Intelligent product recommendation system based on cloud platform

Abstract

The invention discloses an intelligent product recommendation system based on a cloud platform, and the system comprises the steps: obtaining the user portrait information of a target user object, and obtaining the material information of a target candidate medical beauty product from a plurality of candidate medical beauty products; semantic coding is carried out on the user portrait of the target user object and the material information of the target candidate medical beauty product by utilizing a data analysis and processing mode based on artificial intelligence; and intelligently generating a recommendation probability value of the target candidate medical beauty product based on the semantic flow field alignment matching representation between the target user portrait semantic coding feature and the target candidate medical beauty product material semantic coding feature, and determining whether to use the target candidate medical beauty product as the target medical beauty product. In this way, the consistency and objectivity of recommendation results can be ensured. Meanwhile, the latest behavior and preference change of the user can be captured in real time, the recommendation model is updated in time, and it is ensured that the recommendation result always meets the current requirement of the user.

Description

Technical Field

[0001] The present application relates to the field of intelligent recommendation, and more specifically, to a product intelligent recommendation system based on a cloud platform. Background Art

[0002] In recent years, with the rapid expansion of the medical beauty market and the increasing demand for personalized services from consumers, the traditional way of recommending medical beauty products can no longer meet users' expectations. Consumers not only want to receive product recommendations that meet their unique needs, but also expect to find the most suitable medical beauty solutions in a convenient and efficient way.

[0003] Patent CN111859122B proposes a method, device, electronic device and readable storage medium for recommending medical beauty products, which receives images or videos uploaded by users, analyzes and determines body part problems and generates demand information. Based on this information, matching candidate products are selected from multiple medical beauty products. Combined with the user's consumption history, search and treatment records and other features, the most suitable target medical beauty products are further screened for recommendation.

[0004] In this patent, weight values ​​are set for each candidate medical beauty product based on the user's characteristic information, and the target medical beauty product is obtained based on the weight value. However, the setting of weight values ​​is highly dependent on the designer's experience and assumptions. Different experts may set different weights based on different backgrounds and understandings, resulting in deviations in the recommendation results. This subjectivity makes it difficult for the system to form a unified standard, affecting the consistency and accuracy of the recommendation. In addition, user preferences and needs are dynamically changing, and weight values ​​are usually set based on historical data and cannot reflect the user's latest needs in a timely manner. This means that even if the user's interests or circumstances have changed, the system may still make recommendations based on outdated data, resulting in recommendation results that do not match the user's current needs.

[0005] Therefore, an optimized product intelligent recommendation solution is desired. Summary of the invention

[0006] In order to solve the above technical problems, the present application is proposed. The embodiment of the present application provides a product intelligent recommendation system based on a cloud platform, which obtains the user portrait information of the target user object, and obtains the material information of the target candidate medical beauty product from the multiple candidate medical beauty products, and uses artificial intelligence-based data analysis and processing methods to semantically encode the user portrait of the target user object and the material information of the target candidate medical beauty product, thereby intelligently generating the recommendation probability value of the target candidate medical beauty product based on the semantic flow field alignment matching representation between the semantic coding features of the target user portrait and the semantic coding features of the target candidate medical beauty product material, and determining whether to use the target candidate medical beauty product as the target medical beauty product. In this way, the consistency and objectivity of the recommendation results can be ensured. At the same time, it can capture the latest behavior and preference changes of users in real time, update the recommendation model in a timely manner, and ensure that the recommendation results always meet the current needs of users.

[0007] According to one aspect of the present application, a product intelligent recommendation system based on a cloud platform is provided, which includes: a user data to be identified acquisition module, used to acquire the data to be identified input by a target user object; a data to be identified transmission module, used to transmit the data to be identified to a data processor based on a cloud platform; a demand information generation module, used to analyze the data to be identified in the data processor based on the cloud platform to obtain the demand information of the target user object; a medical beauty product candidate module, used to determine multiple candidate medical beauty products based on the demand information; a medical beauty product recommendation module, used to determine a target medical beauty product from the multiple candidate medical beauty products.

