Campus Product Selection Method and Platform Based on User Data Analysis

Abstract

This invention belongs to the field of product recommendation and discloses a method and platform for selecting campus products based on user data analysis. The method includes: dividing students into multiple small groups; generating a normalized consumption behavior matrix for each small group based on the consumption behavior data of its members; obtaining a collective consumption preference model for each small group based on the normalized consumption behavior matrix; obtaining a dynamic demand model for each small group based on the collective consumption preference model; obtaining individual demand models for each member of each small group based on the dynamic demand model; generating a recommendation model based on the dynamic demand model and individual demand models; using the recommendation model to score and rank products in a candidate product set to obtain a recommendation list; and continuously optimizing the dynamic demand model and individual demand model based on member feedback data. This invention achieves a dual response of recommendation results to students' social groups and real-time needs.

Description

Technical Field

[0001] This invention relates to the field of product recommendation, and more particularly to a method and platform for selecting campus products based on user data analysis. Background Technology

[0002] In the current campus consumption environment, with the widespread adoption of e-commerce platforms, an increasing number of on-campus product recommendation systems have emerged. These systems analyze data such as students' historical shopping records, click behavior, and browsing habits to provide product recommendations based on individual preferences. These recommendation systems typically rely on user-based collaborative filtering, content-based recommendation algorithms, or hybrid recommendation algorithms, attempting to generate personalized recommendations for users through historical data and similarity models. However, these existing technologies have many limitations.

[0003] First, traditional recommendation systems primarily rely on individual user behavior data, neglecting the unique environment of students on campus. On a university campus, student consumption behavior is influenced by multiple factors, including not only individual needs but also their collective preferences within social groups. Campus life is typically highly group-oriented; for example, there is frequent interaction between small groups such as dormitory members, club members, classes, or study groups, and group activities also affect individual consumption decisions. For instance, dormitory members might buy shared household items or snacks together, while members of a study group might purchase study-related tools or books together. These social group-based consumption preferences are often overlooked in traditional recommendation systems, resulting in recommendations that fail to reflect users' true needs within the campus environment. Traditional personalized recommendation systems cannot effectively capture the group consumption characteristics of students, making it difficult to achieve recommendation optimization based on group behavior.

[0004] Secondly, the campus environment is highly dynamic. Students have different needs at different times of the day and in different settings; for example, they might need study tools in the library, while in the cafeteria they might prefer food. However, existing recommendation systems often fail to dynamically adjust to the real-time context of the user, relying solely on historical data to make static recommendations, lacking the ability to respond to users' immediate needs. This static recommendation mechanism is clearly inadequate in the context of a campus environment where student behavior is constantly changing and needs are immediate. Especially with the increasing prevalence of the Internet of Things and smart campus environments, campus facilities (such as libraries, cafeterias, and dormitories) can already obtain real-time information about students' activities through sensors and other technologies, but existing recommendation systems have not yet fully utilized this real-time data to optimize recommendation results.

[0005] Therefore, existing campus product recommendation systems have the following main problems:

[0006] 1. Ignoring the collective consumption behavior of small groups: The existing system is mainly based on the historical data of individual users and fails to effectively utilize the consumption preferences and behavioral patterns of the small groups to which students belong on campus, resulting in inaccurate recommendation results.

[0007] 2. Lack of responsiveness to real-time scenarios: Most recommendation systems are based on static data and fail to dynamically adjust recommended content based on the real-time scenarios in which students are on campus (such as libraries, dormitories, canteens, etc.), thus failing to meet the changing needs of students in different scenarios.

[0008] 3. Insufficient timeliness of recommendations: The existing system cannot respond in a timely manner to changes in students' needs at different times and under different activity states, resulting in recommendations that often lag behind the actual needs of users.

[0009] These issues significantly reduce the applicability of existing recommendation systems in campus settings. Therefore, how to dynamically optimize product recommendations based on the consumption behavior of small groups in a campus social environment and the real-time needs of students has become an urgent problem to be solved. Summary of the Invention

[0010] The purpose of this invention is to disclose a method and platform for optimizing campus products based on user data analysis, thereby solving the technical problems mentioned in the background.

[0011] To achieve the above objectives, the present invention adopts the following technical solution:

[0012] On the one hand, the present invention provides a method for optimizing campus products based on user data analysis, including:

[0013] Step 1: Divide students into multiple small groups and generate a normalized consumption behavior matrix for each small group based on the consumption behavior data of its members.

[0014] Step 2: Obtain the collective consumption preference model corresponding to the small group based on the normalized consumption behavior matrix;

[0015] Step 3: Obtain the dynamic demand model of small groups based on the collective consumption preference model;

[0016] Step 4: Obtain the individual demand models of the members in the small group based on the dynamic demand model of the small group;

[0017] Step 5: Generate a recommendation model based on the dynamic demand model of small groups and the demand model of individuals. Use the recommendation model to score and sort the products in the candidate product set to obtain a recommendation list.

