Non-overlapping cross-domain recommendation method and device based on fuzzy prototype migration

Abstract

The invention discloses a non-overlapping cross-domain recommendation method and device based on fuzzy prototype migration. The method comprises the steps that user potential features and project potential features of a source domain and user potential features and project potential features of a target domain are extracted through a potential factor model; performing fuzzy clustering on the user potential features and the project potential features of the source domain and the user potential features and the project potential features of the target domain through a maximum entropy loss function to generate a user fuzzy fusion prototype and a project fuzzy fusion prototype; an asymmetric double-prototype fuzzy migration module is constructed, user potential features and project potential features of a source domain and a target domain, a user fuzzy fusion prototype and a project fuzzy fusion prototype are aligned with a cross-domain prototype, and a user prototype set and a project prototype set are generated; and inputting the user prototype set, the project prototype set and the user potential features and the project potential features of the target domain into a recommendation model to generate a target domain recommendation result. According to the invention, accurate recommendation is realized in the cross-domain recommendation task.

Description

Technical Field

[0001] This application relates to the field of data recommendation technology, and more specifically, to a non-overlapping cross-domain recommendation method and apparatus based on fuzzy prototype transfer. Background Technology

[0002] In the internet age of information overload, effective data recommendation methods have provided convenience for production and life in various industries, enabling people to efficiently extract information of interest. Cross-domain recommendation can extract user evaluation records, feedback information, search records, and click information from different domains to extract data of interest to users and then push it to users. When users have no historical data or very little data, the cold start of cross-domain recommendation becomes a difficult problem for recommendation systems to overcome.

[0003] Current conventional cross-domain recommendation methods typically use user behavior data (such as ratings, clicks, etc.) in the source domain to train a recommendation model and extract user interest representations or item feature representations. Then, in the target domain, these shared representations (such as through transfer learning, joint training, or feature mapping) are used to assist in the recommendation in the target domain, thereby alleviating the problem of data sparsity in the target domain and improving the recommendation effect.

[0004] Current research on cross-domain recommendation (NCDR) faces the following challenges: First, cross-domain knowledge transfer is difficult. In the absence of explicit entity associations (such as shared items or user IDs), traditional methods struggle to establish reliable cross-domain mapping relationships. Second, the clustering granularity does not match the multi-category nature of entities. Forcing users or items into a single category ignores the reality that entities may belong to multiple categories simultaneously, leading to the loss of semantic information and making it difficult to effectively capture the complex interaction relationships between users and items, thus reducing the accuracy of cross-domain recommendations. Summary of the Invention

[0005] To address at least one deficiency or improvement need in the prior art, this invention provides a non-overlapping cross-domain recommendation method and apparatus based on fuzzy prototype transfer. This method solves the problems in the prior art where, in the absence of explicit entity associations, it is difficult to establish reliable cross-domain mapping relationships, the clustering granularity does not match the multi-classification characteristics of entities, and forced classification ignores the actual situation that an entity may belong to multiple categories at the same time, resulting in the loss of semantic information and reduced accuracy of cross-domain recommendations.

[0006] To achieve the above objectives, according to a first aspect of the present invention, a non-overlapping cross-domain recommendation method based on fuzzy prototype transfer is provided, comprising: The latent features of users and projects in the source domain, as well as the latent features of users and projects in the target domain, are extracted using the latent factor model. By using the maximum entropy loss function, fuzzy clustering is performed on the latent user features and project features in the source domain and the latent user features and project features in the target domain to generate user fuzzy fusion prototypes and project fuzzy fusion prototypes. An asymmetric dual-prototype fuzzy transfer module is constructed to align cross-domain prototypes based on the latent features of users and projects in the source and target domains, as well as the fuzzy fusion prototypes of users and projects, and to generate user prototype sets and project prototype sets. Input the user prototype set, the item prototype set, and the latent user features and item features of the target domain into the recommendation model to generate target domain recommendation results.

[0007] In one possible implementation, fuzzy clustering is performed on the latent user and item features of the source domain and the latent user and item features of the target domain using the maximum entropy loss function to generate user fuzzy fusion prototypes and item fuzzy fusion prototypes. This also includes: Initialize prototypes for the source and target domains respectively to obtain the initial user prototype and the initial project prototype for the source and target domains. Calculate the membership distribution based on the initial user prototype and project prototype in the source and target domains, and the latent user and project features in the source and target domains. By constraining the fuzzy membership distribution using the maximum entropy loss function, the initial user prototype and initial project prototype in the source and target domains are optimized. When the preset conditions are met, the optimized user initial prototype and project initial prototype of the source domain and target domain are the user fuzzy fusion prototype and project fuzzy fusion prototype, respectively.

[0008] In one possible implementation, based on the maximum entropy loss function, the membership distribution is calculated according to the similarity between the latent features of users and projects and the prototype, and it also includes: The formula for calculating membership degree distribution is as follows: ; in, It is the first in the source domain i The user on the first k The membership degree of each user, with a value range of [0,1]; Source Domain i The latent feature vector of each user; It is the source domain number k Initial prototype vectors for each user; It is the source domain number l The initial prototype vectors of each user; sim() is the similarity measurement function; τ is the temperature parameter (τ>0).

