Personalized recommendation method and system for animal husbandry bidding information

Abstract

The invention discloses a personalized recommendation method and system for animal husbandry bidding information. The method comprises the following steps: acquiring static attributes and dynamic behavior data of animal husbandry users, and constructing multi-dimensional feature vectors of the users; acquiring bid invitation information in real time and extracting structural features of the bid invitation information; based on historical bidding records and performance evaluation of the user, a benefit evaluation model is adopted to calculate benefit indexes of the user for various bid invitation items, and the indexes fuse a historical bid winning rate, an item profit rate, a performance score and a real-time capability matching degree; generating and pushing a personalized bid invitation information recommendation list through multi-level matching and sorting; and meanwhile, establishing a feedback learning mechanism, and dynamically optimizing the model and the user portrait according to the user interaction behavior. According to the method, the problems of low precision, lack of benefit guidance and poor real-time performance of existing bid invitation information pushing are solved, high-matching and high-expected-income bid invitation opportunities can be accurately recommended for animal husbandry users, and the commercial decision-making efficiency and success rate are improved.

Description

Technical Field

[0001] This invention belongs to the field of livestock bidding information management technology, and in particular relates to a personalized recommendation method and system for livestock bidding information. Background Technology

[0002] With the informatization and large-scale development of animal husbandry, various animal husbandry-related bidding projects (such as feed procurement, live animal sales, equipment import, and ranch construction) are increasing. Bidding information is usually released in a scattered manner through multiple channels such as government procurement platforms, industry websites, and enterprise service platforms, resulting in serious problems of information overload and fragmentation.

[0003] Currently, livestock farmers primarily rely on manual searches, subscriptions to general industry news, or personal connections to obtain bidding information. Existing information push services often rely on simple keyword matching or fixed categories for broadcast-style delivery, lacking deep personalization. This approach has significant drawbacks:

[0004] Low push accuracy: It cannot accurately match the user's actual production scale, breeding type, technical qualifications and bidding ability, resulting in a large amount of irrelevant information interfering with the user.

[0005] Lack of profitability orientation: It fails to consider users' historical bidding success rate, project profitability and performance capabilities, and is unable to screen out high-quality bidding opportunities that are "not only suitable, but also profitable" for users.

[0006] Passive and delayed: Users need to actively search or wait for periodic push notifications, and cannot perceive new bidding opportunities that are highly relevant to them in real time, making it easy to miss opportunities.

[0007] Ignoring dynamic preferences: Users' interests and needs change over time, and traditional static models cannot continuously learn and optimize recommendations based on user feedback.

[0008] Therefore, there is an urgent need for a solution that can deeply integrate the multidimensional characteristics of livestock users with the characteristics of bidding information to achieve intelligent, personalized, and efficient recommendations, so as to improve the efficiency and accuracy of livestock users in obtaining business opportunities. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the prior art and provide a personalized recommendation method and system for livestock bidding information, aiming to solve the problems of low accuracy, lack of benefit consideration, poor real-time performance and inability to adaptively optimize existing bidding information push methods.

[0010] In a first aspect, the present invention provides a personalized recommendation method for livestock bidding information, comprising:

[0011] Acquire user profile data of livestock users, the user profile data including static attribute data and dynamic behavior data;

[0012] The static attribute data is processed by one-hot encoding and normalization, and the dynamic behavior data is processed by calculating the weight of recent behavior through a time decay model. The processed data is then fused into vectors to obtain a multi-dimensional feature vector of the user.

[0013] From the bidding information database, historical bidding information sets and currently published bidding information associated with the user's multi-dimensional feature vector are obtained in real time. The structured features of each bidding information are extracted. The structured features include at least the bidding category, budget amount range, technical standard requirements, product delivery region, bid deadline, and the bidding party's credit rating.

[0014] Based on the user's historical bidding records and performance evaluations, a benefit assessment model is used to calculate the user's benefit index for various bidding projects. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements.

[0015] Perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information;

[0016] Based on the ranking results, a personalized bidding information recommendation list containing recommendation reasons is generated, and the personalized bidding information recommendation list is pushed to the animal husbandry user through the user's designated terminal or messaging channel. The recommendation reasons include at least a matching score, benefit forecast, and time urgency reminder.

[0017] Secondly, the present invention provides a personalized recommendation system for livestock bidding information, comprising:

[0018] The acquisition module is configured to acquire user profile data of livestock users, the user profile data including static attribute data and dynamic behavior data;

[0019] The processing module is configured to use one-hot encoding and normalization to process the static attribute data, calculate the recent behavior weights of the dynamic behavior data through a time decay model, and perform vector fusion on the processed data to obtain a multi-dimensional feature vector of the user.