[0018] Step 6: Continuously optimize the dynamic demand model of the small group and the individual demand model based on the feedback data of members on the products;

[0019] Students are divided into several small groups, including:

[0020] S11, acquire browsing history data, purchase history data, activity participation data, and social interaction data;

[0021] S12 merges browsing history data, purchase history data, activity participation data, and social interaction data into a multi-source behavior matrix;

[0022] S13, construct a weighted adjacency matrix based on the elements in the multi-source behavior matrix;

[0023] S14 uses a dynamic neighborhood expansion algorithm to process the weighted adjacency matrix and divides students into multiple small groups.

[0024] On the other hand, the present invention provides a campus product selection platform based on user data analysis, including a segmentation module, a first acquisition module, a second acquisition module, a third acquisition module, a recommendation module, and an optimization module;

[0025] The segmentation module is used to divide students into multiple small groups and generate a normalized consumption behavior matrix for each small group based on the consumption behavior data of its members.

[0026] The first acquisition module is used to obtain the collective consumption preference model corresponding to the small group based on the normalized consumption behavior matrix.

[0027] The second acquisition module is used to acquire the dynamic demand model of small groups based on the collective consumption preference model;

[0028] The third acquisition module is used to acquire the individual demand models of members in a small group based on the dynamic demand model of the small group.

[0029] The recommendation module is used to generate a recommendation model based on the dynamic demand model of small groups and the demand model of individuals. The recommendation model is used to score and rank the products in the candidate product set to obtain a recommendation list.

[0030] The optimization module is used to continuously optimize the dynamic demand model of small groups and individual demand models based on members' feedback data on products;

[0031] Students are divided into several small groups, including:

[0032] S11, acquire browsing history data, purchase history data, activity participation data, and social interaction data;

[0033] S12 merges browsing history data, purchase history data, activity participation data, and social interaction data into a multi-source behavior matrix;

[0034] S13, construct a weighted adjacency matrix based on the elements in the multi-source behavior matrix;

[0035] S14 uses a dynamic neighborhood expansion algorithm to process the weighted adjacency matrix and divides students into multiple small groups.

[0036] Beneficial effects:

[0037] The main innovations of this invention are reflected in the following aspects:

[0038] 1. Small Group Behavioral Preference Analysis: By analyzing the collective behavioral patterns of students' small groups, the system can provide more accurate product recommendations for the groups, solving the problem that existing systems only focus on individual historical behavior and ignore the consumption characteristics of the group.

[0039] 2. Real-time scene awareness and dynamic recommendation: By acquiring students' real-time activity data through IoT devices and combining it with scene awareness algorithms, the system can dynamically adjust the recommended content according to the different scenes in which the students are located, solving the problem that existing systems cannot respond to immediate needs.

[0040] 3. Demand Priority Ranking and Joint Decision-Making: This invention introduces a demand priority ranking mechanism in the recommendation generation process, which can intelligently decide to prioritize pushing products that meet the current needs in different scenarios, while taking into account the collective preferences of small groups, thereby improving the personalization and real-time response capabilities of the recommendations.

[0041] Through these innovations, this invention not only solves the problems of accuracy and timeliness in product recommendations for campus environments in existing technologies, but also achieves a dual response of recommendation results to students' social groups and real-time needs, making the recommendation system more practically valuable in campus environments. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of a campus product optimization method based on user data analysis according to the present invention.

[0044] Figure 2 This is a schematic diagram of a campus product optimization platform based on user data analysis according to the present invention. Detailed Implementation

[0045] The technical solutions of 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0046] This invention proposes an intelligent product recommendation platform based on small group preference analysis and real-time campus facility perception, aiming to provide more accurate and dynamic product recommendations for specific needs in the campus environment. By analyzing small group behavior and sensing campus facility data in real time, this platform effectively addresses the limitations of personalized recommendation systems in a campus environment, improving the accuracy and real-time performance of recommendation results.

[0047] First, this invention addresses the shortcomings of existing recommendation systems in analyzing group consumption by introducing a small group behavior analysis module. Specifically, the system can identify small groups of students on campus and uncover their collective preferences by analyzing the shared behaviors and consumption records of group members. In this way, the system can provide students with product recommendations that are more relevant to their social environment, especially items suitable for shared use by the group (such as dormitory supplies and equipment for group activities). This recommendation method based on small group preferences effectively improves the relevance of recommendation results and user satisfaction.

