty operation procedure nodes within SOP process layer 520 to cover aspects such as skin testing, product recommendations, usage instructions, and allergy alerts.

[0083] In SOP process layer 520, the SOP process is a directed graph structure, consisting of nodes and edges. Each node in the SOP process represents a step, and each edge represents a transition condition and operational specification.

[0084] Intelligent recommendation systems can be configured with standardized workflow logic for transitions between different steps in a beauty routine. For example... Figure 5 As shown, under the skin test node, if the user's skin test is passed, the intelligent recommendation system can proceed to the product recommendation node in SOP process layer 520 through the process engine. If the test is failed, it will proceed to the allergy warning node in SOP process layer 520. Taking the skincare solution recommendation on a beauty brand's official website as an example, after the user completes the skin type test, the intelligent recommendation system can directly push a matching product recommendation page based on the test results. If the test results show that the user has sensitive skin and is allergic to certain ingredients, an allergy warning will be triggered, reminding the user to choose products with caution.

[0085] The business entity layer 530 can connect to external business systems to obtain real-time information on various business entities within those systems. Each business entity may include an ingredient database, a skin type database, and / or an allergy case database. Figure 5 In the illustrated embodiment, the business entity layer 530 may further include a product efficacy library and a pairing taboo library.

[0086] The ingredient database can be used to collect various beauty ingredients (such as niacinamide, salicylic acid, alcohol, etc.), ingredient characteristics (efficacy, safety, etc.), and related safety regulations (such as lists of prohibited ingredients, concentration limits, etc.). The skin type database can be used to describe in detail the characteristics of each skin type, suitable product types, and key care points. The allergy case database can be used to record allergy reaction cases experienced by users of different skin types due to the use of products with specific ingredients, including allergy symptoms, usage scenarios, and other information.

[0087] The dynamic beauty knowledge graph supports node parameterization. Intelligent recommendation systems can parameterize the nodes within the dynamic beauty knowledge graph, including setting operational specifications for each node, configuring input and output parameters, setting timeout thresholds, and defining dynamic constraints. By finely defining the parameters of each node in the dynamic beauty knowledge graph, the efficiency and accuracy of the intelligent recommendation system at each node can be ensured.

[0088] Setting operational guidelines for each node can clarify the specific instructions to be executed at each node. For example, the operational guidelines for the skin testing node could be "conduct a 28-day patch test and observe the skin reaction." Setting input and output parameters for each node can specify the content and format of data transmission between nodes. For example, the skin testing node outputs the user's skin type (sensitive skin, dry skin, etc.) as input to the product recommendation node, and the product recommendation node outputs a list of recommended products. Setting timeout thresholds for each node can determine the maximum execution time for node operations. For example, the allergy reaction observation period can be set to 72 hours; if the user does not have an abnormal reaction within this period, the process continues. Defining dynamic constraints for each node can constrain node operations based on real-time business rules. For example, "products containing parabens are prohibited from being recommended to users with sensitive skin," which allows the product recommendation node to dynamically filter recommended products based on the user's skin type and product ingredients.

[0089] A dynamic beauty knowledge graph supports real-time data updates. For example, based on the connection between the business entity layer and external business systems, the intelligent recommendation system can extract safety specification data of relevant ingredients from reports published by beauty industry research institutions to update the ingredient database; it can also collect allergy cases from after-sales feedback of beauty brands to expand the allergy case database; or it can adjust sensitive ingredients according to the season, such as dynamically adjusting the detection weight of pollen-related ingredients in winter and spring. Therefore, a dynamic beauty knowledge graph can support real-time updates of data such as new ingredients and new skin types, ensuring that the recommendation logic of the intelligent recommendation system is synchronized with the latest beauty research results.

[0090] Preferably, the nodes of the dynamic beauty knowledge graph are dynamically linked and support dynamic binding updates. The intelligent recommendation system can monitor data updates in the business entity layer in real time using the developed data monitoring module, and update relevant nodes in the customer layer and SOP process layer based on these updates to achieve a dynamic binding update mechanism. For example, when allergen risk information for a certain ingredient is added to the ingredient database of the business entity layer, it immediately triggers parameter updates for the relevant linked nodes in the customer layer and SOP process layer.

[0091] Based on the dynamic beauty knowledge graph constructed above, the intelligent recommendation system can accurately match each user's skin type and safe ingredients. When a user's query is ambiguous, the dynamic beauty knowledge graph can provide compensation for the graph supplementation mechanism. Based on the product name, ingredients, and other entity information extracted from the graph by the feature entities, it assists the dual-model of the dual-path intent recognition module to complete accurate intent recognition. The dual-path intent recognition module combined with the graph compensation mechanism achieves high-accuracy intent recognition. Among them, the accuracy rate of allergy problem recognition reaches 94.6% (the accuracy rate of recognition of ambiguous descriptions is improved by 28%). Furthermore, the dual-path intent recognition module combined with the graph compensation mechanism can effectively handle ambiguous descriptions and complex scenarios, improving the user experience.

