Multi-dimensional user behavior feature mining and recommending method and system for precise business cooperation

Abstract

The invention relates to the technical field of digital data processing, in particular to a multi-dimensional user behavior feature mining and recommendation method and system for precise business cooperation, and the method comprises the steps: building a user portrait type through a first clustering algorithm based on user behavior data; on the basis of the resource feature data, a resource portrait is constructed through a second clustering algorithm, and resource portrait features comprise the industry type of the resource, the resource type and the potential cooperative partner of the resource; extracting a transaction feature group based on transaction data corresponding to a certain type of user portraits; matching the features of the transaction feature group with the resource portraits to obtain recommended resource portraits; pushing the corresponding resource data to the client based on the recommended resource portrait; and scoring the resource data based on the feedback of all clients. According to the invention, matching and recommendation can be automatically carried out in combination with the features of user data and resource data, thereby improving the accuracy of resource recommendation and improving the transaction efficiency.

Description

Technical Field

[0001] The present invention relates to the field of digital data processing technology, and in particular to a multi-dimensional user behavior feature mining and recommendation method and system for precise business cooperation. Background Art

[0002] With the rapid development of information technology, the demand for business cooperation is increasing, especially in the B2B field. Traditional business resource acquisition methods mostly rely on manual recommendations, salesmen and account managers' active recommendations, but these methods have problems such as untimely information acquisition and inaccurate recommendations.

[0003] At present, existing resource recommendation algorithms are mainly concentrated in the e-commerce field, such as collaborative filtering algorithms based on user historical purchase records, content-based recommendations, etc. However, most of these algorithms are optimized for consumer purchase behavior and have not been specifically improved for the business cooperation field. Business cooperation resources often have complex characteristics, including user business needs, industry categories, resource types, etc. Therefore, how to provide accurate cooperation resource recommendations based on user needs and resource characteristics has become a major problem in current technology. Summary of the invention

[0004] The purpose of the present invention is to provide a multi-dimensional user behavior feature mining and recommendation method and system for precise business cooperation, aiming to automatically match and recommend based on the characteristics of user data and resource data, thereby improving the accuracy of resource recommendations and thus improving transaction efficiency.

[0005] To achieve the above-mentioned purpose, in a first aspect, the present invention provides a multi-dimensional user behavior feature mining and recommendation method for precise business cooperation, comprising real-time collection of user behavior data, resource feature data and transaction data, wherein the user behavior data comprises user identity data, user search data, browsing data and click data; the resource feature data is extracted from resource data, wherein the resource feature data comprises industry data, cooperation model data, resource type data and resource demand data;

[0006] Based on the user behavior data, a user portrait type is constructed by a first clustering algorithm, wherein the user portrait features include the user's industry background, browsing preference, click frequency, and demand type;

[0007] Based on the resource feature data, a resource portrait is constructed by a second clustering algorithm, wherein the resource portrait features include the industry type of the resource, the resource type, and potential partners of the resource;

[0008] Extract transaction feature groups based on transaction data corresponding to a certain type of user portrait;

[0009] Match the features of the transaction feature group with the resource profile to obtain a recommended resource profile;

[0010] Push the corresponding resource data to customers based on the recommended resource profile;

[0011] Resource data is scored based on feedback from all customers.

[0012] The specific steps of collecting user behavior data, resource feature data and transaction data in real time include:

[0013] Extract resource feature data in the resource aggregation platform;

[0014] Record the user behavior data generated by users on the resource aggregation platform;

[0015] When a user completes a transaction, the user behavior data and the corresponding resource feature data are linked to form transaction data;

[0016] Use message queues to transmit user behavior data and transaction data to the host computer in real time;

[0017] Preprocess resource feature data, user behavior data, and transaction data.

[0018] The specific steps of constructing a user portrait type based on user behavior data by using a first clustering algorithm include:

[0019] Standardize user behavior data to obtain standard user data;

[0020] Use K-means algorithm to cluster the standard user data to obtain the first clustering result;

[0021] Different types of user portraits are defined based on the first clustering results.

[0022] The specific steps of constructing a resource portrait based on resource feature data by using a second clustering algorithm include:

[0023] Obtain resource profile feature data related to resource profile;

[0024] Use a hierarchical clustering algorithm to cluster the resource portrait feature data to obtain the second clustering result;

[0025] The common characteristics of the resources in each cluster in the second clustering result are analyzed to obtain different types of resource portraits.

[0026] The specific steps of extracting a transaction feature group based on the transaction data corresponding to a certain type of user portrait include:

[0027] Extract resource feature data from transaction data;

[0028] Filter relevant transaction data from resource feature data based on user profile type;

[0029] Normalizing the relevant transaction data to obtain processed transaction data;

[0030] Extract transaction feature groups based on processed transaction data.

[0031] The specific steps of matching the features of the transaction feature group with the resource portrait to obtain the recommended resource portrait include:

[0032] Create a matching matrix, where rows represent users, columns represent resources, and each cell contains the user's behavior data on the resource;

[0033] Select cosine similarity to calculate the similarity between users;

[0034] Based on the similarity score, find a group of users that are most similar to the target user;

[0035] Calculating resource similarity based on the features of the transaction feature group corresponding to the group of users and the resource portrait;

[0036] Sort candidate resources by resource similarity;

[0037] The first N resources are selected from the sorted resources as resource profiles recommended to the user to obtain a resource list.