[0048] Secondly, this invention addresses the lack of responsiveness to dynamic scenarios in traditional systems by incorporating real-time sensing of campus facilities. The system can acquire real-time student activity data from IoT devices within the campus (such as library check-in systems, cafeteria card-swiping systems, and dormitory electricity usage data), and dynamically adjust recommended content based on this data. For example, when students are in the library, the system will prioritize recommending study tools and books, while in the cafeteria, it will suggest drinks or healthy foods. Through this scenario-aware mechanism, the system can provide more timely recommendations based on students' immediate needs, significantly improving the accuracy of recommendations and user experience.

[0049] like Figure 1 As shown in one embodiment, the present invention provides a method for optimizing campus products based on user data analysis, comprising:

[0050] Step 1: Divide students into multiple small groups and generate a normalized consumption behavior matrix for each small group based on the consumption behavior data of its members.

[0051] Step 2: Obtain the collective consumption preference model corresponding to the small group based on the normalized consumption behavior matrix;

[0052] Step 3: Obtain the dynamic demand model of small groups based on the collective consumption preference model;

[0053] Step 4: Obtain the individual demand models of the members in the small group based on the dynamic demand model of the small group;

[0054] Step 5: Generate a recommendation model based on the dynamic demand model of small groups and the demand model of individuals. Use the recommendation model to score and sort the products in the candidate product set to obtain a recommendation list.

[0055] Step 6: Continuously optimize the dynamic demand model of the small group and the individual demand model based on the feedback data of members on the products.

[0056] Preferably, students are divided into several small groups, including:

[0057] S11, Obtain browsing history data D B Purchase record data D P Activity participation data D A and social interaction data D SI ;

[0058] Browsing history data D B This refers to students' browsing history on campus e-commerce platforms, the school website, or other online learning platforms. Each student's browsing behavior data can be represented by access records, including the products, pages, or course content viewed by the student.

[0059] Purchase record data D P Students' shopping records on campus e-commerce platforms, including the products they purchased and the frequency of their purchases.

[0060] Activity participation record data D A Public activity data involving students, including lectures, social events, online courses, etc. This can be represented by activity participation records (such as attendance, frequency of participation, etc.).

[0061] Social interaction data D SI This includes student interaction data on campus social media platforms. Examples include student friendships, likes, and comments on campus social media platforms.

[0062] Before using the above data, authorization from the data management party is obtained. The data management party first performs anonymization processing, and then authorizes the use of the data obtained after anonymization. Therefore, the data obtained during the reanalysis of this invention does not involve specific individual information, thereby achieving the effect of protecting student privacy.

[0063] S12, D B D P D A and D SI Merged into a multi-source behavior matrix D M D M ={D B D P D A D SI};

[0064] S13, based on D M The elements in the matrix are used to construct a weighted adjacency matrix A;

[0065]

[0066] A ab This represents the weighted association value between student a and student b across four dimensions of data. α1, α2, α3, and α4 are the weight coefficients for each dimension, used to control the importance of different data points in clustering, and are automatically determined through experience or training data; A ab For elements in A;

[0067] Each student's social relationships, dormitory, coursework, and activity behavior can be viewed as data from different dimensions.

[0068] S14 uses the dynamic neighborhood expansion algorithm to process A, dividing students into multiple small groups.

[0069] This algorithm starts with each student and expands their neighborhood layer by layer based on the value of A, detecting the correlation of edges during the expansion process until a preset group density threshold τ is met. The result is the generation of multiple small groups, each with a unique ID. τ controls the conditions for group formation.

[0070] The computation process of the Dynamic Neighborhood Expansion (DNE) algorithm:

[0071] enter:

[0072] Weighted adjacency matrix A: This represents the multidimensional correlation value between student a and student b.

[0073] Threshold θ: Minimum correlation value required between students.

[0074] Maximum expansion steps k max : Limit the step size for neighborhood expansion.

[0075] step:

[0076] Initialize the ungrouped student set and empty group collection

[0077] from Randomly select a student 'a' as the seed point to initialize the community. Remove

[0078] right For member b, calculate its neighborhood. Will Students from the middle school joined and from Removed from the middle.

[0079] Repeat the expansion until... Unable to expand further or reach the maximum step size k max .

[0080] Will join in And continue from Select a new seed point and repeat the above process.

[0081] Output:

[0082] The system outputs a set of N small groups. Each ID N This represents the ID of a small group, whose members are automatically identified by a dynamic neighborhood expansion algorithm.

[0083] Using the DNE algorithm, the system automatically identifies multiple small groups, forming N group IDs. G ={ID1,ID2,...,ID N ID N This represents the ID of the Nth group.

