Clothing recommendation method and system based on user preference

Abstract

The invention relates to the technical field of computers, and discloses a costume recommendation method and system based on user preferences, and the method comprises the following steps: constructing a multi-dimensional utility target of the user preferences, the target comprising at least one target or a combination of a costume style matching degree, a diversity index and a novelty index; performing time evolution modeling on user recommendation preferences, and establishing a dynamic weight structure associated with user behavior feedback; constructing a recommendation utility function and introducing a multi-target constraint, wherein the nested structure comprises an outer-layer recommendation target constraint and an inner-layer preference evolution constraint; controlling the continuous evolution process of the user preference weight by using a partial differential equation; user short-term behavior feedback and long-term behavior trend data are fused in the recommendation process. According to the method, the multi-target recommendation utility function is introduced, and the style matching degree, diversity and novelty targets of the user are modeled in a nested optimization structure in a combined manner, so that the costume recommendation effect meeting personalized and exploratory requirements at the same time is achieved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method and system for recommending clothing based on user preferences. Background Technology

[0002] With the rapid development of e-commerce platforms and mobile applications, apparel recommendation systems have become an important tool for users to obtain personalized product information. Traditional apparel recommendation methods often employ collaborative filtering, content matching, or shallow feature modeling, primarily focusing on improving the relevance and matching degree of recommendations. However, these methods often neglect the overall performance of recommendation results in terms of style diversity, novelty, and users' exploratory needs, leading to homogenized recommended content, a monotonous user experience, and a lack of ability to stimulate user interest.

[0003] Some existing improvement methods attempt to introduce multi-objective optimization concepts, considering multiple objective factors simultaneously during the recommendation process, such as maximizing click-through rate and increasing product diversity. However, most of these methods fail to effectively coordinate the conflicting relationships between different objectives, lack structural constraints in the optimization process, and are prone to getting trapped in local optima or objective imbalances. Furthermore, the dependencies and non-independence between recommendation objectives have not been systematically modeled, making it difficult to form an optimization mechanism with overall control capabilities. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a clothing recommendation method and system based on user preferences, aiming to improve the problem that existing clothing recommendation methods are unable to simultaneously take into account users' personalized interests and the diversity and novelty of recommendation results, resulting in high repetition rate of recommended content, monotonous style, and failure to stimulate users' desire for continuous exploration.

[0005] In a first aspect, the present invention provides the following technical solution: a clothing recommendation method based on user preferences, comprising the following steps:

[0006] Construct a multidimensional utility objective for user preferences, wherein the objective includes at least one or a combination of multiple objectives among clothing style matching degree, diversity index and novelty index;

[0007] We model the temporal evolution of user recommendation preferences and establish a dynamic weight structure that is correlated with user behavior feedback.

[0008] A recommendation utility function is constructed and multi-objective constraints are introduced. A nested optimization structure is used to solve the problem. The nested structure includes an outer recommendation objective constraint and an inner preference evolution constraint.

[0009] The continuous evolution process of user preference weights is controlled using partial differential equations;

[0010] The recommendation process integrates short-term user behavior feedback and long-term behavior trend data.

[0011] A recommendation set is generated based on trend data, and clothing recommendations that meet the constraints of each objective are output.

[0012] Preferably, the multidimensional utility objective of the user preference includes the following three items:

[0013] Scoring based on the matching degree between clothing image features and user tag features;

[0014] The measure of differences in style, category, and color among clothing samples in the recommended collection;

[0015] The semantic or vector space difference measure between currently recommended clothing and the user's recent browsing or purchase history.

[0016] Preferably, the evolution process of the preference weights satisfies the following conditions:

[0017] The weights change continuously over time;

[0018] The rate of change of weights is determined by the strength of the current recommendation target's influence on the results, user behavior feedback signals, and the trend of weight changes themselves.

[0019] The system iteratively updates the weight evolution trajectory using analytical differentiation or numerical integration methods.

[0020] Preferably, the outer constraints of the nested optimization structure include recommendation quantity control and recommendation delay constraints, and the inner constraints include user preference evolution stability control term and regularization term. The recommendation preference weight update function introduces user behavior feedback variables, including click behavior, browsing duration, and purchase behavior, and models them according to two time scales: short-term feedback and long-term feedback.