[0009] In one possible implementation, the user initial prototype and project initial prototype in the source and target domains are optimized by constraining the fuzzy membership distribution through the maximum entropy loss function, and further include: Based on the membership degree distribution, determine the negative similarity between the latent features of users and projects and their associated prototypes, and maximize the entropy of the membership degree distribution, and jointly construct the maximum entropy loss function; Calculate the gradient of the maximum entropy loss function and iteratively optimize the user's initial prototype and the project's initial prototype through backpropagation.

[0010] In one possible implementation, an asymmetric dual-prototype fuzzy transfer module is constructed. This module aligns cross-domain prototypes based on latent user and project features from the source and target domains, as well as user and project fuzzy fusion prototypes, and generates user and project prototype sets. The module also includes: Calculate the first minimum distance between the source domain user fuzzy fusion prototype and the target domain user fuzzy fusion prototype, and the second minimum distance between the source domain item fuzzy fusion prototype and the target domain item fuzzy fusion prototype; The loss function of the asymmetric dual-prototype fuzzy transfer module, constructed based on the first and second minimum distances, is aligned with the cross-domain prototype. The aligned cross-domain prototypes are integrated and redundancy-removed to form user prototype sets and project prototype sets.

[0011] In one possible implementation, the loss function of the asymmetric dual-prototype fuzzy transfer module is: ; in, and A prototype for fuzzy fusion of source domain users and projects. and A prototype for fuzzy fusion of target domain users and projects.

[0012] In one possible implementation, the user prototype set, the item prototype set, and the latent user and item features of the target domain are input into the recommendation model to generate target domain recommendation results. This also includes: Input the user prototype set, the item prototype set, the user latent features and the item latent features of the target domain into the recommendation model, and calculate the similarity between the user prototype set and the user latent features, as well as the similarity between the item prototype set and the item latent features. Based on the similarity between the user prototype set and the user latent features, and the similarity between the item prototype set and the item latent features, generate user prototype-level representations and item prototype-level representations in the target domain. The user prototype-level representation in the target domain is matched with the project prototype-level representation to obtain the recommendation list and generate the target domain recommendation results.

[0013] According to a second aspect of the present invention, a non-overlapping cross-domain recommendation apparatus based on fuzzy prototype transfer is also provided, comprising: The feature extraction module is configured to extract latent user features and project features from the source domain, as well as latent user features and project features from the target domain, using a latent factor model. The fuzzy clustering module is configured to perform fuzzy clustering on the latent features of users and items in the source domain and the latent features of users and items in the target domain using the maximum entropy loss function, generating user fuzzy fusion prototypes and item fuzzy fusion prototypes. The fuzzy migration module is configured to build an asymmetric dual-prototype fuzzy migration module, which aligns cross-domain prototypes based on the latent features of users and projects in the source and target domains, as well as the fuzzy fusion prototypes of users and projects, and generates user prototype sets and project prototype sets. The recommendation module is configured to input user prototype sets, item prototype sets, and latent user and item features of the target domain into the recommendation model to generate target domain recommendation results.

[0014] According to a third aspect of the present invention, a non-overlapping cross-domain recommendation device based on fuzzy prototype transfer is also provided, comprising at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program, and when the computer program is executed by the processing unit, the processing unit performs the steps of any of the above-described non-overlapping cross-domain recommendation methods based on fuzzy prototype transfer.

[0015] According to a fourth aspect of the invention, a storage medium is also provided, which stores a computer program executable by a non-overlapping cross-domain recommendation device based on fuzzy prototype transfer, wherein when the computer program is run on the non-overlapping cross-domain recommendation device based on fuzzy prototype transfer, the non-overlapping cross-domain recommendation device based on fuzzy prototype transfer performs the steps of any of the above-described non-overlapping cross-domain recommendation methods based on fuzzy prototype transfer.

[0016] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects: This invention provides a non-overlapping cross-domain recommendation method based on fuzzy prototype transfer. It extracts latent features of users and items from the source and target domains using a latent factor model, mapping users and items to a unified latent space. This captures shared latent structures across domains, overcoming the limitations of directly relying on explicit association information and enabling connections between domains without explicit associations through latent space features. Fuzzy clustering of user and item latent features is performed using the maximum entropy loss function, generating fuzzy fusion prototypes for users and items. This allows users or items to belong to multiple prototypes simultaneously with different weights, avoiding the "either / or" limitations of traditional clustering. It more accurately models the diversity of user interests or item attributes, preserves the semantic ambiguity in cross-domain features, and prevents information truncation caused by forced classification. An asymmetric dual-prototype fuzzy transfer module is constructed to align cross-domain prototypes based on features and initial prototypes from the source and target domains, generating user / item prototype sets. This reduces the differences in cross-domain feature distribution while preserving domain-specific feature patterns, improving the robustness and adaptability of cross-domain feature alignment. By inputting user prototype sets, item prototype sets, and target domain features into the recommendation model to generate recommendation results, and by matching the semantic relationship between target domain features and prototype sets, user preferences can be accurately predicted, significantly improving the hit rate of cross-domain recommendations and user satisfaction. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart illustrating an embodiment of the non-overlapping cross-domain recommendation method based on fuzzy prototype transfer provided by the present invention; Figure 2 Provided by the present invention Figure 1 A schematic flowchart of an embodiment of step S102; Figure 3 Provided by the present invention Figure 1 A schematic flowchart of an embodiment of step S103; Figure 4 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of step S104; Figure 5 A schematic diagram of a structure of an embodiment of the non-overlapping cross-domain recommendation device based on fuzzy prototype transfer provided by the present invention; Figure 6 This is a schematic diagram of the structure of a non-overlapping cross-domain recommendation device based on fuzzy prototype transfer, provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0020] The terms "first," "second," "third," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0021] This invention provides a non-overlapping cross-domain recommendation method and apparatus based on fuzzy prototype transfer, which will be described below.