[0008] Compared with the prior art, the present application provides a product intelligent recommendation system based on a cloud platform, which obtains the user portrait information of the target user object, obtains the material information of the target candidate medical beauty product from the multiple candidate medical beauty products, and uses artificial intelligence-based data analysis and processing methods to semantically encode the user portrait of the target user object and the material information of the target candidate medical beauty product, thereby intelligently generating the recommendation probability value of the target candidate medical beauty product based on the semantic flow field alignment matching representation between the semantic coding features of the target user portrait and the semantic coding features of the target candidate medical beauty product materials, and determining whether to use the target candidate medical beauty product as the target medical beauty product. In this way, the consistency and objectivity of the recommendation results can be ensured. At the same time, it can capture the latest behavior and preference changes of users in real time, update the recommendation model in a timely manner, and ensure that the recommendation results always meet the current needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 A block diagram of a product intelligent recommendation system based on a cloud platform according to an embodiment of the present application; Figure 2 A data flow diagram of a product intelligent recommendation system based on a cloud platform according to an embodiment of the present application; Figure 3 A block diagram of a traditional Chinese medicine beauty product recommendation module in a cloud platform-based product intelligent recommendation system according to an embodiment of the present application; Figure 4 The present invention is a block diagram of a user-product feature matching unit in a cloud platform-based product intelligent recommendation system according to an embodiment of the present application. DETAILED DESCRIPTION

[0011] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0012] As shown in this application and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0013] Although the present application makes various references to certain modules in the system according to the embodiments of the present application, any number of different modules can be used and run on the user terminal and / or server. The modules are only illustrative, and different aspects of the system and method can use different modules.

[0014] Flowcharts are used in the present application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed accurately in order. On the contrary, various steps may be processed in reverse order or simultaneously as required. Meanwhile, other operations may also be added to these processes, or a certain step or several steps of operations may be removed from these processes.

[0015] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0016] In the technical solution of the present application, a product intelligent recommendation system based on a cloud platform is proposed. Figure 1 The block diagram of the cloud platform-based product intelligent recommendation system according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the cloud platform-based product intelligent recommendation system according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the cloud platform-based product intelligent recommendation system 300 includes: a user data acquisition module 310 for acquiring data to be identified input by a target user object; a data transmission module 320 for transmitting the data to be identified to a cloud platform-based data processor; a demand information generation module 330 for analyzing the data to be identified in the cloud platform-based data processor to obtain the demand information of the target user object; a medical beauty product candidate module 340 for determining a plurality of candidate medical beauty products based on the demand information; and a medical beauty product recommendation module 350 for determining a target medical beauty product from the plurality of candidate medical beauty products.

[0017] In particular, the user data to be identified acquisition module 310 is used to obtain the data to be identified input by the target user object. In the example of the present application, the system supports users to upload photos or video files that have been taken, and also allows users to directly use the camera for real-time shooting or recording. After receiving the data uploaded by the user, the system will first make a preliminary judgment on the validity of the data to ensure that the data is suitable for further processing. Specifically, the system will check whether the data contains images of body parts (such as face, head, etc.) that can be used for analysis. If the data is invalid, an error message will be fed back to the user and prompted to re-enter. In addition, the quality of the image or video, such as clarity, lighting conditions, etc., will also be evaluated to ensure the accuracy of subsequent analysis.

[0018] Once the data is confirmed to be valid, the system will automatically identify the body parts involved in the image or video (such as face, legs, etc.) through computer vision technology, and at the same time identify the user's age, gender and other basic information, so as to obtain the corresponding standard image as a comparison benchmark. Based on the above recognition results, the system retrieves the standard image that matches the determined body position, age and gender from the database, and establishes a reference framework so that the system can accurately identify the problems on the user's body (such as spots, skin conditions, etc.). Finally, the system compares the data to be identified with the standard image and generates the user's demand information based on the difference between the two. This process may include specific facial feature extraction (such as wrinkles, acne, etc.), problem location (clearly pointing out the specific problem in a certain part of the user's body) and detailed feature acquisition (for some specific problems, such as spot size, skin dryness, etc., the system may continue to conduct in-depth analysis to obtain more detailed user demand information).

[0019] In particular, the to-be-identified data transmission module 320 is used to transmit the to-be-identified data to a data processor based on a cloud platform. In the example of the present application, after the user submits an image or video through the system interface, the to-be-identified data will first be temporarily stored by the local device. In order to ensure the security of data transmission, the system will perform preliminary encryption processing on the received data, using industry-standard encryption algorithms such as AES (Advanced Encryption Standard) or RSA (Rivest-Shamir-Adleman) to prevent the data from being intercepted or tampered with during transmission. The encrypted data is then packaged into a format suitable for network transmission, such as a multimedia file stream under the HTTP / HTTPS protocol. Then, the system will establish a secure and reliable connection channel for sending the packaged data to the cloud server. This process is usually assisted by mature cloud computing service platforms, which provide a wealth of API interfaces and service options, support a variety of programming languages ​​and development environments, and enable developers to easily integrate data transmission functions. At the same time, they also have powerful network infrastructure and load balancing capabilities to ensure stable data transmission performance even during network fluctuations or peak hours.