[0020] The extraction module is configured to obtain, in real time, a set of historical bidding information associated with the user's multi-dimensional feature vector and currently published bidding information from the bidding information database, and extract the structured features of each bidding information. The structured features include at least the bidding category, budget amount range, technical standard requirements, product delivery region, bid deadline, and the bidding party's credit rating.

[0021] The calculation module is configured to calculate the user's benefit index for various bidding projects based on the user's historical bidding records and performance evaluations using a benefit assessment model. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements.

[0022] The matching module is configured to perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information;

[0023] The push module is configured to generate a personalized bidding information recommendation list containing recommendation reasons based on the sorting results, and push the personalized bidding information recommendation list to the livestock user through the user-specified terminal or message channel. The recommendation reasons include at least a matching score, benefit forecast, and time urgency reminder.

[0024] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the personalized recommendation method for livestock bidding information according to any embodiment of the present invention.

[0025] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the personalized recommendation method for livestock bidding information according to any embodiment of the present invention.

[0026] This application presents a personalized recommendation method and system for livestock bidding information. By constructing a multi-dimensional user feature vector that integrates static attributes and dynamic behaviors, and by performing deep, structured analysis of bidding information, it achieves precise alignment of supply and demand characteristics. Based on this, it innovatively introduces and quantifies the core indicator of "efficiency index." This index comprehensively considers a user's historical bidding success rate, project profit margin, performance reputation, and real-time capability matching degree, while also incorporating dynamic factors such as user activity and market competition intensity. This upgrades the recommendation logic from traditional "content matching" to "value and success probability prediction." Through a multi-level matching and ranking engine, this method can filter and prioritize bidding projects from massive amounts of information in real time, projects that highly match the user's production capabilities and have high expected returns. This solution effectively overcomes the shortcomings of existing technologies, such as low information push accuracy, lack of commercial benefit orientation, passive lag, and inability to adaptively optimize. It significantly improves the efficiency, accuracy, and commercial success rate of livestock users in obtaining high-quality business opportunities, achieving a leap from "information recommendation" to "business decision empowerment." Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart illustrating a personalized recommendation method for livestock bidding information provided in an embodiment of the present invention;

[0029] Figure 2 This is a structural block diagram of a personalized recommendation system for livestock bidding information provided in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0032] Please see Figure 1 The diagram illustrates a flowchart of a personalized recommendation method for livestock bidding information according to this application.

[0033] like Figure 1 As shown, the personalized recommendation method for livestock bidding information specifically includes the following steps:

[0034] Step S101: Obtain user profile data of livestock users, the user profile data including static attribute data and dynamic behavior data.

[0035] In this step, the static attribute data includes the user's production scale, breeding type, enterprise qualification certification information, and geographical location; the dynamic behavior data includes the user's historical bidding records, bidding information browsing records, collection records, and performance evaluation feedback on successful bids.

[0036] Obtain user profile data for livestock users, including:

[0037] By connecting to the livestock production management system or IoT devices, the user's basic production data is automatically obtained. The basic production data includes pasture area, number of animals in stock, average annual output, main livestock species and growth cycle data.

[0038] By connecting to the enterprise credit information platform and the government's public data interface, we can verify and capture users' qualification certification information, including animal disease prevention and control qualification certificates, environmental impact assessment approvals, organic product certification certificates, green food labels, and enterprise credit rating reports.

[0039] By analyzing users' operation logs on the bidding platform, we can extract users' historical bidding records, including bidding project types, bidding frequency, success rate, preferred bidding regions, commonly used bidding strategies, and analysis of reasons for bid rejection.

[0040] Step S102: The static attribute data is processed by one-hot encoding and normalization, the dynamic behavior data is processed by calculating the recent behavior weights through a time decay model, and the processed data is fused into vectors to obtain the user's multi-dimensional feature vector.

[0041] In this step, the static encoding vector, static normalized vector, and dynamic weighted vector are merged into a single user multi-dimensional feature vector.

[0042] The sparse vector generated by one-hot encoding, the normalized continuous value vector, and the time-weighted dynamic behavior feature vector are directly concatenated along the feature dimension.

[0043] For example, suppose the result after processing is:

[0044] Static classification encoding vector: 50 dimensions;

[0045] Static continuous normalized vector: dimension 10;

[0046] Dynamic behavior weighted vector: dimension 30;

[0047] The total dimension of the concatenated user feature vector is 50 + 10 + 30 = 90.

[0048] To ensure that the feature vectors of all users have the same dimension and that the corresponding positions have the same meaning, a global feature mapping table needs to be predefined. This table fixes the index position of each feature in the final vector.

[0049] The fused high-dimensional feature vectors can be stored in an efficient numerical array format (such as NumPy arrays or PyTorch Tensors) and then stored in a vector database (such as Milvus, FAISS) or a traditional database for rapid retrieval and calculation by the subsequent real-time similarity matching module.