[0084] Preferably, a normalized consumption behavior matrix for the small group is generated based on the consumption behavior data of its members, including:

[0085] S21, Collect consumption behavior data of members within the small group; consumption behavior data includes purchase records P it Browsing History B it Shopping cart record C it ;

[0086] P it B represents member i's purchase record at time t; it This represents member i's browsing history at time t; C itThis represents the items in member i's shopping cart at time t;

[0087] S22 integrates the consumption behaviors of all members within a small group into a behavior matrix, where each member's consumption behavior is represented by a time series. For the small group G... k Its consumption behavior matrix at time t Represented as:

[0088] B Gk (t)={P it B it C it |i∈G k}

[0089] S23, normalize the consumption behavior matrix within a time window of length T to obtain the small group G. k Normalized consumer behavior matrix

[0090]

[0091] w t The weight is for time t.

[0092] Since the behavior of different members may vary greatly in terms of time, quantity, etc., the consumption behavior of each member is weighted according to the time window to normalize the consumption frequency, amount, etc., and eliminate the scale difference between different behaviors.

[0093] A time window of length T can be a recent natural time period, such as the most recent day, week, or month.

[0094] w t The weight for time t can be dynamically adjusted based on the consumption frequency of the small group at time t, for example... Where f t This represents the consumption frequency at time t, ensuring that behaviors during high-frequency time periods have higher weights.

[0095] Preferably, step 2 includes:

[0096] S31, based on Calculate the behavioral similarity among members within a small group using a pre-defined similarity calculation formula:

[0097]

[0098] B i and B j Let D represent the normalized consumption behavior vectors of members i and j, respectively. ijγ represents the social distance between member i and member j in a social network, and γ is the weight that controls the influence of social distance on behavioral similarity.

[0099] Since the consumption behaviors of members within a small group share both commonalities and differences, it is first necessary to calculate the similarity of consumption behaviors among members within the small group. To improve the accuracy of the calculation results, social distance information is further introduced, so that the similarity not only considers consumption behavior but also incorporates the influence of social relationships on consumption decisions.

[0100] This calculation method allows the present invention to simultaneously consider the impact of consumer behavior and social interaction on small group consumer behavior.

[0101] Members' consumption behavior vectors are normalized by frequency and amount to eliminate scale differences, as shown in the formula:

[0102] In some embodiments, social distance can be calculated based on a set of friends:

[0103] The distance between members i and j on the social network, such as the number of mutual friends:

[0104]

[0105] in:

[0106] F(i): The set of friends of member i;

[0107] F(j): The set of friends of member j;

[0108] |F(i)∩F(i)|: The number of mutual friends of member i and member j.

[0109] S32, Building the Model

[0110]

[0111] This model not only aggregates the commonalities of group members through similarity measurement, but also reflects the dynamic changes in consumption behavior through a time-weighted mechanism.

[0112]

[0113] w i G represents the weight of member i in the group. Members with higher behavioral similarity will have a greater weight in the collective consumption preference model, so that the final model can represent the preferences of most members in the group. k | represents G k The number of members in;

[0114] Because consumer behavior changes over time, the model needs to consider the time factor. Therefore, a time-weighted adjustment mechanism is proposed, assigning higher weight to recent consumer behavior. The time-weighted function f(t) is defined as:

[0115] f(t) = e -λ1(T-t)

[0116] f(t) represents the time-weighted function, where t is the time when the consumption behavior occurs, used to control the impact of time on consumption behavior; λ1 represents the time decay coefficient; λ1 gives the most recent behavior a higher influence through the decay function, ensuring that the model can dynamically adapt to the recent behavior of the group's members.

[0117] It can capture the shared consumption interests among members of small groups.

[0118] S33, Construct the differential regularization term R Gk :

[0119]

[0120] B avg It is the average behavior vector of members within the group;

[0121] To prevent the consumption behavior of some members from deviating excessively from that of other members and affecting the accuracy of the collective preference model, this invention designs a difference regularization term. Difference Regularization Term This is used to limit the differences between members and the group, so that collective consumption preferences can more accurately represent the needs of the majority of the group members. Punish members whose behavior deviates significantly from the group average to ensure that the model focuses on common group behaviors.

[0122] S34, used right Optimize the model to obtain the final collective consumption preference model.

[0123]

[0124] β is the first regularization coefficient, used to control the impact of the variance regularization term on the model.

[0125] In this way, the optimized model It can more balancedly reflect the commonalities and individualities of group members, and avoid the model being biased towards the abnormal behavior of individual members.

[0126] This model will serve as input for subsequent calculations, ensuring that the recommendations meet the majority of the group's needs and dynamically respond to recent behavioral changes among group members.

[0127] Preferably, step 3 includes:

[0128] S41, Extract feature vector F from the behavioral data of scene C. C F C ={f1,f2,...,f n}, f v Let v represent a feature of behavior v in scenario C, where v∈[1,n]; n represents the total number of behavior types;

[0129] The attributes of the scene include campus scene information, equipment usage information, and activity status information;

[0130] Campus scene information could include the library, dormitory, classroom, etc.