[0020] In particular, the demand information generation module 330 is used to analyze the data to be identified in the cloud platform-based data processor to obtain the demand information of the target user object. In the example of the present application, when the image or video submitted by the user is transmitted to the cloud server through a secure encrypted channel, it will first be sent to a preprocessing module. In this module, the system will perform preliminary processing on the received data, including but not limited to format conversion, size adjustment, and quality optimization. These steps are intended to lay a good foundation for subsequent more in-depth analysis and ensure the consistency and availability of input data. For example, for image files of different resolutions or encoding formats, the system will uniformly convert them into standard specifications; for video streams, key frames may need to be extracted to complete the analysis faster. Next, enter the core analysis stage. The key here is how to effectively identify and understand the body part features contained in the image or video and the differences between them and the ideal state. To this end, the system uses a series of advanced computer vision technologies and deep learning models for meticulous analysis. Specifically, algorithms such as convolutional neural networks (CNN) will be used to automatically locate the face or other relevant body areas in the image, and mark important anatomical landmarks such as the location of the eyes, nose, and mouth. This step is crucial for more accurate problem detection later because it provides a basic reference framework for the user's physical condition. Once the body area of ​​interest is determined, the system will further perform problem detection and feature extraction. In this process, a variety of different methods and techniques may be combined, such as skin texture analysis based on statistical models, color distribution statistics, and morphological transformations to identify specific skin problems (such as spots, acne), degree of wrinkles, or irregular contours. In addition, in order to improve accuracy, some auxiliary tools can be introduced, such as 3D reconstruction technology to generate a more three-dimensional and intuitive facial model, so as to better evaluate certain subtle changes. At the same time, the system will also try to identify basic information such as the user's age and gender to prepare for the next step of standard image matching.

[0021] In particular, the medical beauty product candidate module 340 is used to determine multiple candidate medical beauty products based on the demand information. That is, after the user's demand information is generated, multiple candidate medical beauty products are extracted from the medical beauty product database based on the demand information. In the example of the present application, the system first obtains the material information of all available medical beauty products, including but not limited to the product's ingredient list, applicable skin type, expected effect, usage method, user evaluation, etc. Then, the natural language processing (NLP) technology is used to convert these text descriptions into a structured data format for subsequent analysis. For example, through a word embedding model such as Word2Vec or BERT, the ingredient description of the product can be converted into a high-dimensional vector representation, thereby capturing the semantic similarity between different ingredients. At the same time, for some quantitative attributes, such as applicable age range, price range, etc., they are directly stored in numerical form. Next, the system calculates the matching degree between the user's demand information and the material information of each medical beauty product. Specifically, the system will find products with corresponding effects based on the user's specific demand points (such as freckle removal, anti-wrinkle, moisturizing, etc.), and evaluate whether their ingredients are suitable for the user's skin type. In addition, the safety and side effect risks of the product will be considered to ensure that the recommended product is not only effective but also safe and reliable. In order to improve the accuracy of the matching calculation, the system may introduce some advanced algorithms, such as support vector machine (SVM), random forest or neural network, to simulate expert-level judgment logic.

[0022] In particular, the medical beauty product recommendation module 350 is used to determine a target medical beauty product from the plurality of candidate medical beauty products. Figure 3 As shown, the medical beauty product recommendation module 350 includes: a user portrait information acquisition unit 351, which is used to acquire the user portrait information of the target user object; a material information acquisition unit 352, which is used to acquire the material information of the target candidate medical beauty product from the multiple candidate medical beauty products; a user product information encoding unit 353, which is used to semantically encode the user portrait information of the target user object and the material information of the target candidate medical beauty product respectively to obtain the target user portrait semantic encoding features and the target candidate medical beauty product material semantic encoding features; a user-product feature matching unit 354, which is used to perform semantic flow field-guided feature response matching on the target user portrait semantic encoding features and the target candidate medical beauty product material semantic encoding features to obtain a target user-medical beauty product fine-grained alignment response matching representation; a product judgment unit 355, which is used to determine whether to use the target candidate medical beauty product as the target medical beauty product based on the target user-medical beauty product fine-grained alignment response matching representation.