[0092] like Figure 3 As shown, the intelligent recommendation method 300 for the beauty industry may include step S340: generating response content based on user intent and combining a dynamic beauty knowledge graph. The response content includes personalized product recommendations and / or usage suggestions determined by a knowledge push algorithm.

[0093] Intelligent recommendation systems can directly query dynamic beauty knowledge graphs based on determined user intent and generate answers to respond to users.

[0094] The response can also include recommendations. The intelligent recommendation system can determine the recommended content based on user intent obtained from the dual-path intent recognition module, combined with a dynamic beauty knowledge graph, and through a knowledge-based push algorithm. Recommended content can include personalized product recommendations and / or usage suggestions.

[0095] Here, the knowledge recommendation algorithm can determine and recommend content from the dynamic beauty knowledge graph that is highly relevant to the user's intent based on the user's intent.

[0096] Preferably, the intelligent recommendation system can first perform similarity matching in the dynamic beauty knowledge graph based on user intent, and determine the SOP process nodes with high similarity in the SOP process layer of the dynamic beauty knowledge graph as the optimal SOP process nodes. Then, based on the SOP process node, user historical interaction data, and entity relationships in the knowledge graph, the system calculates the correlation between each knowledge item in the dynamic beauty knowledge graph and the SOP process node, and dynamically pushes the most relevant knowledge content.

[0097] In some embodiments, the knowledge recommendation algorithm can be an improved PersonalRank algorithm. The improved PersonalRank algorithm determines the correlation between a knowledge item and a SOP process node based on the shortest path distance between the knowledge item and the SOP process node in the knowledge graph, the relationship weight between the knowledge item and the SOP process node, and the user's preference for the knowledge item. Here, the user's preference for the knowledge item can be determined based on the user's historical interaction data.

[0098] In some embodiments, based on the correlation between the determined knowledge items and SOP process nodes, knowledge items are pushed as recommended content by the knowledge push algorithm in descending order of correlation. Each knowledge item may include a title, summary, and detailed link.

[0099] Existing beauty recommendation systems have fixed recommendation logic that cannot be dynamically adjusted, and they completely ignore potential allergy risks and changes in skin condition when making recommendations to users.

[0100] The intelligent recommendation system provided by this invention can dynamically adjust the SOP process for recommending products to users based on user profile information in the customer layer and relevant knowledge in the business entity layer. For example, for users with sensitive skin, the intelligent recommendation system can skip the alcohol and fragrance ingredient testing nodes; for users with acne-prone skin, it can add an oil-control ingredient compatibility testing node; and for users with mature skin, it can add an anti-wrinkle ingredient tolerance assessment node.

[0101] Intelligent recommendation systems can also dynamically adjust the Standard Operating Procedures (SOPs) for recommending products to users based on changes in user profiles at the customer level and relevant knowledge at the business entity level. For example, after obtaining the latest skin condition of users through the regular skin self-testing function on the official website of a beauty brand, and responding to a change in skin type from oily to sensitive, the intelligent recommendation system can automatically adjust the product recommendation and usage guidance process, skipping the process node that recommends products containing alcohol.

[0102] Thus, the intelligent recommendation system provided by this invention can accurately match users' skin types and safe ingredients based on a multi-dimensional dynamic beauty knowledge graph, achieving an allergy prevention accuracy rate of 96.8%. Furthermore, the intelligent recommendation system dynamically adjusts the SOP process for recommending products to users based on the dynamic beauty knowledge graph, which can effectively address different user skin types and changes in user skin type.

[0103] Based on a multi-dimensional dynamic beauty knowledge graph, the intelligent recommendation system provided by this invention achieves an ingredient safety retrieval speed of 1.2 seconds (compared to 3.2 seconds for traditional solutions), an allergy prevention accuracy rate of 96.8% (compared to only 78% for manual recommendations), and a personalized recommendation matching rate of 89% (compared to an industry average of 65%). Based on a dual-path intent recognition module and a beauty knowledge graph compensation mechanism, the intelligent recommendation system provided by this invention achieves an allergy problem identification accuracy rate of 94.6%, improves intent recognition accuracy for ambiguous descriptions by 28%, and achieves a recall rate of 91% for the model in the dual-path intent recognition module (compared to an industry average of 72%). The speed of pushing beauty emergency solutions based on the dual-path intent recognition module is 2.3 seconds.

[0104] The intelligent recommendation system can include recommended Standard Operating Procedures (SOPs) and SOP steps in its responses. Users can then operate based on these recommended SOPs and steps. For example, when a user searches for information due to allergies, the intelligent recommendation system can recommend relevant soothing and repair products and guide the user through the recommended usage process. This usage process can include a normal usage sequence, a soothing sequence, and a repair sequence. The intelligent recommendation system, considering the severity of the user's allergy, recommends starting points in the repair sequence to guide the user in applying a thick layer of the product. Preferably, the intelligent recommendation system can visualize the entire process to better guide the user in following the SOP.

[0105] In some embodiments, the intelligent recommendation system can also recommend SOP processes and SOP process nodes in the SOP process layer that best match the application scenario to users based on the application scenario. For example, when the application scenario is live streaming, the intelligent recommendation system can recommend a live streaming-specific SOP process to users whose role is "host," which can guide the host to introduce products to the audience.