[0050] In one specific embodiment, static attribute data typically includes categorical variables (such as farming type) and continuous variables (such as stock quantity), which need to be processed separately.

[0051] One-hot encoding of categorical variables:

[0052] Applicable fields: type of farming, type of enterprise qualification (e.g., whether it has organic certification), main species farmed, etc.

[0053] Implementation process:

[0054] First, determine all possible categories based on user data across the entire platform. For example, possible values ​​for farming types are: {dairy farming, beef cattle farming, sheep farming, poultry farming, integrated farming}.

[0055] For a specific user, if their farming type is "dairy farming", then it is converted into a sparse vector containing only 0s and 1s using one-hot encoding. The length of this vector is equal to the total number of categories (5 in this case), with a value of 1 only at the index position corresponding to "dairy farming", and 0 at the other positions.

[0056] For example: "Dairy cow farming" -> [1,0,0,0,0]; "Meat sheep farming" -> [0,0,1,0,0].

[0057] A similar approach is used for enterprise qualifications. Each qualification (such as "organic certification" or "green food label") is treated as an independent Boolean feature. If a user possesses a qualification, the feature value is 1; otherwise, it is 0. For example, a user who possesses both "organic certification" and "AAA credit" might have an encoding vector of [1, 0, 1, ...] in this dimension.

[0058] Normalization of continuous variables:

[0059] Applicable fields: numerical features such as pasture area, number of animals in stock, and average annual output.

[0060] Implementation process:

[0061] Minimum-maximum normalization or Z-score standardization is used.

[0062] Taking min-max normalization as an example:

[0063] Calculate in advance the maximum and minimum values ​​(max and min) of a certain continuous feature for all users across the platform.

[0064] For a user's specific value, value, it is scaled to the range [0,1] using the formula (value-min) / (max-min).

[0065] For example, the minimum number of animals in stock across the entire platform is 100, and the maximum is 10,000. If a user has 2,000 animals in stock, the normalized value is (2000-100) / (10000-100)≈0.192.

[0066] This processing eliminates the impact of different units (such as "mu" and "head") and order of magnitude differences on distance calculation, enabling the model to treat different features equally.

[0067] Processing of dynamic behavioral data: Calculation of time decay weights

[0068] Dynamic behavioral data is time-sensitive, and recent behavior better reflects a user's current interests and state. Therefore, a time decay function is introduced to assign different weights to behaviors at different points in time.

[0069] Determine the time decay function:

[0070] The commonly used exponential decay function is: weight(t) = exp(-λ*Δt).

[0071] Where Δt represents the interval between the current time and the time the behavior occurred (usually in days), and λ is the decay rate parameter (λ>0), which controls the rate at which the weight decays over time. The larger the value of λ, the faster the weight of the old behavior decreases.

[0072] For example, if λ = 0.01, the weight of an action taken 30 days ago is approximately exp(-0.01*30) ≈ 0.74; the weight of an action taken 90 days ago is approximately exp(-0.01*90) ≈ 0.41.

[0073] Applying decay weights to behavioral statistics:

[0074] For behavioral characteristics such as "bidding frequency" and "number of times browsing a certain category" statistically derived from historical logs, instead of simple counting, a weighted sum is calculated.

[0075] Step S103: Obtain in real time the historical bidding information set and currently published bidding information associated with the user's multi-dimensional feature vector from the bidding information database, and extract the structured features of each bidding information. The structured features include at least the bidding category, budget amount range, technical standard requirements, product delivery region, bid deadline, and the bidding party's credit rating.

[0076] In this step, the official platform's real-time data stream is obtained by calling official data interfaces (such as WebService APIs or push interfaces that follow specific data specifications) of national and provincial electronic bidding and tendering public service platforms and government procurement websites. This involves establishing long connections or conducting high-frequency short polling to achieve data retrieval within seconds to minutes after the tender announcement is published. The system will then parse the raw data packets returned by the interfaces (usually in JSON or XML format).

[0077] Targeted crawling of industry websites: For key livestock industry portals and industry association information platforms, deploy web crawler clusters configured with intelligent parsing rules. These crawlers can simulate browser behavior, bypass simple anti-crawling mechanisms, and accurately locate and extract the core content areas of bidding announcements based on webpage structure templates, enabling scheduled (e.g., every 10 minutes) or incremental crawling.

[0078] Aggregation of publicly available government data: Access to government information disclosure data interfaces of agricultural and rural departments and animal husbandry and veterinary stations at all levels, or structured crawling of their publicly available bidding and announcement columns to ensure coverage of local and professional bidding projects.

[0079] The collected raw bidding information will first enter a buffer queue. The system does not perform in-depth processing on all information, but first performs a quick coarse screening.