[0021] Preferably, the method further includes: in order to determine the independence between different recommendation targets, the recommendation target function satisfies the orthogonality condition in the function space, and the similarity or correlation between different targets is constrained by the system to be close to zero;

[0022] After the independence determination, a reference trajectory function is constructed and the deviation between the user preference weight and the reference trajectory and the square of its derivative are integrally calculated. When the integral value exceeds the stability threshold, the system initializes or suppresses the weight state.

[0023] Preferably, the method further includes: using the alternating direction multiplier method to solve the problem, first iterating in the recommendation objective and preference evolution subproblems respectively, then coupling variables, and finally converging to a solution that satisfies optimality and constraint consistency.

[0024] Preferably, the output of the recommendation set needs to satisfy the necessary condition for optimality, including zero first-order variation of the recommendation utility function, and satisfying the KKT conditions for the existence and uniqueness of the optimal solution.

[0025] Secondly, the present invention provides the following technical solution: a clothing recommendation system based on user preferences, the system comprising:

[0026] The user behavior collection module is used to collect users' historical behavior data.

[0027] The preference modeling module is used to build multi-objective preference models for users;

[0028] The optimization and solution module is used to establish recommended optimization functions for nested structures and perform numerical solutions.

[0029] The preference weight evolution module is used to control the dynamic changes of preference weights based on user behavior feedback.

[0030] The recommendation generation module is used to output a set of clothing recommendations that meet user interests and are diverse and novel, based on trend data.

[0031] Thirdly, the invention provides the following technical solution: a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned method for recommending clothing based on user preferences.

[0032] Fourthly, the present invention provides the following technical solution: a readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-mentioned method for recommending clothing based on user preferences.

[0033] The present invention has the following beneficial effects:

[0034] 1. In this invention, by introducing a multi-objective recommendation utility function and jointly modeling the user's style matching degree, diversity and novelty objectives in a nested optimization structure, the recommendation results achieve a globally optimal trade-off among multi-dimensional performance indicators, thus achieving a clothing recommendation effect that simultaneously satisfies the needs of personalization and exploration.

[0035] 2. In this invention, by constructing a preference weight evolution model driven by user behavior feedback and using partial differential equations to control its dynamic change process, the continuous tracking of user interests over time is achieved, and the recommendation system achieves high responsiveness and long-term adaptability to changes in user behavior.

[0036] 3. In this invention, by introducing KKT conditions and a first-order variational differentiation mechanism into the recommendation optimization problem, the recommendation results satisfy the optimality and feasibility constraints, thereby achieving mathematical rigor and interpretability of the recommendation output and significantly enhancing the system's convergence and stability. Attached Figure Description

[0037] Figure 1 This is a flowchart of a clothing recommendation method based on user preferences proposed in this invention;

[0038] Figure 2 This is a module relationship architecture diagram of a clothing recommendation system based on user preferences proposed in this invention. Detailed Implementation

[0039] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0040] Example 1

[0041] Reference Figure 1 In the first embodiment of the present invention, the present invention provides a clothing recommendation method based on user preferences, comprising the following steps:

[0042] Construct multidimensional utility objectives for user preferences, including at least one or a combination of clothing style matching, diversity indicators, and novelty indicators;

[0043] We model the temporal evolution of user recommendation preferences and establish a dynamic weight structure that is correlated with user behavior feedback.

[0044] A recommendation utility function is constructed and multi-objective constraints are introduced. A nested optimization structure is used to solve the problem. The nested structure includes an outer recommendation objective constraint and an inner preference evolution constraint.

[0045] The method also includes: in order to determine the independence between different recommendation objectives, the recommendation objective function satisfies the orthogonality condition in the function space, and the similarity or correlation between different objectives is constrained by the system to be close to zero;

[0046] After determining independence, a reference trajectory function is constructed, and the deviation between the user preference weight and the reference trajectory and the square of its derivative are integrally calculated. When the integral value exceeds the stability threshold, the system initializes or suppresses the weight state.