[0023] Specifically, the user portrait information acquisition unit 351 is used to acquire user portrait information of the target user object. In particular, the user portrait information includes basic information, consumption information, search information and treatment information. In one example, basic information includes the user's age, gender, geographic location, etc. These basic information help the system to preliminarily screen out the categories of medical beauty products that may be suitable for the user. For example, users of different age groups and genders may pay more attention to different beauty issues. Young people may be more concerned about acne and uneven skin tone, while middle-aged people may pay more attention to anti-aging and wrinkle treatment. Therefore, by understanding these basic attributes of the user, the system can narrow the range of candidate products and improve the pertinence of the recommendation; and consumption information reflects the user's historical purchasing behavior and consumption habits. By analyzing the user's past purchase records, the system can identify the user's preferred brands, price ranges, and products with specific ingredients or efficacy types. For example, if a user often buys high-end brand whitening products, the system may tend to choose new products with the same price range and similar functions when recommending. This recommendation based on historical behavior patterns not only improves user acceptance, but also increases the possibility of sales conversion; in addition, search information is also an important part of user portraits. A user's search history can reveal their current interests and potential needs, even if these needs have not yet been converted into actual purchases. For example, if a user frequently searches for information about freckle removal but has never purchased related products, this may indicate that they have a strong interest in the issue but are still in the wait-and-see stage. At this point, the system can stimulate the user's interest and guide them to make decisions by pushing some popular or well-reviewed freckle removal products in a timely manner. Treatment information should also not be ignored, as it covers the medical beauty services or treatment plans that the user has received. This part of information is particularly valuable to the system because it is directly related to the user's physical condition and treatment effect. For example, if the user has had laser freckle removal surgery before, the system needs to take this into account when recommending new spot care products, and give priority to those products that will not conflict with the original treatment or can play an auxiliary role. In this way, not only the professionalism and scientificity of the recommendation are improved, but also a more comprehensive solution is provided for users.

[0024] Specifically, the material information acquisition unit 352 is used to acquire the material information of the target candidate medical beauty product from the multiple candidate medical beauty products. It should be understood that acquiring the material information of the candidate medical beauty products can help the system fully understand the characteristics of each product, so as to perform more refined screening. These material information covers a wide range of content, including but not limited to the product's ingredient list, applicable skin type, expected effects, usage methods, user reviews, etc. For example, for a freckle cream, the system not only needs to know its main ingredients (such as niacinamide, tranexamic acid, etc.), but also needs to understand the mechanism of action of these ingredients and their adaptability to different skin types. In this way, when facing users with specific needs, the system can select the most suitable product according to their skin type and personal preferences. Further, detailed material information helps the system evaluate the match between each candidate product and user needs. By comparing the user's personalized needs (such as freckle removal, anti-wrinkle, moisturizing, etc.) with the functional characteristics of the product, the system can calculate the similarity or difference between the two, and then determine which products are most likely to meet the user's expectations. For example, if the user's demand is for whitening products for sensitive skin, the system will give priority to products with mild formulas and no irritating ingredients, and give higher weights based on their actual effect scores. This data-driven matching method not only improves the accuracy of recommendations, but also enhances the user experience. In addition, obtaining material information can provide users with more references and increase their trust in the recommended results. When the system recommends a medical beauty product to a user, the accompanying detailed instructions can allow users to more clearly understand the product's background information, advantages and potential risks. For example, displaying user reviews can help other consumers judge the true effect of the product; and providing a detailed guide to use can guide users to apply it correctly and avoid adverse consequences due to incorrect operations. In this way, not only the effectiveness of recommendations is improved, but also better decision-making support is provided to users.

[0025] Specifically, the user product information encoding unit 353 is used to semantically encode the user portrait information of the target user object and the material information of the target candidate medical beauty product, respectively, to obtain the semantic encoding features of the target user portrait and the material semantic encoding features of the target candidate medical beauty product. In the technical solution of the present application, the user portrait information of the target user object is passed through a user portrait semantic encoder containing a Bert model to obtain a target user portrait semantic encoding feature vector as the target user portrait semantic encoding feature. Here, considering that the user portrait information of the target user object contains a large amount of rich semantic information and deep meanings, these information reflect the user's preferences and needs. Therefore, in order to deeply explore language features and capture potential semantic connotations, in the technical solution of the present application, the user portrait information of the target user object is passed through a user portrait semantic encoder containing a Bert model to perform deep semantic understanding of the user portrait information through the Bert model, extract key contextual semantic information and implicit semantic meanings, and obtain a target user portrait semantic encoding feature vector. Similarly, considering that the material information of the target candidate medical beauty product contains multiple aspects of content, its product description, ingredient list, applicable population and other information usually exist in text form, and there are semantic dependencies between the contexts of these text information. Based on this, in order to further capture and extract the semantic information between the key contexts of the product, in the technical solution of the present application, the material information of the target candidate medical beauty product is semantically encoded by a medical beauty product semantic encoder including a Bert model to extract a richer semantic representation and obtain the target candidate medical beauty product material semantic encoding feature vector. In this way, the feature vector of each medical beauty product can accurately reflect its unique attributes, thereby improving the relevance of the recommendation.