[0080] The coarse screening rules are directly derived from the key dimensions of the user's multi-dimensional feature vector. For example:

[0081] By leveraging the user's geographic location, bids can be filtered to include products delivered within the user-defined radius (e.g., 300 kilometers) or within the reach of logistics.

[0082] By utilizing users' farming types and main breeds, we can filter bidding projects that contain relevant keywords such as "dairy cows," "beef cattle," "feed," "pasture," and "farming equipment."

[0083] By leveraging the user's production scale and qualification level, tenders that have a budget far exceeding the user's affordability or are completely mismatched with the qualification requirements are initially filtered out.

[0084] This step allows for the initial screening of "historical bidding information sets" and "currently valid bidding information" that may be relevant to users from massive amounts of information, greatly reducing the computational load of subsequent in-depth processing.

[0085] The selected tender notices are analyzed in depth, and their unstructured text and tabular information are transformed into structured feature fields.

[0086] Tender notices may exist in various formats such as HTML web pages, PDF documents, and Word files. The system will call the corresponding parser (such as a PDF parsing library or an HTML parser) to extract plain text and tabular data.

[0087] Key information extraction using Natural Language Processing (NLP) techniques:

[0088] Named Entity Recognition (NER): Using a pre-trained NER model, identify and classify entities such as "tender category" (e.g., "alfalfa hay" or "veterinary vaccine"), "product delivery region" (e.g., "Korqin District, Tongliao City, Inner Mongolia"), and "bid deadline" (e.g., "before 17:00 on May 20, 2024") from the announcement text.

[0089] Amount and range extraction: Through regular expressions and rule models, identify "budget amount" or "maximum price" and normalize it into a numerical range (e.g., budget_min:500,000, budget_max:800,000, unit: yuan).

[0090] Technical standard structuring: Paragraphs such as "Technical parameter requirements" and "Quality standards" are parsed. Key indicators (such as "protein content ≥18%" and "equipment power: 30kW") are extracted through keyword matching and syntactic analysis and organized into a key-value pair list of {technical indicator: required value}.

[0091] Bidding party credit assessment: Based on the relevant corporate credit database, the bidding party's credit rating (e.g., AAA, AA), administrative penalty records, judicial risks, and other information are queried according to its unified social credit code, and quantified into a credit score.

[0092] The extracted raw features are cleaned and standardized to ensure consistent data quality.

[0093] Category standardization: The extracted multiple expressions (such as "silage corn" and "corn silage feed") are mapped to codes in a unified standard category library (such as "CAT_0012").

[0094] Geographic coding: Converts place names in text form into standard administrative division codes.

[0095] Time formatting: Convert all time information into ISO 8601 standard timestamps.

[0096] Technical parameter vectorization: The extracted list of key-value pairs is converted into a fixed-length vector based on a predefined industry-wide technical indicator dictionary. Each indicator in the dictionary corresponds to a position; if the tender specifies a requirement, that position contains the required value (normalized) or 1 (Boolean); otherwise, it contains 0.

[0097] All the standardized structured features mentioned above are arranged and concatenated according to a predefined global feature schema to form the structured feature vector of each bidding information.

[0098] The vector has a fixed dimension and may contain hundreds of feature bits, covering all key dimensions from category, budget, technology, region, time to the reputation of the bidding party.

[0099] These feature vectors, along with their original announcement IDs, publication times, and other metadata, are stored in a bidding information feature vector library (usually using a database that supports high-performance vector retrieval), and an index is created that is associated with the user's feature vectors to prepare data for the next step of real-time matching and sorting.

[0100] Step S104: Based on the user's historical bidding records and performance evaluations, a benefit assessment model is used to calculate the user's benefit index for various bidding projects. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements.

[0101] In this step, for each type of bidding project, the initial benefit base score is calculated based on the user's historical success rate, average profit margin of the successful bid amount, and the purchaser's performance evaluation score after the project is completed.

[0102] A matching coefficient between user production scale, technical equipment level and project bidding requirements is introduced to dynamically adjust the initial benefit base score, wherein the higher the matching degree, the larger the weighting coefficient.

[0103] By combining factors such as the user's recent bidding activity and the regional competition intensity of the project, a final comprehensive benefit index is generated. This benefit index is used to quantify the user's expected comprehensive benefits and probability of success in participating in the bidding project.

[0104] Furthermore, after calculating the user's benefit index for various bidding projects using a benefit evaluation model based on the user's historical bidding records and performance evaluations, a feedback learning mechanism is established: receiving real-time feedback from users on bidding information in the recommended list, including clicking to view details, saving bidding documents, downloading bidding templates, submitting bidding intentions, or marking as not interested; based on the feedback behavior, using an incremental learning algorithm to dynamically adjust the weights of relevant features in the user's feature vector and update the parameters of the benefit evaluation model to optimize the accuracy of subsequent recommendation results and user satisfaction.