[0047] The continuous evolution process of user preference weights is controlled using partial differential equations;

[0048] The method also includes: using the alternating direction multiplier method to solve the problem, first iterating in the recommendation objective and preference evolution subproblems respectively, then coupling variables, and finally converging to a solution that satisfies optimality and constraint consistency;

[0049] The recommendation process integrates short-term user behavior feedback and long-term behavior trend data.

[0050] A recommendation set is generated based on trend data, and clothing recommendations that meet the constraints of each objective are output.

[0051] Specifically, the determination of goal independence is performed through the following: Specifically, to determine the independence between different recommendation goals, this embodiment defines the recommendation objective function in a function space and introduces orthogonality conditions as a judgment criterion. The core objective function within the recommendation system, such as matching... diversity Novelty Theoretically, it should reflect complementary dimensions. The system evaluates the degree of independence between objective function pairs in the function space by calculating the inner product or correlation index between them. Let f be the standardized output of each objective function in the current sample space. i The similarity between targets can then be expressed as:

[0052]

[0053] The system sets a threshold of ε≈0.05, and forces a constraint Sim(f). i ,f j →0, meaning that the recommendation objectives are nearly orthogonal in terms of output performance, and the policy directions provided by each sub-objective do not overlap, thus avoiding the system misjudging that the value of the same type of recommendation is repeatedly reinforced by multiple objectives;

[0054] Once the goal independence is satisfied, the system further introduces a reference trajectory function to capture the natural evolution path of user preferences in multiple rounds of interaction; this trajectory function α ref (t) can be constructed based on the moving average or low-order fitting function of weight changes in the historical interaction window, reflecting the steady trend of the preference structure under the system's expectation. The deviation between the current preference weight trajectory α(t) and the reference trajectory will be continuously monitored. To measure the strength of the deviation, the system calculates the integral distance between the two, including the trajectory difference itself and its first derivative term:

[0055]

[0056] Here, λ is the derivative deviation adjustment factor, balancing the influence of the offset magnitude and the rate of change. This integral result reflects whether the current policy evolution deviates from the historical trajectory, and is particularly suitable for identifying abrupt policy adjustments or feedback perturbations. When the above integral result... Exceeding the preset stability threshold δ stab When this occurs, the system considers the strategy to be at risk of instability. At this point, the weight state control mechanism is triggered, including the following two methods:

[0057] The first method resets the current preference weights to the current or average stable values ​​of the reference trajectory, which is used to fall back to a controlled state.

[0058] The second method temporarily freezes the update step size of the current weight components, or applies a high damping coefficient to suppress their change amplitude. This is used to protect and repair the recommendation strategy in situations such as inconsistent multi-objective feedback information, external abnormal interference, and short-term drastic fluctuations in user behavior. Ultimately, through technical means such as target orthogonality judgment, preference trajectory monitoring, and integral deviation constraint, effective control of the multi-objective recommendation strategy in terms of both structural independence and dynamic evolution stability is achieved.

[0059] The solution was obtained using the alternating direction multiplier method. The following content is provided:

[0060] In solving the multi-objective recommendation weight optimization problem, the alternating direction multiplier method is adopted as the core optimization algorithm framework to achieve efficient joint solution of the two sub-problems: recommendation objective and user preference evolution. The original optimization problem is a joint objective problem with multiple constraint terms, which can be formalized as:

[0061]

[0062] in:

[0063] Recommendation loss is constructed based on multi-objective functions (such as matching, novelty, etc.);

[0064] Trajectory deviation and regularization term used to constrain the evolution of preference weights;

[0065] Constraints include weight normalization, nonnegativity, orthogonality approximation, and trajectory deviation control.

[0066] To reduce the overall coupling complexity of the problem, an auxiliary variable β is introduced, and the original problem is split into two subproblems:

[0067] Subproblem 1: Minimizing the recommendation objective: Using α as the optimization variable, focusing on the performance of the recommendation objective;

[0068] Subproblem 2: Preference Evolution Constraint: With β as the optimization variable, responsible for controlling the stability and regularity of the trajectory, after introducing the constraint α = β, the original problem is transformed into a form that can be solved alternately.

[0069] The solution process of the alternating direction multiplier method is as follows: let the Lagrange multiplier be λ and the penalty factor be ρ:1.