[0022] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the non-overlapping cross-domain recommendation method based on fuzzy prototype transfer provided by the present invention. In a specific embodiment of the present invention, a non-overlapping cross-domain recommendation method based on fuzzy prototype transfer is disclosed, including: S101. Extract latent user and project features from the source domain and the target domain using the latent factor model. S102. Using the maximum entropy loss function, perform fuzzy clustering on the latent features of users and items in the source domain and the latent features of users and items in the target domain to generate user fuzzy fusion prototypes and item fuzzy fusion prototypes. S103. Construct an asymmetric dual-prototype fuzzy transfer module, align cross-domain prototypes based on the latent features of users and projects in the source and target domains, the user fuzzy fusion prototype and the project fuzzy fusion prototype, and generate user prototype sets and project prototype sets. S104. Input the user prototype set, the item prototype set, the user latent features and the item latent features of the target domain into the recommendation model to generate the target domain recommendation results.

[0023] In the above embodiments, the source domain is a domain with rich data information, and the target domain is the domain where we want to optimize recommendations. Latent factor models can be used to mine the performance of users and items across these latent feature dimensions. For example, in a movie recommendation scenario, the source domain contains a large amount of user rating data for different movies. A latent factor model can represent each user and each movie as a multi-dimensional vector, with each dimension of the vector corresponding to a latent feature. These features reflect the user's interests and preferences. In this way, high-dimensional, sparse, and difficult-to-understand user-item interaction data can be transformed into low-dimensional, dense latent feature representations with clear semantics, enabling a more fundamental characterization of user and item properties.

[0024] The maximum entropy loss function is a commonly used loss function in machine learning. Under certain constraints, it maximizes the entropy of the model, maximizing its uncertainty in prediction and thus better fitting the true distribution of the data. Fuzzy clustering differs from traditional hard clustering; it allows samples to belong to multiple categories with a certain membership degree. Using the maximum entropy loss function to perform fuzzy clustering on the latent features of the source and target domains can more flexibly reflect the distribution of samples in the feature space. Compared to hard clustering, fuzzy clustering can better handle the uncertainty and fuzziness of data. The generated user and item fuzzy fusion prototypes can comprehensively reflect the common features of similar users or items in the source and target domains, helping to improve the accuracy and generalization ability of clustering and making the generated prototypes more representative.

[0025] The asymmetric dual-prototype fuzzy transfer module analyzes potential asymmetries between the source and target domains. Based on the latent user and project characteristics obtained in previous steps, as well as the user and project fuzzy fusion prototypes, it aligns cross-domain prototypes. The most suitable matching relationship is found through fuzzy transformations and other methods. By aligning cross-domain prototypes, targeted knowledge transfer can be achieved, transferring useful information from the source domain to the target domain while avoiding interference from irrelevant information. The generated user and project prototype sets are more accurate and comprehensive, better characterizing the features of users and projects in the target domain.

[0026] Recommendation models can be various common machine learning or deep learning models, such as collaborative filtering models and deep neural networks, and this invention does not impose further limitations on them. These models perform calculations and inferences based on the input feature information using their internal algorithms and structures. By using multi-layer neural networks to fuse and transform features from user prototype sets, item prototype sets, and the potential features of users and items in the target domain, the complex relationships between them are learned, thereby predicting the degree of user preference for items.

[0027] By inputting the rich feature information obtained from the previous processing into the recommendation model, we can fully utilize the knowledge transferred across domains and the feature information of the target domain itself to generate more accurate and personalized recommendation results. Compared with recommendation models trained solely on data from the target domain, recommendation models that incorporate cross-domain information can uncover more potential user interests and item associations, improving the coverage and accuracy of recommendations.

[0028] Compared with existing technologies, this embodiment provides a non-overlapping cross-domain recommendation method based on fuzzy prototype transfer. It extracts latent features of users and items in the source and target domains using a latent factor model, mapping users and items to a unified latent space. This captures shared latent structures across domains, overcoming the limitations of directly relying on explicit association information, allowing domains without explicit associations to establish connections through latent space features. Fuzzy clustering of user and item latent features is performed using the maximum entropy loss function, generating fuzzy fusion prototypes for users and items. This allows users or items to belong to multiple prototypes simultaneously with different weights, avoiding the "either / or" limitations of traditional clustering. It more accurately models the diversity of user interests or item attributes, preserves the semantic ambiguity in cross-domain features, and prevents information truncation caused by forced classification. An asymmetric dual-prototype fuzzy transfer module is constructed to align cross-domain prototypes based on features and initial prototypes from the source and target domains, generating user / item prototype sets. This reduces the differences in cross-domain feature distribution while preserving domain-specific feature patterns, improving the robustness and adaptability of cross-domain feature alignment. By inputting user prototype sets, item prototype sets, and target domain features into the recommendation model to generate recommendation results, and by matching the semantic relationship between target domain features and prototype sets, user preferences can be accurately predicted, significantly improving the hit rate of cross-domain recommendations and user satisfaction.