[0105] In one specific embodiment, the model adopts a phased, multi-factor fusion computational architecture, and its implementation is divided into three phases:

[0106] Phase 1: Calculating the initial benefit base score

[0107] Data preparation: From the user behavior database, filter out all historical bidding records of the user under a specific bidding category (such as "dairy cow feed procurement") and the corresponding performance evaluation of the winning bids.

[0108] Indicator Calculation:

[0109] Historical success rate (R_h): Number of successful bids / Total number of bids. To smooth out data sparsity, Laplace smoothing or Bayesian averaging can be used.

[0110] Average Project Profit Margin (P_avg): For winning bids, calculate the average of (project settlement amount - estimated cost) / estimated cost. Cost data can be estimated from the user's production management system or historical quotation database.

[0111] Average Performance Rating (S_perf): Aggregates the evaluation scores (e.g., on a 5-point scale) of all completed projects by the tendering party for quality, timeliness, service, etc., and calculates the average value.

[0112] Base score calculation: The three normalized indicators mentioned above are combined using a weighted linear combination or a small neural network to calculate an initial benefit base score (Score_base) between 0 and 1. The formula is: Score_base = α*R_h + β*P_avg + γ*S_perf, where α, β, and γ are trainable weights that reflect the company's preference for success rate, profit, and reputation.

[0113] Phase Two: Dynamic Weighting Based on Capability Matching

[0114] Matching degree calculation: This step aims to assess the degree to which the user's current actual capabilities match the target tender requirements.

[0115] Input: The user multi-dimensional feature vector generated in step S102 (especially the static and dynamic dimensions such as production capacity and technical qualifications), and the structured feature vector of the target bidding project extracted in step S103 (technical standards, qualification requirements, production capacity requirements, etc.).

[0116] Matching process:

[0117] Hard match filtering: First, check whether the user's list of qualification certificates fully covers the mandatory qualification requirements of the tender. If not, the match degree can be set to 0.

[0118] Soft capability similarity calculation: For non-mandatory requirements such as technology and production capacity, calculate the cosine similarity between the user feature vector and the bidding requirement vector in the corresponding dimensions or the matching degree based on predefined rules.

[0119] Overall Match Coefficient (C_match): This combines the hard match result and the soft similarity result to generate a coefficient between 0 and 1. For example, C_match = Hard match passed? Soft similarity score: 0.

[0120] Dynamic weighting: The matching degree coefficient is used as a weight to adjust the initial benefit base score. In practice, exponential weighting can be used to amplify the advantage of a high matching degree, for example: Score_adjusted = Score_base * (1 + δ * C_match), where δ is the amplification factor. This means that even if a user's historical performance (base score) is average, if they are highly matched with the current project requirements, their adjusted score will significantly improve.

[0121] Phase 3: Integrating dynamic market factors to generate a comprehensive benefit index;

[0122] User Recent Bidding Activity Factor (A_user): This factor calculates the weighted frequency of a user's browsing, saving, and bidding activities related to similar tenders within a recent time window (e.g., 30 days). It is used to determine the user's current level of interest and willingness to participate in this type of project. Highly active users may respond more quickly to new opportunities.

[0123] Regional Competition Heat Factor (H_comp): This factor is calculated in real-time based on the average number of bidders for similar projects within the target bidding project's region over a past period, or estimated by monitoring recent search popularity of relevant keywords. This factor reflects the intensity of market competition.

[0124] The Overall Benefit Index (EI) is generated by combining the adjusted score with dynamic factors. One implementation is: EI = Score_adjusted * A_user * (1 / (1+η*H_comp)). Here, η is the coefficient of competition intensity. The formula shows that user activity positively adjusts the overall benefit, while market overheating (high competition intensity) will reduce the expected benefit. The final EI value is the user's benefit index for the bidding project, quantifying the expected overall return and probability of success.

[0125] To enable the benefit evaluation model and user profiles to evolve on their own, the following feedback learning loop is established:

[0126] Feedback behavior collection and quantification

[0127] The system captures various user interactions with the recommendation results and quantifies them into "reward signals" of varying intensities:

[0128] Strong positive feedback: Submitting a bid intention and downloading the bid document template (indicating strong interest). Reward value +2.

[0129] Weak positive feedback: Saving tender documents and viewing details for an extended period of time. Reward value +1.

[0130] Neutral / Exploratory Feedback: Click to view details. Reward +0.5.

[0131] Negative feedback: Mark as uninteresting and immediately disable recommendations. Reward value -1.

[0132] These behaviors, along with the corresponding recommended item features (i.e., the feature vector used to calculate the benefit index at that time), were recorded together as a feedback sample (user context, item features, user behavior, reward value).

[0133] Incremental learning and model update

[0134] Online learning algorithms: Contextual bandit algorithms (such as LinUCB or gradient updates of neural networks) are used as incremental learning frameworks.