[0070] Initialize α (0) ,β (0) ,λ (0)

[0071] Recommended target subproblem (fixed β, λ):

[0072]

[0073] The preference evolution subproblem (with fixed α, λ):

[0074]

[0075] Lagrange multipliers update:

[0076] λ (k+1) =λ (k) +α (k+1) -β (k+1) ;

[0077] During the iteration process, the optimization problem at each step is simpler in structure than the original problem, exhibiting convexity or locally convexity, thus making it easier to solve. As the iteration progresses: the sub-variables α and β gradually converge, the consistency constraint α = β is gradually satisfied; the Lagrange multiplier λ converges to the equilibrium point, and finally, the system converges to a set of solutions α. * ,β*, which simultaneously satisfies the optimality of the recommendation objective and the consistency of the evolutionary constraints.

[0078] The multidimensional utility objectives of user preferences include the following three items:

[0079] Scoring based on the matching degree between clothing image features and user tag features;

[0080] The measure of differences in style, category, and color among clothing samples in the recommended collection;

[0081] The semantic or vector space difference measure between currently recommended clothing and the user's recent browsing or purchase history.

[0082] Specifically, firstly, the system extracts features from the user's static tags, including gender, age, style preference, and color preference, and embeds them into a vector representation through a feature encoding module, denoted as . Simultaneously, the image information of all clothing samples to be recommended will be processed through a visual neural network to extract mid-level features, resulting in image feature vectors. Based on this, the matching score function between the recommendation set R and the user tag vector is defined as follows:

[0083]

[0084] This function represents the average similarity in semantic space between each garment in the recommendation set and the user's static interest tags, reflecting the degree to which the recommended content matches the user's preferences. A higher score indicates that the recommendation set as a whole better matches the user's explicit preferences.

[0085] Define the difference function between any two garments in the set in three dimensions: style, category, and color, denoted as d. s (r i ,r j ),d p (r i ,r j ),d c (r i ,r j );

[0086] Where r i ,r j ∈R represents any two garments in the set, whose three-dimensional differences are aggregated into a set diversity scoring function using a weighted combination:

[0087]

[0088] Where, λ s ,λ p ,λ c ∈[0,1] represents the weight coefficients for the three dimensions of style, category, and color, satisfying λ s +λ p +λ c =1, used to control the contribution of each dimension to diversity. This function measures the visual and semantic differences of the recommendation set. A higher score indicates a richer set structure, which can reduce user visual fatigue and enhance the novelty of the recommendations;

[0089] Finally, we measure the difference between the current recommendation set and the user's historical interest trajectory. Let H be the set of the user's recent interaction behaviors. u ={h1,h2,...,h m Each historical garment sample possesses a corresponding image feature vector f. h The system defines the average semantic distance function between the recommendation set and the historical set as follows:

[0090]

[0091] Where cos(f) r ,f hLet be the standard cosine similarity function. This score measures the semantic deviation between the current recommendation and past behavior. A higher score indicates a greater difference between the recommended clothing and historical content in terms of visual style, category characteristics, or color matching, i.e., stronger novelty. After integrating the above three objective functions, the system constructs a recommendation utility functional in a weighted manner as a comprehensive evaluation index, which is used for the final ranking and selection of the recommendation set, defined as follows:

[0092]

[0093] Among them, α=(α1,α2,α3)∈[0,1] 3 The target weighting vector of the system, satisfying the constraint α1+α2+α3=1, is used to control the relative importance among preference fit, diversity, and novelty. In the specific implementation, this weight can be dynamically updated driven by user behavior characteristics; factors such as recent click density, repeat click rate, and recent activity will jointly affect the real-time adjustment of α.

[0094] The evolution of preference weights satisfies the following conditions:

[0095] The weights change continuously over time;

[0096] The rate of change of weights is determined by the strength of the current recommendation target's influence on the results, user behavior feedback signals, and the trend of weight changes themselves.

[0097] The system iteratively updates the weight evolution trajectory using analytical differentiation or numerical integration methods.

[0098] Specifically, by constructing a weight model that evolves over time, the system achieves real-time adjustment of the fusion structure of recommendation objectives. The weights are not only static hyperparameters, but also continuous variables with dynamic attributes. Their changes are influenced by the combined effects of the recommendation result structure, user behavior feedback, and the historical state of the weights. During the recommendation process, the system periodically calculates the performance of the recommendation set under multiple sub-objectives.