[0029] Please see Figure 2 , Figure 2 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of step S102. In some embodiments of the present invention, fuzzy clustering is performed on the latent user features and item features of the source domain and the latent user features and item features of the target domain using the maximum entropy loss function to generate user fuzzy fusion prototypes and item fuzzy fusion prototypes. The method further includes: S201. Initialize prototypes for the source domain and target domain respectively to obtain the user initial prototype and project initial prototype for the source domain and target domain. S202. Calculate the membership distribution based on the initial user prototype, initial project prototype, latent user features, and latent project features in the source and target domains; S203. Optimize the user initial prototype and project initial prototype in the source domain and target domain by constraining the fuzzy membership distribution through the maximum entropy loss function. S204. When the preset conditions are met, the optimized user initial prototype and project initial prototype of the source domain and target domain are the user fuzzy fusion prototype and project fuzzy fusion prototype.

[0030] In the above embodiments, the initial user prototype and item prototype are the basis for subsequent clustering and optimization. The initial user prototype can be initialized based on information such as the statistical mean or mode of certain basic attributes of users in the source and target domains. The initial item prototype can be set based on attributes such as item category and average rating. Reasonable initialization enables the subsequent iterative optimization process to converge to better results more quickly, avoiding the algorithm getting trapped in local optima, and laying the foundation for discovering potential patterns of users and items in the source and target domains.

[0031] By utilizing existing initial user and project prototypes, and combining the actual latent user and project features in the source and target domains, the degree to which each user and project belongs to each prototype is determined, i.e., the membership distribution. This is achieved by calculating the distance or similarity between user or project features and prototype features, which is then converted into membership values. For example, if a user's feature vector is close to the feature vector of a user's initial prototype, then that user has a higher membership degree to that prototype.

[0032] Membership distribution reflects the degree of association between users and items in the source and target domains and each initial prototype. Membership information can clearly show the distribution of users and items in the prototype space, which helps to discover the similarities and differences of users and items in different domains.

[0033] The basic idea of ​​the maximum entropy principle is to maximize entropy under certain constraints, thereby maximizing the model's uncertainty or generalization ability. In this invention, the initial user prototype and initial project prototype in the source and target domains are adjusted to ensure that the membership distribution satisfies the requirements of the maximum entropy loss function. Specifically, during the optimization process, the prototype parameters are continuously iteratively updated to minimize the value of the maximum entropy loss function while ensuring that the membership distribution conforms to a certain degree of rationality.

[0034] By constraining the maximum entropy loss function, the optimized membership distribution becomes more reasonable and uniform, preventing the model from overfitting to certain samples or patterns. This helps improve the model's generalization ability, enabling the generated user and item fuzzy fusion prototypes to better adapt to different data and scenarios. The optimized prototypes more accurately represent the latent features of users and items in the source and target domains, thereby improving the performance of applications such as cross-domain recommendation.

[0035] Preset conditions may include reaching the maximum number of iterations, the maximum entropy loss function being less than a certain threshold, and the rate of change of the membership distribution being less than a certain value. When any one or more of these preset conditions are met, the optimization process is considered to have converged. At this point, the optimized initial user prototype and initial item prototype of the source and target domains are respectively determined as the user fuzzy fusion prototype and the item fuzzy fusion prototype. The user fuzzy fusion prototype and item fuzzy fusion prototype obtained in this way are fully optimized and adjusted; they integrate information from both the source and target domains, enabling a more accurate characterization of the potential features of users and items, and providing high-quality input for subsequent cross-domain analysis, recommendation, and other tasks.

[0036] In some embodiments of the present invention, based on the maximum entropy loss function, the membership distribution is calculated according to the similarity between the latent features of users and projects and the prototype, and the method further includes: The formula for calculating membership degree distribution is as follows: ; in, It is the first in the source domain i The user on the first k The membership degree of each user, with a value range of [0,1]; Source Domain i The latent feature vector of each user; It is the source domain number k Initial prototype vectors for each user; It is the source domain number l The initial prototype vectors of each user; sim() is the similarity measurement function; τ is the temperature parameter (τ>0).

[0037] In the above embodiments, in cross-domain scenarios, the user features of the source domain and the target domain may be inconsistent. By using membership weighting, the model only learns the source domain features that are highly related to users in the target domain, thereby reducing noise interference caused by inter-domain differences.

[0038] We construct a membership degree calculation method based on user-prototype similarity using the maximum entropy loss function. The aim is to improve the generalization and accuracy of recommendations in cross-domain recommendation scenarios by quantifying the degree of association between users and prototypes, thereby optimizing the representation of user features.