[0026] Specifically, the user-product feature matching unit 354 is used to perform semantic flow field-guided feature response matching on the semantic coding features of the target user portrait and the semantic coding features of the target candidate medical beauty product materials to obtain a fine-grained aligned response matching representation of the target user and the medical beauty product. Considering that the semantic coding features of the target user portrait and the semantic coding features of the target candidate medical beauty product materials respectively express the user's specific needs for medical beauty products and the detailed semantic information of the products. If these features are simply merged, information may be lost or fail to accurately reflect the actual situation. Based on this, the present application introduces a semantic flow field-guided feature response matching mechanism to perform response matching on the semantic coding features of the target user portrait and the semantic coding features of the target candidate medical beauty product materials to obtain a fine-grained aligned response matching representation of the target user and the medical beauty product. That is, by building a matching mechanism that can capture the complex relationship between users and medical beauty products, fine-grained matching from user needs to specific product features can be achieved, which includes not only explicit needs (such as specific ingredients, applicable skin types), but also implicit preferences (such as user treatment history or potential skin problems), so that the user's unique needs can be more accurately identified and matched with the most suitable product features, thereby providing highly personalized recommendations and improving user satisfaction. In a specific example of this application, Figure 4 As shown, the user-product feature matching unit 354 includes: a target user-product feature alignment subunit 3541, which is used to perform feature alignment based on the semantic field on the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector to obtain the aligned target user portrait semantic encoding feature vector and the aligned target candidate medical beauty product material semantic encoding feature vector; a target user portrait linear transformation subunit 3542, which is used to perform linear transformation on the aligned target user portrait semantic encoding feature vector to obtain a target user portrait query vector and a target user portrait value vector; a target candidate medical beauty product material linear transformation subunit 3543, which is used to perform linear transformation on the aligned target candidate medical beauty product material semantic encoding feature vector to obtain a target candidate medical beauty product material key vector; a target user product feature fine-grained matching subunit 3544, which is used to perform fine-grained response encoding on the target user portrait query vector, the target user portrait value vector and the target candidate medical beauty product material key vector to obtain a target user-medical beauty product fine-grained alignment response matching representation vector as the target user-medical beauty product fine-grained alignment response matching representation.

[0027] More specifically, the target user-product feature alignment subunit 3541 is used to perform feature alignment based on the semantic field on the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector to obtain the aligned target user portrait semantic encoding feature vector and the aligned target candidate medical beauty product material semantic encoding feature vector. In an embodiment of the present application, first, a target user-medical beauty product semantic flow field is constructed between the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector. Among them, the target user-medical beauty product semantic flow field can capture the complex semantic relationship space between the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector. By constructing the target user-medical beauty product semantic flow field, the subsequent feature alignment process can more accurately identify and adjust the subtle differences between features, thereby laying the foundation for deeper feature interactions. In a specific implementation, the process of constructing a semantic flow field between the first feature vector and the second feature vector generally includes: performing feature activation based on point convolution and Sigmoid function on the semantic coding feature vector of the target user portrait and the semantic coding feature vector of the target candidate medical beauty product material to obtain the target user portrait semantic coding activation feature vector and the target candidate medical beauty product material semantic coding activation feature vector; performing semantic interaction based on convolution kernel attention on the semantic coding activation feature vector of the target user portrait and the semantic coding activation feature vector of the target candidate medical beauty product material to obtain the target user-medical beauty product semantic flow field. More specifically, the target user-medical beauty product semantic flow field between the semantic coding feature vector of the target user portrait and the semantic coding feature vector of the target candidate medical beauty product material is constructed using the following semantic flow field construction formula; wherein, the semantic flow field construction formula is: ;in, is the semantic encoding feature vector of the target user portrait, is the semantic coding feature vector of the target candidate medical beauty product material, is the point convolutional coding, is the activation function, is the target user portrait semantic encoding activation feature vector, is the semantic encoding activation feature vector of the target candidate medical beauty product material, yes The transposed vector of is matrix multiplication, yes Length, It is a convolutional code with a convolution kernel of 5x5. is a convolutional code with a convolution kernel of 3×3. is the upsampling operation, It is the semantic flow field of target users - medical beauty products.

[0028] Then, based on the target user-medical beauty product semantic flow field, the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector are feature aligned, so that the target user portrait semantic encoding features and the target candidate medical beauty product material semantic encoding features from different sources or modalities can be accurately matched in the same semantic space to obtain the aligned target user portrait semantic encoding feature vector and the aligned target candidate medical beauty product material semantic encoding feature vector. More specifically, based on the target user-medical beauty product semantic flow field, the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector are feature aligned using the following feature alignment formula to obtain the aligned target user portrait semantic encoding feature vector and the aligned target candidate medical beauty product material semantic encoding feature vector; wherein, the feature alignment formula is: ;in, is the semantic encoding feature vector of the aligned target user portrait, It is the semantic encoding feature vector of the target candidate medical beauty product material after alignment.