[0099] The first factor is that the weight value at any given time is a natural continuation of its state at the previous time, without any jumps or resets. When the system detects that the improvement in the performance index of a certain sub-objective significantly contributes to the overall recommendation quality, it will increase the weight corresponding to that object. The degree of contribution is usually difficult to judge based on experience, so the system needs to calculate the sensitivity of the recommendation utility function to each sub-objective. In this case, if the utility function is differentiable, introducing partial derivatives is necessary.

[0100]

[0101] in, Let α represent the current recommendation utility function. iIt is the weight of the i-th recommendation target. This partial derivative represents the strength of the recommendation performance in response to the weight of the current target. This partial derivative is not for gradient descent, but for perceiving which part of the recommendation structure plays a major role, i.e. how much trust is given to different targets;

[0102] The second factor comes from user behavior feedback. If the system detects that a decrease in diversity in the recommendation results leads to an increase in user bounce rate and a shorter dwell time, it concludes that diversity affects user satisfaction and thus increases its weight. This feedback signal does not have a fixed expression and can be any measurable behavioral trend, including:

[0103] The percentage increase in clicks for a specific target item;

[0104] The difference between the average time spent on the page and the historical baseline;

[0105] Rapid changes in the skip rate of a certain type of recommended content are signals that are fed into the system via the rate of change, including sliding window averages or differential values, to dynamically adjust the target emphasis.

[0106] The third factor is that if a certain weight is consistently increasing, it indicates that the target is consistently preferred. Conversely, if it fluctuates wildly, it may indicate unstable feedback signals or high noise in user behavior. Therefore, a mechanism is introduced to adjust the weight's trend. When the system detects a rapid increase in a weight over two periods without a corresponding increase in the feedback signal, it will proactively reduce the rate of increase to maintain a stable strategy. In this case, if system resources allow or high recommendation accuracy is required, the recommendation engine will use a numerical integration method to calculate the weight value. Through explicit Euler's method, its iterative form is as follows:

[0107] α i (t+Δt)=α i (t)+Δt·v i (t);

[0108] Among them, v i (t) represents the target change rate of the i-th term, which is a weighted synthesis of the three driving factors mentioned above. Whether to use integral updates depends on the balance between response speed and computational cost for the scenario. For scenarios with high-frequency fluctuations in user behavior, including short video streams, numerical integration is more robust; if the recommendation update cycle is long, fixed-step iteration can be used.

[0109] The outer constraints of the nested optimization structure include recommendation quantity control and recommendation delay constraints, while the inner constraints include user preference evolution stability control terms and regularization terms. User behavior feedback variables are introduced into the recommendation preference weight update function, including click behavior, browsing duration, and purchase behavior, and are modeled separately for short-term feedback and long-term feedback.

[0110] Specifically, the evolution function of preference weights is the core of the nested structure optimization, while the constraints on the number of recommendations and time delay control are located in the outer layer. Within each recommendation cycle, the system first generates a set of candidate recommendations based on the user context and behavioral input, and then performs fusion ranking and target combination scoring on the set according to the current recommendation preference weights α(t). The recommendation output must satisfy two outer constraints:

[0111] First, there is a constraint on the number of recommendations, which is used to limit the size of the recommendation set to prevent information overload due to too much content or reduced attractiveness due to insufficient content.

[0112] Second, there is a recommendation delay constraint, meaning that the generation and updating of the system's recommendation strategy must be completed within an acceptable time window to ensure the immediacy of the user interface response. The delay constraint can be expressed as:

[0113] T compute (α(t),x u )≤τ max ;

[0114] Among them, T compute τ represents the time consumed in calculating and sorting the current strategy. max This is the maximum response time set by the platform service level agreement. After satisfying the above outer constraints, the system enters the inner optimization stage of the nested structure. This inner optimization includes two key sub-constraints:

[0115] First, the stationarity control term for user preference evolution does not allow weights to change abruptly without significant behavioral signals. Therefore, the system introduces a smoothing term or change penalty term to limit the magnitude of weight changes over continuous time.