[0039] The membership distribution represents the similarity weight between users in the source domain and users in the target domain, with values ​​ranging from [0, 1]. Values ​​closer to 1 indicate stronger similarity between users in the source domain and users in the target domain. i With target domain users k The characteristics of the two users are highly similar; the closer the value is to 0, the greater the difference in their characteristics. By dynamically adjusting the membership degree, the model prioritizes retaining user information with high similarity during the transfer process, thus avoiding negative transfer.

[0040] It should be noted that the similarity metric function can quantify the degree of similarity between two vectors, and can be cosine similarity, Euclidean distance, etc. The temperature parameter can be adjusted according to actual needs, which can balance the model's dependence on highly similar users and avoid overfitting or underfitting.

[0041] In some embodiments of the present invention, optimizing the user initial prototype and project initial prototype in the source and target domains by constraining the fuzzy membership distribution using the maximum entropy loss function further includes: Based on the membership degree distribution, determine the negative similarity between the latent features of users and projects and their associated prototypes, and maximize the entropy of the membership degree distribution, and jointly construct the maximum entropy loss function; Calculate the gradient of the maximum entropy loss function and iteratively optimize the user's initial prototype and the project's initial prototype through backpropagation.

[0042] In the above embodiments, based on the obtained membership distribution information, the similarity relationships between user latent features and their respective user prototypes, and between project latent features and their respective project prototypes, are analyzed. The degree of difference between them is measured by calculating their negative similarity. Minimizing negative similarity encourages user latent features to align more closely with their respective prototypes, and the same applies to project latent features, thereby improving the prototypes' ability to generalize user and project features.

[0043] Due to the inherent uncertainty of membership distribution, it is necessary to maximize its entropy to measure information uncertainty and increase the model's generalization ability. By combining the objectives of minimizing negative similarity and maximizing entropy, the maximum entropy loss function is constructed. This comprehensively considers these two important aspects, guiding the model towards a better direction during training and making feature learning and prototype optimization more coordinated and effective.

[0044] After determining the maximum entropy loss function, the gradient of this loss function with respect to the parameters of the user's initial prototype and the project's initial prototype is calculated using the principle of differentiation in calculus. This gradient reflects the direction and rate of change of the loss function in the parameter space. Then, the calculated gradient is propagated from the output layer to the input layer using the backpropagation algorithm, gradually updating the parameters of the user's initial prototype and the project's initial prototype. In each iteration, the parameters are adjusted according to the direction of the gradient, causing the value of the loss function to gradually decrease.

[0045] Through backpropagation and iterative optimization, the initial user prototype and the initial item prototype are continuously adjusted in the direction of minimizing the maximum entropy loss function. As the iteration progresses, the prototypes increasingly accurately reflect the true feature distribution of users and items, thereby improving the model's understanding and representation of users and items, and helping to improve the model's accuracy and performance in recommendations.

[0046] Please see Figure 3 , Figure 3 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of step S103 is provided. In some embodiments of the present invention, an asymmetric dual-prototype fuzzy transfer module is constructed. Based on the latent features of users and items in the source and target domains, and the fuzzy fusion prototypes of users and items, cross-domain prototypes are aligned, and user prototype sets and item prototype sets are generated. The module also includes: S301. Calculate the first minimum distance between the source domain user fuzzy fusion prototype and the target domain user fuzzy fusion prototype, and the second minimum distance between the source domain item fuzzy fusion prototype and the target domain item fuzzy fusion prototype. S302. The loss function of the asymmetric dual-prototype fuzzy transfer module is constructed based on the first minimum distance and the second minimum distance to align the cross-domain prototype. S303. Integrate and remove redundancy from the aligned cross-domain prototypes to form user prototype sets and project prototype sets.

[0047] In the above embodiments, in order to establish connections between user-project interaction data in different domains and improve the adaptability and generalization ability of the entire model in cross-domain scenarios, an asymmetric dual-prototype fuzzy transfer module is constructed. Based on the potential user features and potential project features of the source and target domains, as well as the user fuzzy fusion prototype and the project fuzzy fusion prototype, cross-domain prototypes are aligned, and user prototype sets and project prototype sets are generated.

[0048] By calculating the first minimum distance and the second minimum distance, the similarity or difference between the source domain and the target domain in the dimensions of user prototype and project prototype can be quantified. This reflects the proximity of the two domains in the representation of key features, providing a quantitative feedback metric for the model. This allows the model to more accurately measure the gap between the current state and the ideal state during training, thereby guiding the adjustment of the model's parameters.

[0049] The purpose of constructing the loss function is to transform the distance information calculated earlier into an optimizable objective. By minimizing this loss function, the model can continuously adjust its parameters during training to align the prototypes of the source and target domains as closely as possible, thereby achieving better cross-domain feature transfer effects, learning more suitable cross-domain feature representations, improving the model's ability to transfer knowledge between different domains, and thus improving task performance in the target domain, such as recommendation accuracy.

[0050] The integration operation merges aligned prototypes from different domains into a unified and more representative prototype set. Redundancy removal eliminates duplicate or irrelevant prototype information that doesn't significantly improve model performance, making the prototype set more concise and efficient. The resulting user and project prototype sets are of higher quality and usability, providing the model with more accurate and discriminative information, further enhancing its performance in user-project related tasks.