[0106] However, the monitoring of the safety of beauty product use has long been weak. Traditional beauty platforms often only respond passively after users report allergic reactions, lacking real-time monitoring and proactive early warning mechanisms. Existing technologies have introduced simple monitoring systems, but the rule matching speed is slow (average response time exceeds 1 second), and the early warning suggestions lack personalization. For example, for users with sensitive skin who use products containing alcohol—a high-risk behavior—they only give a vague warning of "possible allergy," without providing precise alternatives based on the user's specific skin condition, thus failing to effectively prevent allergic events.

[0107] Please continue to refer to this. Figure 3 The intelligent recommendation method 300 used in the beauty industry may include step S350: performing compliance detection and risk warning on user operation behavior through a real-time monitoring mechanism based on a rule engine.

[0108] The intelligent recommendation system provided by this invention can monitor the user's actions during the execution of the Standard Operating Procedure (SOP) process in real time, thereby monitoring the user's use of beauty products. In some embodiments, the real-time monitoring of the SOP process can be performed by... Figure 2 The real-time monitoring mechanism in 230 is implemented.

[0109] Please refer to Figure 6 , Figure 6 A schematic diagram of a real-time monitoring mechanism provided according to some embodiments of the present invention is shown.

[0110] Intelligent recommendation systems can monitor user behavior in real time. Then, they use monitoring rules to detect whether the user's behavior is abnormal. For example... Figure 6 As shown, the intelligent recommendation system can use the Drools rule engine as a monitoring engine. This monitoring engine is used to match user operation data with monitoring rules in the rule base to detect whether the user operation data is abnormal.

[0111] Using the Drools rule engine as a monitoring engine and matching user operation data with monitoring rules in the rule base can reduce the response time of intelligent recommendation systems to user operation data to within 80ms.

[0112] In some embodiments, the intelligent recommendation system can transform beauty industry standards such as the "Cosmetic Safety Technical Specifications" into executable rules, and can also transform business requirements into monitoring rules to monitor user actions at the millisecond level. These executable rules can include multiple monitoring dimensions covering ingredient safety, usage compliance, symptom severity, and / or the rationality of combinations, forming a multi-dimensional monitoring system.

[0113] For example, the rules for ingredient safety include "warning when the product contains allergenic ingredients and the user has sensitive skin," the rules for usage compliance include "reminding the user of daily use exceeding the recommended frequency," and the rules for symptom severity include "discontinuing use if redness / burning sensation occurs."

[0114] When users perform beauty-related actions, the intelligent recommendation system monitors operation parameters in real time through its monitoring engine and matches them with a rule base. For example, if a user adds a product containing alcohol to their shopping cart, and the user profile information indicates that the user has sensitive skin, the intelligent recommendation system will match a warning rule based on the ingredient safety dimension: "Warning: Contains allergenic ingredients and the user has sensitive skin."

[0115] In response to detected abnormal user behavior, an alert is triggered and corrective suggestions are pushed to the user. The intelligent recommendation system can dynamically generate targeted and personalized suggestions by combining user profile information such as skin type and usage history. For example, in the embodiment where the user adds alcohol-containing products to their shopping cart and the user has sensitive skin, the intelligent recommendation system can provide the user with the corrective suggestion: "Sensitive skin users should use alcohol-containing products with caution; it is recommended to change products."

[0116] For example, when user activity data shows that a user is simultaneously using a product containing Vitamin C and a product containing acids, the system can match this to the warning rule "contains conflicting ingredients" under the compatibility dimension of the product combination. Based on this detected anomaly in the activity data, the intelligent recommendation system can trigger a warning and push corrective suggestions to the user: "The current combination may irritate the skin; adjustment is recommended."

[0117] Preferably, the intelligent recommendation system can dynamically adjust the node parameters of each SOP process in the dynamic beauty knowledge graph based on abnormal operation behavior data during real-time monitoring, thereby improving the SOP process.

[0118] Furthermore, the intelligent recommendation system can adjust the subsequent Standard Operating Procedures (SOPs) recommended to users based on abnormal operation data. For example, in some embodiments, when user operation data shows that the user has redness and swelling, it can be matched to the warning rule "redness / burning sensation occurs" in the symptom severity dimension. Based on the abnormal operation data, the intelligent recommendation system can trigger an alert and push a corrective suggestion to the user: "Immediately stop using and perform soothing and repair." In this case, the intelligent recommendation system can dynamically adjust the content of the SOP recommended to the user based on the abnormal operation data and relevant knowledge in the business entity layer of the knowledge graph. For example, it can suspend recommending alcohol-containing products to the user and instead recommend alternative products suitable for sensitive skin that can soothe and repair.

[0119] like Figure 6 As shown, in response to normal operation data detection, the intelligent recommendation system can continue to execute the next process according to the currently recommended SOP process.