[0135] Update process:

[0136] When a new batch of feedback samples is received, the system inputs the "user context" and "project characteristics" from the samples into the current benefit assessment model to obtain the predicted benefit index (i.e., expected reward).

[0137] The predicted reward is compared with the actual observed reward signal, and the loss (such as mean squared error) is calculated.

[0138] Using this loss, the parameters of the benefit evaluation model (such as the weights α, β, γ in the first stage, and the coefficients η in the third stage) can be fine-tuned through stochastic gradient descent (SGD) or mini-batch gradient descent.

[0139] At the same time, strong feedback behaviors (such as submitting a bid) will also trigger an immediate update of the user's dynamic behavioral characteristics (in step S102), such as increasing the weight of their interest in bidding for relevant product categories.

[0140] Dynamic adjustment of image weight

[0141] Feedback learning not only updates the benefit model, but is also used to adjust the weights of different features in the user's multi-dimensional feature vector.

[0142] For example, if a user repeatedly shows strong interest in items requiring "organic certification" (strong positive feedback), the system will gradually increase the weight of "organic certification" related features in the user profile during matching calculations through a feature importance analysis algorithm, so that it will have a greater influence in future recommendations.

[0143] Step S105: Perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information.

[0144] In this step, the first level of matching is to quickly filter based on the semantic similarity between the user's breeding type, main products and bidding categories, as well as the geographical proximity between the user's geographical location and the product delivery area, to form a preliminary candidate set.

[0145] Second-level matching: For the primary candidate set, calculate the cosine similarity between the user's production capacity feature vector and the tender technical requirements feature vector, and perform refined matching;

[0146] The third level of ranking: For the bidding information that has passed the fine matching, a multi-objective weighted comprehensive ranking is performed based on its corresponding user benefit index, the proximity of the bidding deadline, and the size of the bidding budget, to generate the final recommendation sequence.

[0147] Step S106: Based on the sorting results, generate a personalized bidding information recommendation list containing recommendation reasons, and push the personalized bidding information recommendation list to the livestock user through the user-specified terminal or message channel. The recommendation reasons include at least a matching score, benefit forecast, and time urgency reminder.

[0148] In this step, after generating the personalized bidding information recommendation list, personalized subscription and monitoring services are provided: allowing users to customize subscription conditions, including the bidding categories they are interested in, budget range, target region, and keywords; based on the user's subscription conditions, the bidding information database is continuously monitored in a streaming manner; when new bidding information is found to be highly matched with the user profile and subscription conditions, and the comprehensive score exceeds a preset threshold, a real-time alarm engine is triggered to generate a push notification containing key information and send it to the user's terminal.

[0149] In summary, the method and system for personalized recommendation of livestock bidding information proposed in this application achieves precise alignment of supply and demand characteristics by constructing a multi-dimensional feature vector of users that integrates static attributes and dynamic behaviors, and by performing deep structured analysis of bidding information. Based on this, it innovatively introduces and quantifies the core indicator of "efficiency index," which comprehensively considers users' historical bidding success rate, project profit margin, performance reputation, and real-time capability matching degree, while incorporating dynamic factors such as user activity and market competition intensity. This upgrades the recommendation logic from traditional "content matching" to "value and success probability prediction." Through a multi-level matching and ranking engine, this method can filter and prioritize bidding projects from massive amounts of information in real time, ensuring a high degree of matching with users' production capabilities and high expected returns. The system's built-in feedback learning mechanism dynamically adjusts user profile weights and optimizes efficiency evaluation model parameters based on users' real-time interactive behavior, achieving continuous self-evolution of the recommendation effect. Ultimately, this solution effectively overcomes the shortcomings of existing technologies, such as low information push accuracy, lack of commercial benefit orientation, passive lag, and inability to adaptively optimize. It significantly improves the efficiency, accuracy, and business success rate of livestock users in obtaining high-quality business opportunities, achieving a leap from "information recommendation" to "business decision empowerment".

[0150] Please see Figure 2 The diagram shows a structural block diagram of a personalized recommendation system for livestock bidding information according to this application.

[0151] like Figure 2 As shown, the personalized recommendation system 200 includes an acquisition module 210, a processing module 220, an extraction module 230, a calculation module 240, a matching module 250, and a push module 260.