[0131] Device usage information can be from mobile phones, computers, etc.

[0132] Activity status information can include activities such as studying and shopping.

[0133] f v These can be metrics such as product browsing frequency, time spent on shopping pages, and product click-through rate, which correspond to browsing behavior, time spent on shopping pages, and click behavior, respectively.

[0134] S42, define a scene weighting function g(C) to transform scene features into weights for model adjustment:

[0135] g(C) = W C ·F C

[0136] W C This is a weight vector that reflects the weights of features in different scenarios;

[0137] W C The value can be learned from historical data or set manually to ensure that the adjustment of the preference model is appropriate for different scenario characteristics.

[0138] S43, a model of collective consumption preferences for small groups based on the scenario weighting function g(C). Perform weighted adjustments to generate the model. The formula is as follows:

[0139]

[0140] This formula allows the small group's long-term preference model to be dynamically adjusted based on behavioral characteristics in the current scenario, ensuring that the small group's dynamic demand model can adapt to the immediate needs of the current scenario. For example, in a library scenario, recommended products might lean towards school supplies, while in a dormitory scenario, they might lean towards daily necessities.

[0141] S44, Define the historical preference regularization term RH :

[0142]

[0143] This regularization term is used to measure the dynamic scenario requirement model. With long-term consumption preference model The degree of difference. This avoids the model from relying too much on immediate scenario requirements and ensures a balance between scenario requirements and the group's long-term preferences.

[0144] S45, using R H right Optimize to obtain a dynamic demand model:

[0145]

[0146] This represents a dynamic demand model for small groups, where α1 represents the second regularization coefficient, used to control the impact of historical preferences on scenario-specific demands. This model ensures a reasonable balance between scenario-specific demands and group preferences, generating the final recommendation results.

[0147] Preferably, step 4 includes:

[0148] S51, Construct an individual consumption preference model P i :

[0149]

[0150] w t It is the weight of time t; H i (t) represents the consumption behavior data of member i at time t;

[0151] Consumer behavior includes member i's past behavior such as purchases, browsing, and clicks.

[0152] The model described above extracts preference features from an individual's historical behavior. It then assigns higher weight to recent consumption behavior through a time-weighted approach.

[0153] w t The design uses an exponential decay function for weighting to ensure that recent consumption behavior has a greater impact and to capture changes in an individual's current interests.

[0154] To enable individual preference models to more accurately describe long-term needs, preference features can be further refined by considering multiple dimensions (such as product type, brand, price, etc.). Individual preference vector P i It can be further extended to a feature matrix. Different rows represent preferences in different dimensions, for example:

[0155] Individual preferences in product categories

[0156] Individual brand preferences

[0157] Individual preferences in price range

[0158] This ensures that the preference model accurately captures an individual's long-term interests across multiple dimensions.

[0159] S52, Calculate P i and cosine similarity S iG :

[0160]

[0161] S iG This value is used to measure the similarity of consumption preferences between member i and their small group; the higher the value, the closer the individual's needs are to the group's needs. Based on this, the similarity value will serve as the core basis for the weighted fusion model.

[0162] S53, Constructing a weighted fusion model:

[0163]

[0164] α2=1-S iG

[0165] This represents a weighted fusion model; α² represents the degree to which individual needs deviate from group needs. If S iG When individual needs are relatively large (i.e., individual needs are relatively close to group needs), α2 is relatively small, and group needs have a larger weight in the individual needs model; conversely, if individual needs and group needs differ significantly, individual preferences will have a larger weight in the final model.

[0166] S54, Constructing Real-Time Adjustment Items for the Scene:

[0167]

[0168] f v β represents the feature of behavior v in scenario C. v f v Weighting of individual needs;

[0169] Individual needs not only depend on historical consumption behavior and dynamic group demands, but also require adjustment based on real-time behavior in the current scenario. To address this, this invention designs a scenario-based real-time adjustment term to capture specific behavioral characteristics of individuals in the current scenario and refine the demand model. Through this adjustment term, the individual demand model can more flexibly adapt to changes in the current scenario.

[0170] Can be ordered by behavior f v The weights are allocated using proportional normalization, which ensures that the sum of the contributions of all behaviors to the demand model is 1, and reflects the importance of each behavior.

[0171] S55, Construct individual needs models for members within a small group:

[0172]

[0173] This represents the individual demand model of member i.

[0174] This model adjusts the individual demand model in real time through scenario-based adjustment items, so that the recommendation results can better reflect the needs in the current scenario.