[0120] Therefore, the real-time monitoring mechanism based on the rule engine in the intelligent recommendation system matches monitoring rules with the Drools rule engine and the built rule base to determine whether the response time is normal (less than 80ms) and the deviation rate is 3.8%, significantly better than the industry average deviation rate of 37%. Furthermore, because the intelligent recommendation system monitors user operations in real time, it can promptly detect violations or risky behaviors and alert users to take preventative measures. The end-to-end processing time is 22 minutes (compared to 40 minutes for traditional methods), and the user complaint rate is reduced by 68%.

[0121] In current technologies, the optimization of beauty recommendation systems relies on manual parameter tuning, with iteration cycles lasting several weeks. Optimizing existing recommendation algorithms requires manually collecting user feedback and analyzing data features to adjust algorithm parameters, a process that takes approximately 14 days. This inefficient optimization method struggles to adapt to the rapidly changing market environment and user needs in the beauty industry, leading to stagnant recommendation performance, low user retention rates, and limited platform competitiveness. Faced with the continuous emergence of new ingredients and skin types, the drawbacks of traditional optimization mechanisms are becoming increasingly apparent, severely hindering the intelligent development of the beauty industry.

[0122] like Figure 3 As shown, the intelligent recommendation method 300 for the beauty industry may include step S360: dynamically adjusting the recommendation strategy and system parameters based on user feedback data using a closed-loop reinforcement learning optimization mechanism.

[0123] Furthermore, the intelligent recommendation system provided by this invention can also update the recommendation strategy weight parameters of the knowledge push algorithm and the parameters of the intelligent recommendation system in real time based on user feedback data through a closed-loop reinforcement learning optimization mechanism, thereby accelerating the iteration of the intelligent recommendation system, improving user satisfaction with the pushed content, and increasing user retention. In some embodiments, the closed-loop reinforcement learning optimization mechanism can be... Figure 2 The closed-loop reinforcement learning module 240 is implemented.

[0124] Please refer to Figure 7 , Figure 7 A schematic diagram of a closed-loop reinforcement learning optimization mechanism provided according to some embodiments of the present invention is shown.

[0125] like Figure 7As shown, intelligent recommendation systems can acquire user feedback data, which can include user interaction data. User feedback data can include multi-source feedback data such as knowledge click-through rate, SOP dwell time, user ratings, collection behavior, and / or repurchase behavior. For example, an intelligent recommendation system can use the number of times a user clicks on a recommended product knowledge card as knowledge click-through rate data, the time a user spends on an SOP process page (such as a product usage guide page) as SOP dwell time data, and the user's rating of the recommended products and services as user rating data.

[0126] Intelligent recommendation systems can design corresponding reward signal weights in the reward function of reinforcement learning algorithms based on these user feedback data. For example, the reward signal weight for knowledge click rate can be 35, the reward signal weight for SOP dwell time can be 25, the reward signal weight for user rating can be 20, the reward signal weight for collection behavior can be 10, and the reward signal weight for repurchase behavior can be 10.

[0127] Intelligent recommendation systems can update the recommendation strategy weights of knowledge push algorithms based on user feedback data and reinforcement learning algorithms. Figure 7 In the illustrated embodiment, the intelligent recommendation system can update different weights in the knowledge push algorithm based on different dimensions of user feedback data. Here, the knowledge push algorithm may include weight parameters such as α weight, β weight, γ weight, δ weight, and ε weight, which can be used to represent efficacy factors, combination weights, user preferences, ingredient suitability, and scenario suitability, etc.

[0128] In some embodiments, the intelligent recommendation system can also dynamically adjust system parameters based on user feedback data. For example, the intelligent recommendation system can adjust the dynamic beauty knowledge graph based on user feedback data. In some embodiments, the intelligent recommendation system can adjust the confidence threshold for SOP process matching in the dynamic beauty knowledge graph based on user feedback data, lowering the confidence threshold from 0.6 to 0.58 to adapt to user needs and market changes, thereby improving recommendation accuracy and the performance of the intelligent recommendation system. When beauty brands promote new products, the intelligent recommendation system can quickly adjust its recommendation strategy through a closed-loop reinforcement learning optimization mechanism, accurately pushing new products to potential users, improving the effectiveness of new product promotion and user purchase conversion rates.

[0129] In some embodiments, the intelligent recommendation system can also optimize the trigger threshold of the monitoring rules used to detect abnormal data in the real-time monitoring mechanism based on user feedback data.

[0130] Furthermore, the intelligent recommendation system can adjust the seasonal weight parameters in the dynamic beauty knowledge graph based on the seasonal characteristics of the beauty industry. For example, it can adjust the weight of moisturizing needs in winter and the weight of sun protection needs in summer.

[0131] Thus, the intelligent recommendation system provided by this invention can improve the click-through rate of recommended content by users (click-through rate of 52%, baseline of 32%) through a closed-loop reinforcement learning optimization mechanism, improve the iteration speed of the intelligent recommendation system (the iteration speed of multiple models in the intelligent recommendation system is 22 hours, while the existing manual parameter tuning requires 1 week), increase the user retention rate by 47%, and optimize the overall average response time to 1.8 seconds.