[0116]

[0117] This term can be directly added to the objective function or used as a constraint boundary to suppress rapid oscillations and maintain the natural evolution of the preference curve. Secondly, there's the regularization term, mainly used to prevent weights from being biased towards a single objective for a long time, thus disrupting the balance of the multi-objective recommendation structure. Commonly used terms are entropy regularization or equilibrium constraints.

[0118]

[0119] The recommendation weights evolve continuously over time, and the system uses a numerical integration method to iteratively update them, ensuring that the recommendation strategy gradually converges in the continuous time domain. After each update, the weight vector is normalized to keep its sum at 1 and to prevent deviation from the target space.

[0120] The output of the recommendation set must satisfy the necessary condition for optimality, including zero first-order variation of the recommendation utility function, and satisfy the KKT conditions for the existence and uniqueness of the optimal solution.

[0121] Specifically, firstly, the recommendation set must satisfy the condition that the first variation of the recommendation utility function is zero, to ensure that the current output reaches the optimal state under the multi-objective function. The recommendation utility function is usually a weighted combination of the objective functions, and its variation being zero means that under the current recommendation set, small perturbations no longer bring about utility improvement, thus constituting a necessary first-order condition for optimality;

[0122] Secondly, to further ensure the existence and feasibility of the optimal solution, the system introduces the KKT condition as a constraint criterion. This condition covers factors such as first-order derivative optimality, feasibility of the original constraints including the number of recommendations, project legality, weight normalization and non-negativity, complementary relaxation and multiplier non-negativity, etc. Satisfying the KKT condition means that the current recommended solution is a local optimum within the constraint domain.

[0123] Finally, in the solution process, the system employs the alternating direction multiplier method for optimization, decomposing and iteratively solving the recommendation objective and preference weight evolution problem. After each subproblem is optimized separately, variable coupling and Lagrange multiplier updates are used to gradually converge the recommendation variables to the optimal solution while satisfying constraint consistency.

[0124] Example 2:

[0125] Reference Figure 2 In a second embodiment of the present invention, the present invention provides a clothing recommendation system based on user preferences, the system comprising:

[0126] The user behavior collection module is used to collect users' historical behavior data.

[0127] The preference modeling module is used to build multi-objective preference models for users;

[0128] The optimization and solution module is used to establish recommended optimization functions for nested structures and perform numerical solutions.

[0129] The preference weight evolution module is used to control the dynamic changes of preference weights based on user behavior feedback.

[0130] The recommendation generation module is used to output a set of clothing recommendations that meet user interests and are diverse and novel, based on trend data.

[0131] Specifically, the user behavior collection module is used to collect users' historical behavior data, including click records, browsing duration, favorites, purchase behavior, bounce rate, and other interaction information. This module categorizes and stores the behavior data into short-term and long-term categories, and extracts key behavioral features as the basic input for subsequent preference modeling and dynamic feedback updates.

[0132] The preference modeling module receives feature data output from the user behavior collection module, constructs a multi-objective preference model for users, and represents users' preference tendencies in multiple dimensions such as matching, diversity, novelty, and conversion potential. The results are represented in a vectorized structure and used as input features for the recommendation optimization module.

[0133] The optimization and solution module receives the multi-objective preference model provided by the preference modeling module and, in conjunction with the system-configured recommendation objectives and constraints, establishes a nested recommendation optimization function. It iteratively optimizes the recommendation objectives and preference evolution subproblems using the alternating direction multiplier method, completing variable coupling and solving for recommendation results that satisfy the first-order variational conditions and KKT optimality conditions. Simultaneously, the results are output to the preference weight evolution module for weight adjustment.

[0134] The preference weight evolution module takes the current recommendation effect output by the optimization solution module and real-time user behavior feedback as input, and dynamically adjusts the evolution trajectory of preference weights. The feedback behavior includes short-term click fluctuations and long-term conversion trends. By introducing a trajectory stability control mechanism and orthogonality constraints, the updated preference weight results are fed back to the optimization solution module.

[0135] The recommendation generation module, based on the final convergence result of the optimization solution module, generates a set of clothing recommendations that meet the user's interest structure. The recommendation results take into account matching, diversity, and novelty, covering the richness of clothing in multiple dimensions such as quality, style, and category, thereby improving the overall user satisfaction and exploratory experience of the recommendations.