[0175] This approach effectively integrates long-term individual preferences with the dynamic needs of small groups through similarity metrics, and further adjusts individual demand models in real time using context-aware behavioral modifiers. Through this integration, individual demand models reflect both their independent needs and align with group needs, while also adapting to dynamic changes in the context. The entire solution combines long-term group preferences, historical individual behavior, and dynamic contextual features, ensuring the recommendation system provides more accurate and real-time recommendations for each user.

[0176] Preferably, step 5 includes:

[0177] S61, Calculation and Cosine similarity between them:

[0178]

[0179] S iG2 express and Cosine similarity between them;

[0180] The similarity measure S iG2 This is used to measure the degree of matching between individual needs and group needs in the current context. High similarity indicates that individual and group needs are relatively consistent, while low similarity indicates significant differences.

[0181] S62, Constructing the fusion weights:

[0182] λ2=1-S iG2

[0183] λ² represents the fusion weight; when the similarity S iG When λ2 is large, λ2 is small, indicating that individual needs and group needs are close, with group needs having a larger weight; when S iG2 When the size is small, individual needs take a larger weight.

[0184] S63, Building the Model

[0185]

[0186] This model balances individual and group needs, ensuring that recommendations satisfy individual preferences while reflecting the overall needs of the group.

[0187] S64, Construct the requirement difference regularization term:

[0188]

[0189] This represents the preference value of individual demand model i for behavior v in the scenario; Representation Model The preference value of i for behavior v.

[0190] Demand difference regularization can prevent excessive differences between individual and group demands from causing recommendation imbalances. This regularization reduces the deviation between individual and collective demands, avoiding over-reliance on group demands.

[0191] S65, Generate Recommendation Model:

[0192]

[0193] α3 is the third regularization coefficient, used to control the impact of the regularization term and ensure a reasonable balance between individual and group needs.

[0194] This represents the recommendation model.

[0195] S66, using For candidate product set C rec The products in the list are scored. For product cj, cj∈C rec CJ's recommended rating icj The calculation formula is:

[0196]

[0197] V cj The feature vector representing product cj;

[0198] Candidate Product Set C rec The products include items that are popular in an individual's history, items related to a particular scenario, and items that are commonly preferred by a group.

[0199] V cj Used to describe product attributes (such as category, price, brand, etc.).

[0200] S66, based on the recommended score for C rec The products in the list are sorted, and the top K products with the highest recommendation scores are selected to generate the final recommendation list.

[0201] This recommendation list combines individual historical preferences, group needs, and scenario characteristics to ensure that the recommendations meet individual user needs while remaining consistent with the overall consumption trends of the group, thus ensuring the accuracy and personalization of the recommendations.

[0202] Preferably, step 6 includes:

[0203] S71, Obtain member i's feedback data F on product cj. icj F icj Including F click F time and F purchase ;

[0204] F click This indicates the number of times member i clicked on product cj, representing initial interest; F time This indicates the duration member i spends browsing the product description page on cj; the longer the time, the stronger the interest. F purchase This indicates whether member i purchased the product, representing the final strong positive feedback; if the product was purchased, F... purchase The value is 1, no purchase was made, F purchase It is 0. S72, based on F icj Constructing the feedback matrix

[0205] bβ1, bβ2, and bβ3 represent F, respectively. click F time and F purchase The weights;

[0206] bβ1: Represents the weight of the click action, since a click is only a signal of initial user interest, and its weight is usually low;

[0207] β2: Represents the weight of browsing time. The longer the browsing time, the greater the user's interest in the product, and therefore the higher the weight.

[0208] β3: Represents the weight of the purchase behavior. The purchase behavior is the strongest positive feedback from users to the recommendation, and has the largest weight.

[0209] To improve recommendation performance, the model's recommendation score r needs to be increased. ij Based on actual user feedback A comparison is performed, and the model is optimized by minimizing the deviation between the two. To this end, a feedback optimization objective function L is constructed. opt .

[0210] S73, Construct the feedback optimization objective function:

[0211]

[0212] L opt This represents the feedback optimization objective function;

[0213] S74, based on L opt The recommendation model is updated using gradient descent.

[0214]

[0215] and Let represent the individual demand models before and after the update, respectively; η represents the learning rate, used to control the update step size. L represents opt right The partial derivative; and These represent the dynamic requirements models of the small group before and after the update, respectively. L represents opt right The partial derivative;

[0216] S75, Generate the final optimized individual demand model

[0217] To enable the model to flexibly respond to changes in users' short-term needs while preserving long-term preferences, a time-weighted function f(t) is introduced:

[0218] f(t7)=e -λ1(T-t7)

[0219] t7 represents the time when the feedback occurs, and λ1 represents the time decay coefficient, which is used to control the rate at which the influence of the feedback weakens over time.

[0220] f(t) is used to assign higher weights to recent feedback, enabling the model to respond promptly to changes in users' short-term needs.