[0132] In summary, this invention constructs a complete intelligent beauty system covering recommendation, parsing, monitoring, and optimization, comprehensively overcoming the shortcomings of existing technologies and enabling more accurate, safe, and efficient intelligent beauty recommendations. By constructing a multi-dimensional dynamic knowledge graph, employing a dual-path intent recognition and compensation mechanism in the dual-path intent recognition module, a real-time monitoring mechanism, and a closed-loop reinforcement learning optimization mechanism, it comprehensively solves the deficiencies of traditional beauty recommendation systems in allergy prevention, intent recognition, real-time monitoring, and system optimization, significantly improving user safety when using beauty products, the accuracy of recommended beauty products, and user satisfaction.

[0133] The following are several non-limiting preferred embodiments, which are used to illustrate the use and effects of the intelligent recommendation system for the beauty industry proposed in this invention.

[0134] In Example 1, the user inputs "My skin feels a little uncomfortable after using this serum" as the user query data. After the intelligent recommendation system obtains the user's intent through the dual-path intent recognition module, it activates the graph compensation mechanism because the confidence level of the intent recognition is lower than the preset threshold of 0.65. The graph compensation mechanism queries the dynamic beauty knowledge graph based on the feature entity "serum" to obtain the association results of the feature entity "serum", such as the ingredient list of "serum", and finds that the ingredient list contains ingredients such as niacinamide and alcohol. The intelligent recommendation system further identifies the user's intent based on the query association results, selects the highest confidence associated intent, and overwrites the original output of the dual-path intent recognition module. Then, based on the selected highest confidence associated intent, through the knowledge push algorithm, and combined with the user profile information that the user is allergic to alcohol, it directly replies to the user, "You are allergic to alcohol, it is recommended to choose an alcohol-free serum product." In this way, the graph compensation mechanism can accurately answer user questions and effectively deal with the situation in beauty consultation scenarios where users have many vague inquiries about product ingredients and effects.

[0135] Preferably, the intelligent recommendation system can further determine the content that can be recommended to the user based on the user's profile information and relevant knowledge in the dynamic beauty knowledge graph, such as recommending SOP procedures or products related to "soothing and repair".

[0136] Example 2 illustrates the application of an intelligent recommendation system in a beauty e-commerce livestreaming scenario. In Example 2, the intelligent recommendation system can collect viewer profiles in advance on the beauty e-commerce livestreaming platform, including information such as skin type and historical purchasing behavior. It then integrates the skin type and allergy history data of registered users into the customer layer of a dynamic beauty knowledge graph.

[0137] The intelligent recommendation system can add a live-stream-specific Standard Operating Procedure (SOP) to the SOP (Standard Operating Procedure) layer of the dynamic beauty knowledge graph, providing it to users acting as "hosts." Hosts can query the live-stream process, and the intelligent recommendation system can recommend live-stream-specific SOPs to them. Alternatively, the intelligent recommendation system can directly recommend live-stream-specific SOPs to users based on the detected live-stream scenario. Hosts then recommend products according to the nodes of this live-stream-specific SOP. For example, after the live stream begins, viewers can be guided to perform a skin type test; based on the viewer's skin type, the next product recommendation node in the SOP process retrieves suitable product information from the dynamic beauty knowledge graph in real time. In response to a viewer's comment indicating dry skin, the intelligent recommendation system retrieves suitable product recommendation nodes for dry skin from the SOP layer and pushes corresponding product information to the host.

[0138] Within the business entity layer of the dynamic beauty knowledge graph, relevant product knowledge can be added for live streaming. During live streams, the intelligent recommendation system can retrieve relevant information from the business entity layer in real time to assist the host in product introductions. For example, when a host explains a face cream, the intelligent recommendation system can retrieve detailed information (efficacy, safety, etc.) about the niacinamide ingredient in that cream from the ingredient database, helping the host accurately highlight the product's ingredient advantages.

[0139] During live streaming, the intelligent recommendation system can acquire real-time query data from viewers in the live stream, such as inquiries sent in the chat. In some examples, the intelligent recommendation system can retrieve a viewer's query, "Is this serum suitable for sensitive skin?", as user query data. The dual-path intent recognition module can capture keywords such as "serum" and "sensitive skin" through a local feature extraction model, and understand whether the semantics are related to inquiries about the safety of product ingredients through a contextual semantic understanding model.

[0140] When the confidence level of the user intent output by the dual-channel intent recognition module is lower than the threshold, i.e., the user's query data is relatively ambiguous, the intelligent recommendation system can activate the graph compensation mechanism. The intelligent recommendation system can query related ingredients in the dynamic beauty knowledge graph based on the feature entity "essence," and combine this with the feature entity "sensitive skin" to obtain compatibility information between related ingredients and sensitive skin types, thereby determining the user's accurate intent and providing precise answers to viewers.

[0141] In addition, the intelligent recommendation system can set monitoring rules based on beauty industry standards and the rules of the live streaming platform. According to the set monitoring rules, it monitors the Standard Operating Procedure (SOP) process in real time through a monitoring mechanism. For example, it can monitor the live streaming-specific SOP process in real time to check whether the products recommended by the host meet the requirements of monitoring rules such as ingredient safety. In some examples, if a host recommends a product containing a prohibited ingredient, the intelligent recommendation system can trigger an alert, promptly reminding the host to correct the recommendation.