[0029] More specifically, the target user portrait linear transformation subunit 3542 and the target candidate medical beauty product material linear transformation subunit 3543 are used to linearly transform the aligned target user portrait semantic encoding feature vector to obtain the target user portrait query vector and the target user portrait value vector; and linearly transform the aligned target candidate medical beauty product material semantic encoding feature vector to obtain the target candidate medical beauty product material key vector. Among them, the query vector and the key vector are used to locate relevant features, while the value vector represents the specific information content. In order to ensure the effectiveness of the transformation, the linear transformation matrix is ​​usually defined by the parameters learned during the training process to ensure that the aligned feature vector can maximize the retention of the semantic information of the original features and has stronger expressiveness in the new feature space. More specifically, the aligned target user portrait semantic encoding feature vector and the aligned target candidate medical beauty product material semantic encoding feature vector are linearly transformed by the following linear transformation formula to obtain the target user portrait query vector, the target user portrait value vector and the target candidate medical beauty product material key vector; wherein, the linear transformation formula is: ;in, and They are the target user portrait query embedding matrix and the target user portrait query bias vector, is the target user portrait query vector, and are the target user portrait value embedding matrix and the target user portrait value bias vector, respectively. is the target user portrait value vector, and They are the target candidate medical beauty product material key embedding matrix and the target candidate medical beauty product material key bias vector, is the material key vector of the target candidate medical beauty products.

[0030] More specifically, the target user product feature fine-grained matching subunit 3544 is used to perform fine-grained response encoding on the target user portrait query vector, the target user portrait value vector and the target candidate medical beauty product material key vector to obtain a target user-medical beauty product fine-grained alignment response matching representation vector as the target user-medical beauty product fine-grained alignment response matching representation vector. In a specific example of the present application, the target user portrait query vector, the target user portrait value vector and the target candidate medical beauty product material key vector are subjected to fine-grained response matching based on a converter structure to obtain the target user-medical beauty product fine-grained alignment response matching representation vector. The converter structure can handle long-distance dependencies without relying on sequence order and model the interaction between features through a multi-head self-attention mechanism and a feedforward neural network. In the fine-grained response encoding process, the distribution of attention weights is guided by calculating the similarity between the query vector and the key vector, thereby determining which parts of the value vector should be emphasized or suppressed. After multiple layers of attention mechanisms and nonlinear transformations, the target user-medical beauty product fine-grained alignment response matching representation vector is obtained. More specifically, the target user portrait query vector, the target user portrait value vector and the target candidate medical beauty product material key vector are subjected to fine-grained response matching based on a converter structure using the following fine-grained response matching formula to obtain the target user-medical beauty product fine-grained alignment response matching representation vector; wherein the fine-grained response matching formula is: ;in, yes The transposed vector of yes Length, yes function, It is the target user-medical beauty product fine-grained alignment response matching representation vector.

[0031] Specifically, the product judgment unit 355 is used to determine whether to use the target candidate medical beauty product as the target medical beauty product based on the target user-medical beauty product fine-grained alignment response matching representation. That is, in the technical solution of the present application, based on the target user-medical beauty product fine-grained alignment response matching representation vector, a recommendation result is obtained, and the recommendation result is used to represent the recommendation probability value of the target candidate medical beauty product; that is, the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector are used to perform semantic flow field response matching to obtain the target user-medical beauty product fine-grained alignment response matching representation for classification processing, so as to intelligently generate the recommendation probability value of the target candidate medical beauty product. In this way, the latest behavior and preference changes of users can be captured in real time, and the recommendation model can be updated in time to ensure that the recommendation results always meet the current needs of users. Furthermore, the recommendation result is compared with a predetermined threshold to determine whether to use the target candidate medical beauty product as the target medical beauty product. Therefore, by setting a reasonable threshold, it can be ensured that only those medical beauty products that are highly matched with user needs will be finally recommended, which helps to improve the quality of recommendations and ensure that the recommended products do meet user expectations and needs.