[0221] Through the above feedback optimization mechanism, not only the individual demand model is improved. Optimization also requires gradually adjusting the dynamic requirements model of small groups. This ensures that long-term changes in the needs of individuals and groups are reflected in the model, thereby improving the accuracy and personalization of recommendations.

[0222] Optimize dynamic demand model Method:

[0223] Objective function:

[0224] According to L opt Define the update direction of the dynamic demand model and optimize the model by minimizing the error.

[0225] Optimization methods:

[0226] Update the dynamic demand model of the small group using gradient descent:

[0227]

[0228] η: Learning rate, which controls the update step size;

[0229] L opt right The gradient reflects the sensitivity of the dynamic demand model to feedback errors.

[0230] Gradient calculation:

[0231] gradient By using the chain rule, according to Expand:

[0232]

[0233] r icj Recommended score;

[0234] User feedback signals;

[0235] The rate of change of the recommended score with respect to the dynamic demand model.

[0236] Using the methods described above, the dynamic demand model With feedback optimization objective L opt Combine and complete the update.

[0237] The final output is the optimized individual demand model. Dynamic demand model for small groups The optimized model can generate more accurate recommendations. The system continuously improves the effectiveness of personalized recommendations through continuous updates of feedback signals.

[0238] like Figure 2As shown in one embodiment, the present invention provides a campus product selection platform based on user data analysis, including a segmentation module, a first acquisition module, a second acquisition module, a third acquisition module, a recommendation module, and an optimization module;

[0239] The segmentation module is used to divide students into multiple small groups and generate a normalized consumption behavior matrix for each small group based on the consumption behavior data of its members.

[0240] The first acquisition module is used to obtain the collective consumption preference model corresponding to the small group based on the normalized consumption behavior matrix.

[0241] The second acquisition module is used to acquire the dynamic demand model of small groups based on the collective consumption preference model;

[0242] The third acquisition module is used to acquire the individual demand models of members in a small group based on the dynamic demand model of the small group.

[0243] The recommendation module is used to generate a recommendation model based on the dynamic demand model of small groups and the demand model of individuals. The recommendation model is used to score and rank the products in the candidate product set to obtain a recommendation list.

[0244] The optimization module is used to continuously optimize the dynamic demand model of small groups and individual demand models based on members' feedback data on products;

[0245] Students are divided into several small groups, including:

[0246] S11, acquire browsing history data, purchase history data, activity participation data, and social interaction data;

[0247] S12 merges browsing history data, purchase history data, activity participation data, and social interaction data into a multi-source behavior matrix;

[0248] S13, construct a weighted adjacency matrix based on the elements in the multi-source behavior matrix;

[0249] S14 uses a dynamic neighborhood expansion algorithm to process the weighted adjacency matrix and divides students into multiple small groups.

[0250] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for optimizing campus products based on user data analysis, characterized in that, include: Step 1: Divide students into multiple small groups and generate a normalized consumption behavior matrix for each small group based on the consumption behavior data of its members. Step 2: Obtain the collective consumption preference model corresponding to the small group based on the normalized consumption behavior matrix; Step 3: Obtain the dynamic demand model of small groups based on the collective consumption preference model; Step 4: Obtain the individual demand models of the members in the small group based on the dynamic demand model of the small group; Step 5: Generate a recommendation model based on the dynamic demand model of small groups and the demand model of individuals. Use the recommendation model to score and sort the products in the candidate product set to obtain a recommendation list. Step 6: Continuously optimize the dynamic demand model of the small group and the individual demand model based on the feedback data of members on the products; Step 3 includes: S41, Extract feature vector F from the behavioral data of scene C. C F C ={f1,f2,...,f n }, f v Let v represent a feature of behavior v in scenario C, where v∈[1,n]; n represents the total number of behavior types; S42, define a scene weighting function g(C), which transforms scene features into weights for model adjustment through a weight vector; S43, based on the scenario weighting function g(C), the collective consumption preference model of small groups is weighted and adjusted to generate a new model. S44, based on the collective consumption preference model and model Error accumulation calculation analysis, designing historical preference regularization term R H ; S45, using the historical preference regularization term R H For the model Make corrections and optimizations to obtain a dynamic demand model.

2. The campus product optimization method based on user data analysis according to claim 1, characterized in that, A normalized consumption behavior matrix for the small group is generated based on the consumption behavior data of its members, including: S21, Collect consumption behavior data of members within the small group; consumption behavior data includes purchase records, browsing history, and shopping cart records; S22 integrates the consumption behavior of all members within a small group into a consumption behavior matrix; S23 will normalize the consumption behavior matrix within a time window of length T to obtain the normalized consumption behavior matrix of the small group.