[0142] Meanwhile, the intelligent recommendation system can continuously collect audience feedback data and adjust recommendation strategies and system parameters based on audience behavior data such as click-through rates and viewing time during the live stream. For example, if a particular product has a high click-through rate in a live stream, its corresponding knowledge push weight can be appropriately increased to optimize the recommendation strategy for subsequent live streams.

[0143] Example 3 illustrates the application of an intelligent recommendation system in the intelligent shopping guide system of a beauty brand's offline stores. In Example 3, the intelligent recommendation system can connect to a customer relationship management system via in-store shopping guide devices, collecting customer information and integrating it into the customer layer of a dynamic beauty knowledge graph. When a customer enters the store, customer layer data such as skin type and allergy history are synchronized through membership information. For example, when a customer enters the store, the shopping guide tablet equipped with the intelligent recommendation system can automatically retrieve information from the knowledge graph, such as the customer having sensitive skin and being allergic to alcohol.

[0144] The intelligent recommendation system can add a designed sales guide SOP process to the SOP process layer of the dynamic beauty knowledge graph to assist users acting as "sales guides." Sales guides can provide services to customers according to the SOP process. For example, a sales guide can first recommend a skin type test to the customer; based on the results, in the next product recommendation node of the SOP process, a matching product recommendation plan can be retrieved from the dynamic beauty knowledge graph. If the test shows that the customer has oily skin, the sales guide can retrieve skincare set recommendations suitable for oily skin from the graph according to the SOP process.

[0145] Within the business entity layer of the dynamic beauty knowledge graph, data in the ingredient database and allergy case database can be enriched based on the sales guide scenario, and monitoring rules related to allergy risks can be formulated based on the allergy case database. When sales guides execute their SOP (Standard Operating Procedure) processes, the intelligent recommendation system can monitor their actions. When a sales guide is about to recommend a product containing a common allergen, the intelligent recommendation system can trigger an alert and retrieve information on allergy cases previously caused by that allergen from the allergy case database, reminding the sales guide to recommend the product with caution.

[0146] In addition, the intelligent recommendation system can handle customer inquiries. In some embodiments, customers input their questions via an interactive screen, such as "Can this face mask fade acne scars?" The intelligent recommendation system can extract keywords such as "face mask" and "fade acne scars" using the TextCNN model in the dual-path intent recognition module, and analyze whether the user's query data is an inquiry about product efficacy using the BiLSTM model. If the intent recognition confidence is lower than a threshold, the intelligent recommendation system can also use a graph compensation mechanism to query related efficacy information in the dynamic beauty knowledge graph based on the extracted feature entity "face mask" to further determine the user's intent. The more accurate user intent is then used to determine the response content, thereby making the responses provided by the intelligent recommendation system to customers more accurate.

[0147] In the context of offline stores, intelligent recommendation systems can be configured with alert rules, including product display guidelines and the removal of expired products. For example, based on the standard operating procedure (SOP) for displaying expired products, the intelligent recommendation system can use a monitoring engine to monitor in real time whether expired products are present in the store's product display area. If expired products are found still on display, an alert is triggered, reminding store staff to handle the situation promptly.

[0148] Intelligent recommendation systems can also collect customer feedback data to optimize and iterate subsequent sales recommendations. Based on customer purchasing behavior in-store and feedback on recommended products, intelligent recommendation systems can update recommendation strategies. For example, if a particular sunscreen is selling well in a store, the intelligent recommendation system can increase the weight of its corresponding knowledge posts and optimize subsequent sales recommendations.

[0149] Example 4 illustrates the application of an intelligent recommendation system in a beauty industry online community. In Example 4, the intelligent recommendation system collects information such as skin type and makeup preferences filled in by community users during registration and integrates it into the customer layer of a dynamic beauty knowledge graph. For example, if a user registers as having acne-prone skin and prefers acne treatment products, the intelligent recommendation system can construct a user profile based on this information.

[0150] Intelligent recommendation systems can establish content recommendation SOP processes adapted to online communities within the SOP process layer of a dynamic beauty knowledge graph. For example, the intelligent recommendation system can determine the recommended content to send to users based on data such as user activity times and browsing history. For instance, when a user frequently browses content related to acne treatment serums, the intelligent recommendation system can retrieve relevant acne treatment serum knowledge from the dynamic beauty knowledge graph and push it to the user at the recommended product node of the SOP process. Furthermore, the intelligent recommendation system can push relevant information to users based on keywords used by creators when publishing content on the online community. For example, when a creator is writing an article about niacinamide, the intelligent recommendation system can retrieve detailed information about niacinamide from the ingredient database at the business entity layer for the creator's reference, ensuring the professionalism of the content.