[0032] In a preferred example, passing the target user-medical beauty product fine-grained alignment response matching representation vector through a classifier-based medical beauty product recommender to obtain a recommendation result includes: Determine the number of non-zero eigenvalues ​​in the target user-medical beauty product fine-grained alignment response matching representation vector, and subtract the number of non-zero eigenvalues ​​from the length of the target user-medical beauty product fine-grained alignment response matching representation vector to obtain the target user-medical beauty product fine-grained alignment response matching zero-dimensional value ; Calculate the zero-dimensional value of the target user-medical beauty product fine-grained alignment response matching minus one, and the zero-differentiable value of the target user-medical beauty product fine-grained alignment response matching The first target user-medical beauty product fine-grained alignment response matching field fitting value obtained by multiplying and dividing by the zero-dimensional value of the target user-medical beauty product fine-grained alignment response matching Fine-grained alignment response matching field fitting value with the second target user - medical beauty products ; Calculate the square root of the sum of squares of all eigenvalues ​​of the target user-medical beauty product fine-grained alignment response matching representation vector to obtain the target user-medical beauty product fine-grained alignment response matching pattern representation value ; Match the zero differentiable value with the target user-medical beauty product fine-grained alignment response Calculate each eigenvalue of the target user-medical beauty product fine-grained alignment response matching representation vector as an index The power function , and multiplied by the first target user-medical beauty product fine-grained alignment response matching field fitting value To obtain the local representation value of the fine-grained alignment response matching between the target user and the medical beauty product ; Calculate each eigenvalue of the target user-medical beauty product fine-grained alignment response matching representation vector Fine-grained alignment response matching field fitting value with the second target user-medical beauty product and the target user-medical beauty product fine-grained alignment response matching pattern representation value The product of obtains the target user-medical beauty product fine-grained alignment response matching center pointing value ; Match the local representation value of the target user-medical beauty product fine-grained alignment response Fine-grained alignment response matching center pointing value with the target user-medical beauty product The weighted difference Composition-optimized target user-medical beauty product fine-grained alignment response matching representation vector; and The optimized target user-medical beauty product fine-grained alignment response matching representation vector is passed through a classifier-based medical beauty product recommender to obtain a recommendation result.

[0033] Here, when the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector respectively represent the encoded semantic features of the target user portrait information and the target candidate medical beauty product material information, when performing feature interactive response matching based on semantic flow field modulation, the misalignment of the source sample semantic flow differentiation field modulation will also cause the interactive response matching richness of the target user-medical beauty product fine-grained aligned response matching representation vector, resulting in repetition of unstructured probability mapping, affecting the accuracy of the classification results.

[0034] Therefore, based on the vector field fitting law of high-dimensional manifolds, the isolated zeros in the target user-medical beauty product fine-grained alignment response matching representation vector are taken as the differentiable dimension of the high-dimensional manifold, so as to fix the normal direction pointing to the distribution center near the local coordinates represented by the eigenvalues ​​of the target user-medical beauty product fine-grained alignment response matching representation vector. In this way, the invalid repetition in the probability mapping process caused by the unstructured characteristics of the semantically rich distribution of the target user-medical beauty product fine-grained alignment response matching representation vector is avoided by effectively aligning the characteristic distribution pattern of the target user-medical beauty product fine-grained alignment response matching representation vector to the probability density differentiable field, thereby improving the accuracy of the recommendation results obtained by the classifier-based medical beauty product recommender.

[0035] As described above, the product intelligent recommendation system 300 based on the cloud platform according to the embodiment of the present application can be implemented in various wireless terminals, such as a server with a product intelligent recommendation algorithm based on the cloud platform. In one possible implementation, the product intelligent recommendation system 300 based on the cloud platform according to the embodiment of the present application can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the product intelligent recommendation system 300 based on the cloud platform can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the product intelligent recommendation system 300 based on the cloud platform can also be one of the many hardware modules of the wireless terminal.

[0036] Alternatively, in another example, the cloud-based product intelligent recommendation system 300 and the wireless terminal may also be separate devices, and the cloud-based product intelligent recommendation system 300 may be connected to the wireless terminal via a wired and / or wireless network and transmit interactive information in accordance with an agreed data format.

[0037] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A product intelligent recommendation system based on a cloud platform, characterized in that: include: A user data acquisition module for obtaining data to be identified, used to obtain the data to be identified input by the target user object; A module for transmitting data to be identified, used for transmitting the data to be identified to a data processor based on a cloud platform; a module for generating demand information, used for analyzing the data to be identified on the data processor based on a cloud platform to obtain demand information of a target user object; a module for candidate medical beauty products, used for determining a plurality of candidate medical beauty products based on the demand information; A medical beauty product recommendation module is used to determine a target medical beauty product from the plurality of candidate medical beauty products; wherein the medical beauty product recommendation module includes: A user portrait information acquisition unit, used to acquire user portrait information of a target user object; A material information acquisition unit, used to acquire material information of a target candidate medical beauty product from the plurality of candidate medical beauty products; A user product information encoding unit, used to semantically encode the user portrait information of the target user object and the material information of the target candidate medical beauty product to obtain a target user portrait semantic encoding feature and a target candidate medical beauty product material semantic encoding feature; A user-product feature matching unit, configured to perform semantic flow field-guided feature response matching on the semantic coding features of the target user portrait and the semantic coding features of the target candidate medical beauty product material to obtain a target user-medical beauty product fine-grained alignment response matching representation; A product judgment unit is used to determine whether to use the target candidate medical beauty product as the target medical beauty product based on the target user-medical beauty product fine-grained alignment response matching representation.