[0051] In some embodiments of the present invention, the loss function of the asymmetric dual-prototype fuzzy migration module is: ; in, and A prototype for fuzzy fusion of source domain users and projects. and A prototype for fuzzy fusion of target domain users and projects.

[0052] In the above embodiments, the main purpose of the loss function is to measure the difference between the source domain and the target domain in the user and project fuzzy fusion prototypes. By minimizing the loss function, the model will be guided to adjust its own parameters so that the prototypes of the source domain and the target domain are as close as possible.

[0053] Reducing the value of the loss function means that the prototypes of the source and target domains are more similar, which helps the model better utilize knowledge from the source domain in the target domain, thereby improving performance in the target domain, such as increasing recommendation accuracy and recall. It also encourages the model to align the feature spaces of the source and target domains, resulting in better compatibility and consistency of data from both domains at the feature level, thus enhancing the model's generalization ability to data from different domains.

[0054] Please see Figure 4 , Figure 4 Provided by the present invention Figure 1 A flowchart illustrating an embodiment of step S104. In some embodiments of the present invention, the user prototype set, the item prototype set, the user latent features of the target domain, and the item latent features are input into the recommendation model to generate a target domain recommendation result, and the method further includes: S401. Input the user prototype set, the item prototype set, the user latent features and the item latent features of the target domain into the recommendation model, and calculate the similarity between the user prototype set and the user latent features, as well as the similarity between the item prototype set and the item latent features. S402. Generate user prototype-level representations and project prototype-level representations in the target domain based on the similarity between the user prototype set and user latent features, and the similarity between the project prototype set and project latent features. S403. Match the user prototype-level representation with the project prototype-level representation in the target domain to obtain the recommendation list and generate the target domain recommendation results.

[0055] In the above embodiments, the user prototype set, item prototype set, and latent user features and item features of the target domain are input into the recommendation model again. The similarity calculation module or algorithm in the recommendation model is then used to calculate the similarity between the user prototype set and the latent user features, and the similarity between the item prototype set and the latent item features, respectively. Various methods can be used for similarity calculation, such as cosine similarity and Euclidean distance.

[0056] By calculating similarity, we can quantify the degree of association between user prototypes and user latent features, and between project prototypes and project latent features. This helps to more accurately characterize the user's position in the prototype space and the prototype attributes of the project, thereby improving the performance of the recommendation system and the understanding of user-project relationships.

[0057] Based on the similarity between the user prototype set and user latent features, and the similarity between the item prototype set and item latent features calculated above, user prototype-level representations and item prototype-level representations are constructed in the target domain. A weighted averaging method can be used, assigning different weights to each prototype according to the magnitude of similarity, and then fusing the relevant feature information to generate new prototype-level representations.

[0058] The generated user prototype-level representation and item prototype-level representation in the target domain are input into the matching module, and a similarity score is calculated between them. Then, the items are sorted according to the similarity score, with higher-scoring items being prioritized for recommendation to their corresponding users. This results in a final recommendation list, yielding the target domain recommendation results. This matching method accurately recommends items that best match the user's prototype characteristics, fully utilizing the key user and item information contained in the generated prototype-level representations. This makes the recommendations more aligned with the user's true interests and needs, effectively improving the targeting and personalization of the recommendations, and enhancing user trust and willingness to use the recommendation system.

[0059] To better implement the non-overlapping cross-domain recommendation method based on fuzzy prototype transfer in this invention embodiment, based on the corresponding method, please refer to [link to relevant documentation]. Figure 5 , Figure 5 This is a schematic diagram of an embodiment of the non-overlapping cross-domain recommendation device based on fuzzy prototype transfer provided by the present invention. The embodiment of the present invention provides a non-overlapping cross-domain recommendation device 500 based on fuzzy prototype transfer, comprising: Feature extraction module 510 is configured to extract latent user features and item features of the source domain, as well as latent user features and item features of the target domain, through a latent factor model. The fuzzy clustering module 520 is configured to perform fuzzy clustering on the latent features of users and items in the source domain and the latent features of users and items in the target domain using the maximum entropy loss function, and generate user fuzzy fusion prototypes and item fuzzy fusion prototypes. Fuzzy transfer module 530 is configured to build an asymmetric dual-prototype fuzzy transfer module, align cross-domain prototypes based on user latent features and project latent features in the source and target domains, user fuzzy fusion prototypes and project fuzzy fusion prototypes, and generate user prototype sets and project prototype sets. The recommendation module 540 is configured to input the user prototype set, the item prototype set, the user latent features and the item latent features of the target domain into the recommendation model to generate target domain recommendation results.

[0060] It should be noted that the device 500 provided in the above embodiments can implement the technical solutions described in the above method embodiments. The specific implementation principles of the above modules or units can be found in the corresponding content in the above method embodiments, and will not be repeated here.