[0151] Furthermore, the intelligent recommendation system can handle user questions and queries in online communities. For example, if a user asks, "What should I do about dry skin after using salicylic acid?", the intelligent recommendation system can collaboratively process this question using a dual-model dual-path intent recognition module. The TextCNN model extracts keywords such as "salicylic acid" and "dry skin," while the BiLSTM model determines whether the user is inquiring about skin problems resulting from the use of the ingredient. If the confidence level of the identified user intent is low, a graph compensation mechanism is triggered. Based on the extracted feature entity "salicylic acid," related knowledge is queried in the dynamic beauty knowledge graph to determine the accurate user intent. Then, based on the user intent, related skin problem solutions are queried in the dynamic beauty knowledge graph to provide personalized answers to the user.

[0152] Furthermore, considering the application scenario of online communities, intelligent recommendation systems can set monitoring rules related to these communities, such as prohibiting the posting of content containing prohibited elements. When a user posts a message containing recommendations with prohibited elements, the monitoring system triggers an alert, notifying relevant personnel to promptly address the violation.

[0153] Intelligent recommendation systems can also collect feedback data from community users to optimize and iterate subsequent content recommendation strategies. Based on user behavior data such as likes, comments, and shares within the community, the intelligent recommendation system updates the weight of the knowledge push algorithm. For example, if an article about sensitive skin care has high engagement, its corresponding knowledge push weight is increased, optimizing subsequent content recommendation strategies.

[0154] Example 5 illustrates the application of an intelligent recommendation system in the cosmetics R&D process. In Example 5, the intelligent recommendation system can increase the data update frequency of the business entity layer in the dynamic cosmetics knowledge graph, collecting data such as industry trends and competitor ingredient information.

[0155] Furthermore, the intelligent recommendation system can be integrated with the cosmetics R&D process to set up standard operating procedures (SOPs). The R&D team can then follow these SOPs to design formulas. In some embodiments, the R&D team can first query the skin characteristics of the target audience from a knowledge graph, and then select suitable ingredients. For example, when developing new products for mature skin, the R&D team can obtain information on the characteristics of mature skin and suitable anti-aging ingredients from the knowledge graph, and then design the formula based on the relevant knowledge retrieved.

[0156] The intelligent recommendation system can handle queries from R&D personnel. In some embodiments, when R&D personnel inquire about the compatibility of a new ingredient with sensitive skin, the intelligent recommendation system extracts keywords such as "new ingredient" and "sensitive skin" through a dual-model approach in the dual-path intent recognition model and understands their semantics. If the confidence level is insufficient, a knowledge graph compensation mechanism is triggered to query related information in the knowledge graph to accurately understand the user's intent. Then, based on the user's intent, a response is generated to provide a reference for the R&D personnel.

[0157] For research and development applications, intelligent recommendation systems can be configured with corresponding monitoring rules. In some embodiments, intelligent recommendation systems can be configured with monitoring rules such as limits on ingredient usage and prohibited ingredients. When an ingredient in the research and development formula is used in excess, the monitoring system triggers an alert, reminding researchers to correct the formula.

[0158] Intelligent recommendation systems can adjust recommendation strategies based on R&D feedback. In some embodiments, intelligent recommendation systems can adjust the weight of knowledge push algorithms and optimize recommendations for future R&D directions based on feedback from new product trials in the market, such as user acceptance of a certain ingredient.

[0159] In summary, the intelligent recommendation method and system for the beauty industry provided by this invention can accurately understand user intent, generate response content by combining dynamic beauty knowledge graphs, improve user query efficiency, and enhance user experience. Furthermore, the intelligent recommendation method and system for the beauty industry can monitor user behavior in real time, improving the security of user operations, and optimize recommendation strategies and system parameters based on user feedback data to adapt to the rapidly changing market demands of the beauty industry.

[0160] Although the methods described above are illustrated and depicted as a series of actions for the sake of simplicity, it should be understood and appreciated that these methods are not limited by the order of the actions, as some actions may occur in a different order and / or concurrently with other actions from the illustrations and descriptions herein or not illustrated and described herein but which may be understood by those skilled in the art, according to one or more embodiments.

[0161] Those skilled in the art will understand that information, signals, and data can be represented using any of a variety of different techniques and arts. For example, the data, instructions, commands, information, signals, bits, symbols, and chips described throughout the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, light fields or optical particles, or any combination thereof.

[0162] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in a generalized manner in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the invention.

[0163] The various illustrative logic modules and circuits described in conjunction with the embodiments disclosed herein may be implemented or performed using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. The general-purpose processor may be a microprocessor, but in alternatives, it may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors cooperating with a DSP core, or any other such configuration.

[0164] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of both. The software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor such that the processor can read and write information to / from the storage medium. In an alternative, the storage medium may be integrated into the processor. The processor and storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and storage medium may reside as discrete components in the user terminal.

[0165] In one or more exemplary embodiments, the described functionality may be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functionality may be stored or transmitted as one or more instructions or code on or through a computer-readable medium. A computer-readable medium includes both computer storage media and communication media, encompassing any medium that facilitates the transfer of a computer program from one location to another. A storage medium may be any available medium accessible to a computer. By way of example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disc storage, disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and is accessible to a computer. Any connection is also legitimately referred to as a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of a medium. As used in this article, disk and disc include compact discs (CDs), laser discs, optical discs, digital multi-purpose discs (DVDs), floppy disks, and Blu-ray discs. Disks typically reproduce data magnetically, while discs reproduce data optically using lasers. Combinations of these should also be included within the scope of computer-readable media.