[0136] Example 3

[0137] In a third embodiment of the present invention, based on the same inventive concept, the present invention proposes a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of a user preference-based clothing recommendation method as described in the above embodiments.

[0138] Example 4

[0139] In a fourth embodiment of the present invention, based on the same inventive concept, a computer device is proposed, comprising: a processor and a memory; the processor and the memory communicate with each other; the memory is used to store instructions; the processor is used to execute the instructions in the memory to execute a user preference-based clothing recommendation method of the above embodiment.

[0140] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0141] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A clothing recommendation method based on user preferences, characterized in that, Includes the following steps: Construct a multidimensional utility objective for user preferences, wherein the objective includes at least one or a combination of multiple objectives among clothing style matching degree, diversity index and novelty index; We model the temporal evolution of user recommendation preferences and establish a dynamic weight structure that is correlated with user behavior feedback. A recommendation utility function is constructed and multi-objective constraints are introduced. A nested optimization structure is used to solve the problem. The nested structure includes an outer recommendation objective constraint and an inner preference evolution constraint. The continuous evolution process of user preference weights is controlled using partial differential equations; The recommendation process integrates short-term user behavior feedback and long-term behavior trend data. A recommendation set is generated based on trend data, and clothing recommendations that meet the constraints of each objective are output.

2. The clothing recommendation method based on user preferences according to claim 1, characterized in that, The multidimensional utility objectives of user preferences include the following three items: Scoring based on the matching degree between clothing image features and user tag features; The measure of differences in style, category, and color among clothing samples in the recommended collection; The semantic or vector space difference measure between currently recommended clothing and the user's recent browsing or purchase history.

3. The clothing recommendation method based on user preferences according to claim 1, characterized in that, The evolution of the preference weights satisfies the following conditions: The weights change continuously over time; The rate of change of weights is determined by the strength of the current recommendation target's influence on the results, user behavior feedback signals, and the trend of weight changes themselves. The system iteratively updates the weight evolution trajectory using analytical differentiation or numerical integration methods.

4. The clothing recommendation method based on user preferences according to claim 1, characterized in that, The outer constraints of the nested optimization structure include recommendation quantity control and recommendation delay constraints, while the inner constraints include user preference evolution stability control terms and regularization terms. The recommendation preference weight update function introduces user behavior feedback variables, including click behavior, browsing duration, and purchase behavior, and models them separately according to two time scales: short-term feedback and long-term feedback.

5. The clothing recommendation method based on user preferences according to claim 1, characterized in that, The method further includes: in order to determine the independence between different recommendation targets, the recommendation target function satisfies the orthogonality condition in the function space, and the similarity or correlation between different targets is constrained by the system to be close to zero; After the independence determination, a reference trajectory function is constructed and the deviation between the user preference weight and the reference trajectory and the square of its derivative are integrally calculated. When the integral value exceeds the stability threshold, the system initializes or suppresses the weight state.

6. The clothing recommendation method based on user preferences according to claim 1, characterized in that, The method further includes: using the alternating direction multiplier method to solve the problem, first iterating in the recommendation objective and preference evolution subproblems respectively, then coupling variables, and finally converging to a solution that satisfies optimality and constraint consistency.

7. The clothing recommendation method based on user preferences according to claim 1, characterized in that, The output of the recommendation set must satisfy the necessary condition for optimality, including zero first-order variation of the recommendation utility function, and satisfy the KKT conditions for the existence and uniqueness of the optimal solution.

8. A clothing recommendation system based on user preferences, characterized in that, The system for a user preference-based clothing recommendation method according to any one of claims 1-7, the system comprising: The user behavior collection module is used to collect users' historical behavior data. The preference modeling module is used to build multi-objective preference models for users; The optimization and solution module is used to establish recommended optimization functions for nested structures and perform numerical solutions. The preference weight evolution module is used to control the dynamic changes of preference weights based on user behavior feedback. The recommendation generation module is used to output a set of clothing recommendations that meet user interests and are diverse and novel, based on trend data.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a clothing recommendation method based on user preferences as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a computer program that, when executed by a processor, implements a user preference-based clothing recommendation method as described in any one of claims 1 to 7.

Images