[0061] Please see Figure 6 , Figure 6 This is a schematic diagram of the structure of a non-overlapping cross-domain recommendation device based on fuzzy prototype transfer provided in an embodiment of the present invention. Based on the above-described non-overlapping cross-domain recommendation method based on fuzzy prototype transfer, the present invention also provides a non-overlapping cross-domain recommendation device based on fuzzy prototype transfer. This device can be a computing device such as a mobile terminal, desktop computer, laptop, handheld computer, or server. The non-overlapping cross-domain recommendation device 600 based on fuzzy prototype transfer includes a processor 610, a memory 620, and a display 630. Figure 6 Only a portion of the components of the non-overlapping cross-domain recommendation device based on fuzzy prototype transfer are shown. However, it should be understood that implementation of all shown components is not required, and more or fewer components may be implemented instead.

[0062] In some embodiments, memory 620 may be an internal storage unit of the non-overlapping cross-domain recommendation device 600 based on fuzzy prototype migration, such as a hard disk or memory of the non-overlapping cross-domain recommendation device 600 based on fuzzy prototype migration. In other embodiments, memory 620 may also be an external storage device of the non-overlapping cross-domain recommendation device 600 based on fuzzy prototype migration, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the non-overlapping cross-domain recommendation device 600 based on fuzzy prototype migration. Furthermore, memory 620 may include both internal storage units and external storage devices of the non-overlapping cross-domain recommendation device 600 based on fuzzy prototype migration. Memory 620 is used to store application software and various types of data installed on the non-overlapping cross-domain recommendation device 600 based on fuzzy prototype migration, such as program code installed on the non-overlapping cross-domain recommendation device 600 based on fuzzy prototype migration. Memory 620 may also be used to temporarily store data that has been output or will be output. In one embodiment, the memory 620 stores a non-overlapping cross-domain recommendation program 640 based on fuzzy prototype transfer. The non-overlapping cross-domain recommendation program 640 based on fuzzy prototype transfer can be executed by the processor 610 to implement the non-overlapping cross-domain recommendation method based on fuzzy prototype transfer of the various embodiments of this application.

[0063] In some embodiments, processor 610 may be a central processing unit (CPU), a microprocessor, or other data processing chip, used to run program code stored in memory 620 or process data, such as executing a non-overlapping cross-domain recommendation method based on fuzzy prototype migration.

[0064] Display 630 may, in some embodiments, be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. Display 630 is used to display information from the fuzzy prototype transfer-based non-overlapping cross-domain recommendation device 600 and to display a user interface for visualization. Components 610-630 of the fuzzy prototype transfer-based non-overlapping cross-domain recommendation device 600 communicate with each other via a system bus.

[0065] In one embodiment, the steps of the above-described non-overlapping cross-domain recommendation method based on fuzzy prototype transfer are implemented when the processor 610 executes the non-overlapping cross-domain recommendation program 640 based on fuzzy prototype transfer in the memory 620.

[0066] This embodiment also provides a computer-readable storage medium storing a non-overlapping cross-domain recommendation program based on fuzzy prototype transfer, which, when executed by a processor, performs the following steps: The latent features of users and projects in the source domain, as well as the latent features of users and projects in the target domain, are extracted using the latent factor model. By using the maximum entropy loss function, fuzzy clustering is performed on the latent user features and project features in the source domain and the latent user features and project features in the target domain to generate user fuzzy fusion prototypes and project fuzzy fusion prototypes. An asymmetric dual-prototype fuzzy transfer module is constructed to align cross-domain prototypes based on the latent features of users and projects in the source and target domains, as well as the fuzzy fusion prototypes of users and projects, and to generate user prototype sets and project prototype sets. Input the user prototype set, the item prototype set, and the latent user features and item features of the target domain into the recommendation model to generate target domain recommendation results.

[0067] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method. The computer-readable storage medium may include, but is not limited to, any type of disk, including floppy disks, optical disks, DVDs, CD-ROMs, microdrives, as well as magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic cards or optical cards, nanosystems (including molecular memory ICs), or any type of medium or device suitable for storing instructions and / or data.

[0068] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0069] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0070] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0071] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0072] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0073] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0074] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, which may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0075] The foregoing description is merely an exemplary embodiment of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Those skilled in the art will readily conceive of embodiments of this disclosure upon considering the specification and practicing the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

[0076] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0077] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A non-overlapping cross-domain recommendation method based on fuzzy prototype transfer, characterized in that, include: The latent features of users and projects in the source domain, as well as the latent features of users and projects in the target domain, are extracted using the latent factor model. By using the maximum entropy loss function, fuzzy clustering is performed on the latent user features and project features in the source domain and the latent user features and project features in the target domain to generate user fuzzy fusion prototypes and project fuzzy fusion prototypes. An asymmetric dual-prototype fuzzy transfer module is constructed to align cross-domain prototypes based on the latent features of users and projects in the source and target domains, as well as the fuzzy fusion prototypes of users and projects, and to generate user prototype sets and project prototype sets. Input the user prototype set, the item prototype set, and the latent user features and item features of the target domain into the recommendation model to generate target domain recommendation results.