[0166] The prior description of this disclosure is provided to enable any person skilled in the art to make or use this disclosure. Various modifications to this disclosure will be apparent to those skilled in the art, and the general principles defined herein may be applied to other variations without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not intended to be limited to the examples and designs described herein, but should be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent recommendation method for the beauty industry, characterized in that, Including the following steps: Obtain user query data; Based on the user query data, the user intent is obtained through a dual-path intent recognition module, which includes a local feature extraction model for extracting keywords and a contextual semantic understanding model for understanding the usage scenario context. If the confidence level of intent recognition is less than a preset threshold, a graph compensation mechanism is triggered. This mechanism queries related information in a dynamic beauty knowledge graph based on the feature entities extracted by the dual-path intent recognition module to correct the user intent. The dynamic beauty knowledge graph includes a customer layer, a SOP process layer, and a business entity layer, and supports node parameterization and real-time updates. Based on the user intent, a response is generated using the dynamic beauty knowledge graph. The response includes personalized product recommendations and / or usage suggestions determined by a knowledge push algorithm. A real-time monitoring mechanism based on a rules engine is used to perform compliance checks and risk warnings on user behavior; and The recommendation strategy and system parameters are dynamically adjusted based on user feedback data using a closed-loop reinforcement learning optimization mechanism.

2. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, The local feature extraction model is the TextCNN model, and the context semantic understanding model is the BiLSTM model.

3. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, The steps for constructing the dual-path intent recognition module include: Based on product ingredient vocabulary and skin symptom description phrases in the beauty industry, the local feature extraction model and the contextual semantic understanding model are specifically optimized.

4. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, The customer layer includes user profile information, which includes stored user skin type and allergy history; The SOP process layer includes a standardized beauty operation process that includes testing, recommendation, and early warning nodes; The business entity layer includes an ingredient database, a skin type database, and / or an allergy case database.

5. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, The dynamic beauty knowledge graph supports dynamic binding and updates, which include the following steps: Real-time updates of the data in the business entity layer; and In response to the data update in the business entity layer, the relevant nodes in the customer layer and the SOP process layer are updated.

6. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, The steps for determining recommended content, including personalized product recommendations and / or usage suggestions, through knowledge-based recommendation algorithms include: Based on the user intent, similarity matching is performed in the dynamic beauty knowledge graph to determine the optimal SOP process node in the SOP process layer; Based on the SOP process nodes, user historical interaction data, and entity relationships in the dynamic beauty knowledge graph, calculate the correlation degree between each knowledge item in the dynamic beauty knowledge graph and the SOP process node; and Based on the correlation between the knowledge item and the SOP process node, each knowledge item is pushed as recommended content in descending order of correlation.

7. The intelligent recommendation method for the beauty industry as described in claim 6, characterized in that, The steps for determining recommended content, including personalized product recommendations and / or usage suggestions, through knowledge-based recommendation algorithms include: Based on the user profile information in the customer layer of the dynamic beauty knowledge graph, and combined with the relevant knowledge in the business entity layer, the SOP process recommended to users is dynamically adjusted.

8. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, It also includes the following steps: Based on the application scenario, the system recommends the SOP process node in the SOP process layer that best matches the application scenario to the user.

9. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, The steps for compliance detection and risk warning of user behavior through a real-time monitoring mechanism based on a rule engine include: The user's operational behavior data is matched with monitoring rules to detect whether the operational behavior data is abnormal; In response to the detection of abnormalities in the aforementioned operational behavior data, an alert is triggered and corrective suggestions are pushed to the user; and If the operation data detection is normal, proceed to the next step.

10. The intelligent recommendation method for the beauty industry as described in claim 9, characterized in that, The steps for determining recommended content, including personalized product recommendations and / or usage suggestions, through knowledge-based recommendation algorithms include: In response to the detection of anomalies in the operational behavior data, and in conjunction with the relevant knowledge in the business entity layer of the dynamic beauty knowledge graph, the subsequent SOP process recommended to the user is dynamically adjusted.

11. The intelligent recommendation method for the beauty industry as described in claim 1, characterized in that, The steps of dynamically adjusting the recommendation strategy and system parameters based on user feedback data using the closed-loop reinforcement learning optimization mechanism include: Obtain user feedback data, including user interaction data; and Based on the user feedback data, the weight parameters of the knowledge push algorithm are updated using a reinforcement learning algorithm.

12. An intelligent recommendation system for the beauty industry, characterized in that, include: Memory, on which computer instructions are stored; as well as A processor, connected to the memory, and configured to execute computer instructions stored on the memory to implement the intelligent recommendation method for the beauty industry as described in any one of claims 1 to 11.

13. A computer-readable storage medium storing computer instructions thereon, characterized in that, When the computer instructions are executed by the processor, they implement the intelligent recommendation method for the beauty industry as described in any one of claims 1 to 11.

Images