[0152] The acquisition module 210 is configured to acquire user profile data of livestock users, including static attribute data and dynamic behavior data. The processing module 220 is configured to perform one-hot encoding and normalization on the static attribute data, calculate recent behavior weights on the dynamic behavior data using a time decay model, and perform vector fusion on the processed data to obtain a multi-dimensional feature vector of the user. The extraction module 230 is configured to acquire, in real time, a set of historical bidding information associated with the multi-dimensional feature vector of the user from a bidding information database and currently published bidding information, and extract the structured features of each bidding information, including at least the bidding category, budget range, technical standard requirements, product delivery region, bid deadline, and the credit rating of the bidding party. The calculation module 240 is configured to calculate the user's benefit index for various bidding projects based on the user's historical bidding records and performance evaluations using a benefit assessment model. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements. The matching module 250 is configured to perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information. The push module 260 is configured to generate a personalized bidding information recommendation list containing recommendation reasons based on the ranking results, and push the personalized bidding information recommendation list to the livestock user through the user's designated terminal or message channel. The recommendation reasons include at least a matching degree score, benefit prediction, and time urgency reminder.

[0153] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0154] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the personalized recommendation method for livestock bidding information in any of the above method embodiments.

[0155] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0156] Acquire user profile data of livestock users, the user profile data including static attribute data and dynamic behavior data;

[0157] The static attribute data is processed by one-hot encoding and normalization, and the dynamic behavior data is processed by calculating the weight of recent behavior through a time decay model. The processed data is then fused into vectors to obtain a multi-dimensional feature vector of the user.

[0158] From the bidding information database, historical bidding information sets and currently published bidding information associated with the user's multi-dimensional feature vector are obtained in real time. The structured features of each bidding information are extracted. The structured features include at least the bidding category, budget amount range, technical standard requirements, product delivery region, bid deadline, and the bidding party's credit rating.

[0159] Based on the user's historical bidding records and performance evaluations, a benefit assessment model is used to calculate the user's benefit index for various bidding projects. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements.

[0160] Perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information;

[0161] Based on the ranking results, a personalized bidding information recommendation list containing recommendation reasons is generated, and the personalized bidding information recommendation list is pushed to the animal husbandry user through the user's designated terminal or messaging channel. The recommendation reasons include at least a matching score, benefit forecast, and time urgency reminder.

[0162] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created by the use of a personalized recommendation system based on livestock bidding information. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely located relative to a processor, which can be connected to the personalized recommendation system for livestock bidding information via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0163] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the personalized recommendation method for livestock bidding information described in the above embodiment. The input device 330 can receive input digital or character information, and generate key signal inputs related to user settings and function control of the personalized recommendation system for livestock bidding information. The output device 340 may include a display screen or other display device.

[0164] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0165] In one implementation, the above-described electronic device is applied to a personalized recommendation system for livestock bidding information, for a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0166] Acquire user profile data of livestock users, the user profile data including static attribute data and dynamic behavior data;

[0167] The static attribute data is processed by one-hot encoding and normalization, and the dynamic behavior data is processed by calculating the weight of recent behavior through a time decay model. The processed data is then fused into vectors to obtain a multi-dimensional feature vector of the user.

[0168] From the bidding information database, historical bidding information sets and currently published bidding information associated with the user's multi-dimensional feature vector are obtained in real time. The structured features of each bidding information are extracted. The structured features include at least the bidding category, budget amount range, technical standard requirements, product delivery region, bid deadline, and the bidding party's credit rating.

[0169] Based on the user's historical bidding records and performance evaluations, a benefit assessment model is used to calculate the user's benefit index for various bidding projects. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements.

[0170] Perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information;

[0171] Based on the ranking results, a personalized bidding information recommendation list containing recommendation reasons is generated, and the personalized bidding information recommendation list is pushed to the animal husbandry user through the user's designated terminal or messaging channel. The recommendation reasons include at least a matching score, benefit forecast, and time urgency reminder.

[0172] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A personalized recommendation method for livestock bidding information, characterized in that, include: Acquire user profile data of livestock users, the user profile data including static attribute data and dynamic behavior data; The static attribute data is processed by one-hot encoding and normalization, and the dynamic behavior data is processed by calculating the weight of recent behavior through a time decay model. The processed data is then fused into vectors to obtain a multi-dimensional feature vector of the user. From the bidding information database, historical bidding information sets and currently published bidding information associated with the user's multi-dimensional feature vector are obtained in real time. The structured features of each bidding information are extracted. The structured features include at least the bidding category, budget amount range, technical standard requirements, product delivery region, bid deadline, and the bidding party's credit rating. Based on the user's historical bidding records and performance evaluations, a benefit assessment model is used to calculate the user's benefit index for various bidding projects. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements. Perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information; Based on the ranking results, a personalized bidding information recommendation list containing recommendation reasons is generated, and the personalized bidding information recommendation list is pushed to the animal husbandry user through the user's designated terminal or messaging channel. The recommendation reasons include at least a matching score, benefit forecast, and time urgency reminder.