3. The campus product optimization method based on user data analysis according to claim 2, characterized in that, Step 2 includes: S31, based on the normalized consumer behavior matrix, uses a pre-defined similarity calculation formula to calculate the behavioral similarity S among members within a small group. ij ; S32, Building the Model f(t)=e -λ1(T-t) w i G represents the weight of member i in the group. k | Represents a small group G k The number of members in the data, f(t) represents the time-weighted function, t is the time when the consumption behavior occurs, used to control the impact of time on consumption behavior; T is the total duration; λ1 represents the time decay coefficient; B i norm (t) is the normalized consumer behavior matrix; S ij The similarity of behavior among members within a group; S33, construct a differential regularization term based on the normalized consumer behavior vector and the average behavior vector of members within the group; S34, using a difference regularization term on the model Optimize the model to obtain the final collective consumption preference model.

4. The campus product optimization method based on user data analysis according to claim 1, characterized in that, Step 4 includes: S51, construct an individual consumption preference model P based on member i's consumption behavior data at time t. i ; S52, Calculate the individual consumption preference model P i and model cosine similarity S iG This is used to measure the similarity of consumption preferences between member i and the subgroup to which they belong; S53, Based on the individual consumption preference model P i and model A weighted fusion model is constructed through weighted fusion analysis of multi-source data; wherein the weights of the multi-source data are related to the cosine similarity S. iG Relatedly, if the weights are closest to 1, the fusion result depends more on the individual consumption preference model P. i ; S54, Construct an immediate adjustment term for the scenario based on the weights of individual needs according to the features of behavior v in scenario C. S55, based on the weighted fusion model and the scene real-time adjustment item Construct individual needs models for members within a small group.

5. The campus product optimization method based on user data analysis according to claim 4, characterized in that, Step 5 includes: S61, Calculating the individual needs model of members in a small group. and dynamic demand model Cosine similarity S between iG2 ; S62, based on cosine similarity S iG2 Construct the fusion weight λ2: S63, based on the individual needs model of members in a small group. and dynamic demand model Building Model S64, based on the individual needs model of members in a small group. and model Construct the demand difference regularization term R i ; S65, Based on the demand difference regularization term R i Generate recommendation model S66, using a recommendation model For candidate product set C rec The products in the list are scored. For product cj, cj∈C rec CJ's recommended rating icj The calculation formula is: V cj The feature vector representing product cj; S66, based on the recommended score for C rec The products in the list are sorted, and the top K products with the highest recommendation scores are selected to generate the final recommendation list.

6. The campus product optimization method based on user data analysis according to claim 5, characterized in that, Step 6 includes: S71, Obtain member i's feedback data F on product cj. icj The feedback data F icj This includes: the number of times member i clicked on product cj, the browsing time member i spent on product description page of cj, and whether member i purchased the product; S72, based on feedback data F icj The feedback matrix is ​​constructed by weighted summation. S73, Constructing the feedback optimization objective function L based on recommendation scoring and feedback matrix. opt ; S74, Optimizing the objective function L based on feedback opt The recommendation model is updated using gradient descent. S75, Generate the final optimized individual demand model 7. A campus product selection platform based on user data analysis, characterized in that: It includes a segmentation module, a first acquisition module, a second acquisition module, a third acquisition module, a recommendation module, and an optimization module; The segmentation module is used to divide students into multiple small groups and generate a normalized consumption behavior matrix for each small group based on the consumption behavior data of its members. The first acquisition module is used to obtain the collective consumption preference model corresponding to the small group based on the normalized consumption behavior matrix. The second acquisition module is used to acquire the dynamic demand model of small groups based on the collective consumption preference model; The third acquisition module is used to acquire the individual demand models of members in a small group based on the dynamic demand model of the small group. The recommendation module is used to generate a recommendation model based on the dynamic demand model of small groups and the demand model of individuals. The recommendation model is used to score and rank the products in the candidate product set to obtain a recommendation list. The optimization module is used to continuously optimize the dynamic demand model of small groups and individual demand models based on members' feedback data on products; The dynamic demand model for small groups based on the collective consumption preference model includes: S41, Extract feature vector F from the behavioral data of scene C. C F C ={f1,f2,...,f n }, f v Let v represent a feature of behavior v in scenario C, where v∈[1,n]; n represents the total number of behavior types; S42, define a scene weighting function g(C), which transforms scene features into weights for model adjustment through a weight vector; S43, based on the scenario weighting function g(C), the collective consumption preference model of small groups is weighted and adjusted to generate a new model. S44, based on the collective consumption preference model and model Error accumulation calculation analysis, designing historical preference regularization term R H ; S45, using the historical preference regularization term R H For the model Make corrections and optimizations to obtain a dynamic demand model.

Images