2. The non-overlapping cross-domain recommendation method based on fuzzy prototype transfer as described in claim 1, characterized in that, The step of performing fuzzy clustering on the latent user and item features of the source domain and the latent user and item features of the target domain using the maximum entropy loss function to generate user fuzzy fusion prototypes and item fuzzy fusion prototypes also includes: Initialize prototypes for the source and target domains respectively to obtain the initial user prototype and the initial project prototype for the source and target domains. Calculate the membership distribution based on the initial user prototype and project prototype in the source and target domains, and the latent user and project features in the source and target domains. By constraining the membership distribution using the maximum entropy loss function, the initial user prototype and initial project prototype in the source and target domains are optimized. When the preset conditions are met, the optimized user initial prototype and project initial prototype of the source domain and target domain are the user fuzzy fusion prototype and project fuzzy fusion prototype, respectively.

3. The non-overlapping cross-domain recommendation method based on fuzzy prototype transfer as described in claim 2, characterized in that, The method of calculating the membership distribution based on the maximum entropy loss function and the similarity between the latent features of users and projects and the prototype also includes: The formula for calculating the membership degree distribution is as follows: ； in, It is the first in the source domain i The user on the first k The membership degree of each user, with a value range of [0,1]; Source Domain i The latent feature vector of each user; It is the source domain number k Initial prototype vectors for each user; It is the source domain number l The initial prototype vectors of each user; sim() is the similarity measurement function; τ is the temperature parameter (τ>0).

4. The non-overlapping cross-domain recommendation method based on fuzzy prototype transfer as described in claim 2, characterized in that, The method of constraining the membership distribution by the maximum entropy loss function to optimize the user initial prototype and project initial prototype in the source and target domains also includes: Based on the membership distribution, determine the negative similarity between the latent features of users and projects and their respective prototypes, and the entropy of the membership distribution is maximized, and jointly construct the maximum entropy loss function; Calculate the gradient of the maximum entropy loss function, and iteratively optimize the user initial prototype and the project initial prototype through backpropagation.

5. The non-overlapping cross-domain recommendation method based on fuzzy prototype transfer as described in claim 1, characterized in that, The asymmetric dual-prototype fuzzy transfer module, which aligns cross-domain prototypes based on latent user and project features in the source and target domains, as well as user and project fuzzy fusion prototypes, and generates user and project prototype sets, further includes: Calculate the first minimum distance between the source domain user fuzzy fusion prototype and the target domain user fuzzy fusion prototype, and the second minimum distance between the source domain item fuzzy fusion prototype and the target domain item fuzzy fusion prototype; Based on the first minimum distance and the second minimum distance, the loss function of the asymmetric dual-prototype fuzzy transfer module is constructed to align the cross-domain prototype; The aligned cross-domain prototypes are integrated and redundancy-removed to form user prototype sets and project prototype sets.

6. The non-overlapping cross-domain recommendation method based on fuzzy prototype transfer as described in claim 5, characterized in that, The loss function of the asymmetric dual-prototype fuzzy transfer module is: ； in, and A prototype for fuzzy fusion of source domain users and projects. and A prototype for fuzzy fusion of target domain users and projects.

7. The non-overlapping cross-domain recommendation method based on fuzzy prototype transfer as described in claim 1, characterized in that, The step of inputting the user prototype set, the item prototype set, and the latent user and item features of the target domain into the recommendation model to generate target domain recommendation results also includes: Input the user prototype set, the item prototype set, the user latent features and the item latent features of the target domain into the recommendation model, and calculate the similarity between the user prototype set and the user latent features, as well as the similarity between the item prototype set and the item latent features. Based on the similarity between the user prototype set and the user latent features, and the similarity between the item prototype set and the item latent features, generate user prototype-level representations and item prototype-level representations in the target domain. The user prototype-level representation in the target domain is matched with the project prototype-level representation to obtain the recommendation list and generate the target domain recommendation results.

8. A non-overlapping cross-domain recommendation device based on fuzzy prototype transfer, characterized in that, include: The feature extraction module is configured to extract latent user features and project features from the source domain, as well as latent user features and project features from the target domain, using a latent factor model. The fuzzy clustering module is configured to perform fuzzy clustering on the latent features of users and items in the source domain and the latent features of users and items in the target domain using the maximum entropy loss function, generating user fuzzy fusion prototypes and item fuzzy fusion prototypes. The fuzzy migration module is configured to build an asymmetric dual-prototype fuzzy migration module, which aligns cross-domain prototypes based on the latent features of users and projects in the source and target domains, as well as the fuzzy fusion prototypes of users and projects, and generates user prototype sets and project prototype sets. The recommendation module is configured to input user prototype sets, item prototype sets, and latent user and item features of the target domain into the recommendation model to generate target domain recommendation results.

9. A non-overlapping cross-domain recommendation device based on fuzzy prototype transfer, characterized in that, It includes at least one processing unit and at least one storage unit, wherein the storage unit stores a computer program that, when executed by the processing unit, causes the processing unit to perform the steps of the non-overlapping cross-domain recommendation method based on fuzzy prototype transfer as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, It stores a computer program executable by a non-overlapping cross-domain recommendation device based on fuzzy prototype transfer. When the computer program is run on the non-overlapping cross-domain recommendation device based on fuzzy prototype transfer, the non-overlapping cross-domain recommendation device based on fuzzy prototype transfer performs the steps of the non-overlapping cross-domain recommendation method based on fuzzy prototype transfer according to any one of claims 1 to 7.

Images