2. The personalized recommendation method for livestock bidding information according to claim 1, characterized in that, The static attribute data includes the user's production scale, breeding type, enterprise qualification certification information, and geographical location; the dynamic behavior data includes the user's historical bidding records, bidding information browsing records, collection records, and performance evaluation feedback on successful bids. The acquisition of user profile data for livestock users includes: By connecting to the livestock production management system or IoT devices, the user's basic production data is automatically obtained. The basic production data includes pasture area, number of animals in stock, average annual output, main livestock species and growth cycle data. By connecting to the enterprise credit information platform and the government's public data interface, we can verify and capture users' qualification certification information, including animal disease prevention and control qualification certificates, environmental impact assessment approvals, organic product certification certificates, green food labels, and enterprise credit rating reports. By analyzing users' operation logs on the bidding platform, we can extract users' historical bidding records, including bidding project types, bidding frequency, success rate, preferred bidding regions, commonly used bidding strategies, and analysis of reasons for bid rejection.

3. The personalized recommendation method for livestock bidding information according to claim 1, characterized in that, The step involves calculating the user's benefit index for various bidding projects using a benefit assessment model based on the user's historical bidding records and performance evaluations, including: For each type of bidding project, the initial benefit base score is calculated based on the user's historical success rate, average profit margin of the successful bid amount, and the purchaser's performance evaluation score after project completion. A matching coefficient between user production scale, technical equipment level and project bidding requirements is introduced to dynamically adjust the initial benefit base score, wherein the higher the matching degree, the larger the weighting coefficient. By combining factors such as the user's recent bidding activity and the regional competition intensity of the project, a final comprehensive benefit index is generated. This benefit index is used to quantify the user's expected comprehensive benefits and probability of success in participating in the bidding project.

4. The personalized recommendation method for livestock bidding information according to claim 1, characterized in that, The step of performing multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information includes: First-level matching: Based on the semantic similarity between the user's breeding type, main products and bidding categories, as well as the geographical proximity between the user's geographical location and the product delivery area, a preliminary candidate set is quickly filtered. Second-level matching: For the primary candidate set, calculate the cosine similarity between the user's production capacity feature vector and the tender technical requirements feature vector, and perform refined matching; The third level of ranking: For the bidding information that has passed the fine matching, a multi-objective weighted comprehensive ranking is performed based on its corresponding user benefit index, the proximity of the bidding deadline, and the size of the bidding budget, to generate the final recommendation sequence.

5. The personalized recommendation method for livestock bidding information according to claim 1, characterized in that, After calculating the user's benefit index for various bidding projects using a benefit assessment model based on the user's historical bidding records and performance evaluations, the method further includes: Establish a feedback learning mechanism: Receive real-time feedback from users on the bidding information in the recommended list. The feedback behaviors include clicking to view details, saving bidding documents, downloading bidding templates, submitting bidding intentions, or marking as not interested. Based on the feedback behavior, the weights of relevant features in the user feature vector are dynamically adjusted using an incremental learning algorithm, and the parameters of the benefit evaluation model are updated to optimize the accuracy of subsequent recommendation results and user satisfaction.

6. The personalized recommendation method for livestock bidding information according to claim 1, characterized in that, After generating the personalized bidding information recommendation list, the method further includes: Offers personalized subscription and monitoring services: allows users to customize subscription conditions, including the categories of tenders they are interested in, budget range, target region, and keywords; Based on user subscription conditions, the bidding information database is continuously monitored. When new bidding information is found to be highly matched with user profiles and subscription conditions, and the comprehensive score exceeds a preset threshold, a real-time alarm engine is triggered to generate a push notification containing key information and send it to the user's terminal.

7. A personalized recommendation system for livestock bidding information, characterized in that, include: The acquisition module is configured to acquire user profile data of livestock users, the user profile data including static attribute data and dynamic behavior data; The processing module is configured to use one-hot encoding and normalization to process the static attribute data, calculate the recent behavior weights of the dynamic behavior data through a time decay model, and perform vector fusion on the processed data to obtain a multi-dimensional feature vector of the user. The extraction module is configured to obtain, in real time, a set of historical bidding information associated with the user's multi-dimensional feature vector and currently published bidding information from the bidding information database, and extract the structured features of each bidding information. The structured features include at least the bidding category, budget amount range, technical standard requirements, product delivery region, bid deadline, and the bidding party's credit rating. The calculation module is configured to calculate the user's benefit index for various bidding projects based on the user's historical bidding records and performance evaluations using a benefit assessment model. The benefit index comprehensively considers historical success rate, average project profit rate, performance completion score, and the matching degree between the user's production capacity and project requirements. The matching module is configured to perform multi-level similarity matching between the user's multi-dimensional feature vector and the structured feature vector of the current bidding information; The push module is configured to generate a personalized bidding information recommendation list containing recommendation reasons based on the sorting results, and push the personalized bidding information recommendation list to the livestock user through the user-specified terminal or message channel. The recommendation reasons include at least a matching score, benefit forecast, and time urgency reminder.

8. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements the method described in any one of claims 1 to 6.

Images