2. The product intelligent recommendation system based on the cloud platform according to claim 1, characterized in that: The user portrait information includes basic information, consumption information, search information and treatment information.

3. The product intelligent recommendation system based on the cloud platform according to claim 2 is characterized in that: The user product information encoding unit comprises: A target user portrait semantic encoding subunit, used for passing the user portrait information of the target user object through a user portrait semantic encoder including a Bert model to obtain a target user portrait semantic encoding feature vector as the target user portrait semantic encoding feature; The target candidate medical beauty product material semantic coding subunit is used to pass the material information of the target candidate medical beauty product through a medical beauty product semantic encoder including a Bert model to obtain a target candidate medical beauty product material semantic coding feature vector as the target candidate medical beauty product material semantic coding feature.

4. The product intelligent recommendation system based on the cloud platform according to claim 3 is characterized in that: The user-product feature matching unit includes: A target user product feature alignment subunit, used for performing feature alignment on the target user portrait semantic coding feature vector and the target candidate medical beauty product material semantic coding feature vector based on the semantic field to obtain an aligned target user portrait semantic coding feature vector and an aligned target candidate medical beauty product material semantic coding feature vector; A target user portrait linear transformation subunit, used for performing a linear transformation on the aligned target user portrait semantic encoding feature vector to obtain a target user portrait query vector and a target user portrait value vector; A target candidate medical beauty product material linear transformation subunit, used to perform a linear transformation on the aligned target candidate medical beauty product material semantic encoding feature vector to obtain a target candidate medical beauty product material key vector; The target user product feature fine-grained matching subunit is used to perform fine-grained response encoding on the target user portrait query vector, the target user portrait value vector and the target candidate medical beauty product material key vector to obtain a target user-medical beauty product fine-grained alignment response matching representation vector as the target user-medical beauty product fine-grained alignment response matching representation.

5. The product intelligent recommendation system based on the cloud platform according to claim 4 is characterized in that: The target user product feature alignment subunit includes: The target user product semantic flow field constructs a secondary subunit, which is used to construct a target user-medical beauty product semantic flow field between the target user portrait semantic coding feature vector and the target candidate medical beauty product material semantic coding feature vector; The feature alignment secondary sub-unit is used to perform feature alignment on the target user portrait semantic encoding feature vector and the target candidate medical beauty product material semantic encoding feature vector based on the target user-medical beauty product semantic flow field to obtain the aligned target user portrait semantic encoding feature vector and the aligned target candidate medical beauty product material semantic encoding feature vector.

6. The product intelligent recommendation system based on the cloud platform according to claim 5, characterized in that: The target user product semantic flow field constructs a secondary subunit for: Performing feature activation based on point convolution and Sigmoid function on the semantic coding feature vector of the target user portrait and the semantic coding feature vector of the target candidate medical beauty product material to obtain the semantic coding activation feature vector of the target user portrait and the semantic coding activation feature vector of the target candidate medical beauty product material; The semantic encoding activation feature vector of the target user portrait and the semantic encoding activation feature vector of the target candidate medical beauty product material are subjected to semantic interaction based on convolution kernel attention to obtain the target user-medical beauty product semantic flow field.

7. The product intelligent recommendation system based on the cloud platform according to claim 6, characterized in that: The target user product feature fine-grained matching subunit is used to: perform fine-grained response matching on the target user portrait query vector, the target user portrait value vector and the target candidate medical beauty product material key vector based on the converter structure to obtain the target user-medical beauty product fine-grained alignment response matching representation vector.

8. The product intelligent recommendation system based on the cloud platform according to claim 7, characterized in that: The product judgment unit includes: A recommendation probability value generating subunit, used to obtain a recommendation result based on the target user-medical beauty product fine-grained alignment response matching representation vector, wherein the recommendation result is used to represent the recommendation probability value of the target candidate medical beauty product; A threshold comparison subunit is used to compare the recommendation result with a predetermined threshold to determine whether to use the target candidate medical beauty product as a target medical beauty product.

9. The cloud platform-based product intelligent recommendation system according to claim 8, characterized in that: The recommendation probability value generating subunit is used to: pass the target user-medical beauty product fine-grained alignment response matching representation vector through a classifier-based medical beauty product recommender to obtain the recommendation result, and the recommendation result is used to represent the recommendation probability value of the target candidate medical beauty product.

Images