[0038] The specific steps of pushing the corresponding resource data to the client based on the recommended resource portrait include:

[0039] Get a list of recommended resources;

[0040] The resource list is further customized according to the customer's preferences to obtain a customized list;

[0041] Send customized lists to customers according to the corresponding push platform.

[0042] The specific steps of scoring resource data based on feedback from all customers include:

[0043] Collect customer feedback information in the resource aggregation platform;

[0044] Remove invalid feedback records from customer feedback information;

[0045] The resource data score is calculated based on each score and corresponding weight in the customer feedback information.

[0046] In a second aspect, the present invention also provides a multi-dimensional user behavior feature mining and recommendation system for precise business cooperation, including a data collection module, a user portrait generation module, a resource portrait generation module, a transaction feature generation module, a recommended resource portrait generation module, a push module and a scoring module;

[0047] The data collection module is used to collect user behavior data, resource feature data and transaction data in real time. The user behavior data includes user identity data, user search data, browsing data and click data. The resource feature data is extracted from resource data. The resource feature data includes industry data, cooperation model data, resource type data and resource demand data.

[0048] The user portrait generation module is used to construct a user portrait type based on the user behavior data through a first clustering algorithm, wherein the user portrait features include the user's industry background, browsing preference, click frequency, and demand type;

[0049] The resource portrait generation module is used to construct a resource portrait based on the resource feature data through a second clustering algorithm, wherein the resource portrait features include the industry type of the resource, the resource type, and the potential partners of the resource;

[0050] The transaction feature generation module is used to extract a transaction feature group based on the transaction data corresponding to a certain type of user portrait;

[0051] The recommended resource portrait generation module is used to match the features of the transaction feature group with the resource portrait to obtain the recommended resource portrait;

[0052] The push module is used to push the corresponding resource data to the client based on the recommended resource portrait;

[0053] The scoring module is used to score the resource data based on the feedback from all customers.

[0054] The present invention provides a multi-dimensional user behavior feature mining and recommendation method and system for precise business cooperation. The system first deploys a series of data collection mechanisms to capture all activities of users on the platform, such as browsing history, number of clicks, dwell time, etc.; it also collects information about various resources on the platform, such as industry classification, product features and service details; in addition, it also tracks each successful transaction, including but not limited to transaction amount, transaction frequency and partner information. Based on the collected user behavior data, the first clustering algorithm is used to segment users to form different types of user portraits. Each user portrait will contain detailed descriptive information, such as industry background, preferred product category, page access mode, specific demand type, etc. This helps to deeply understand the unique needs and interests of each user. Similarly, for various resources provided on the platform, clustering algorithms will also be used to analyze them to establish detailed resource portraits. For a specific type of user portrait, the transaction cases related to it are further analyzed to extract a set of representative transaction features. These features can reveal which factors are most likely to promote the occurrence of transactions and provide a basis for subsequent recommendations. The above-obtained transaction feature group is compared with the existing resource portrait to find the best match, thereby generating a recommended resource portrait.

[0055] According to the finalized recommended resource portrait, select the corresponding resources and send detailed information to the target customer group. This step uses personalized customization to improve the relevance and attractiveness of the information and increase the response rate of potential customers. Finally, let all customers who have participated in the recommendation process give their evaluations of the resources they received. This feedback can not only serve as an important indicator for evaluating the effectiveness of the recommendation, but also help optimize future recommendation strategies to make them more in line with changes in market demand and personal preferences. Through such a cyclical iterative process, the method can continuously learn and adapt to new trends, providing users with an increasingly accurate service experience. In this way, the characteristics of user data and resource data can be combined to automatically match and recommend, thereby improving the accuracy of resource recommendations and thus improving transaction efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0057] Figure 1 It is a flow chart of a multi-dimensional user behavior feature mining and recommendation method and system for precise business cooperation of the present invention.

[0058] Figure 2 It is a flow chart of the present invention for collecting user behavior data, resource feature data and transaction data in real time.

[0059] Figure 3 It is a flowchart of the present invention for constructing user portrait types based on user behavior data through a first clustering algorithm.

[0060] Figure 4 It is a flow chart of the present invention for constructing a resource portrait based on resource feature data through a second clustering algorithm.

[0061] Figure 5 It is a flow chart of the present invention for extracting a transaction feature group based on transaction data corresponding to a certain type of user portrait.

[0062] Figure 6 It is a flow chart of the present invention that matches the features of the transaction feature group with the resource portrait to obtain the recommended resource portrait.

[0063] Figure 7 It is a flow chart of the present invention for pushing corresponding resource data to customers based on recommended resource portraits.

[0064] Figure 8 It is a flow chart of the present invention for scoring resource data based on feedback from all customers. DETAILED DESCRIPTION

[0065] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0066] First embodiment

[0067] See also Figures 1 to 8 The present invention provides a multi-dimensional user behavior feature mining and recommendation method for precise business cooperation, comprising:

[0068] S101 collects user behavior data, resource feature data and transaction data in real time, wherein the user behavior data includes user identity data, user search data, browsing data and click data; the resource feature data is extracted from resource data, and the resource feature data includes industry data, cooperation model data, resource type data and resource demand data;

[0069] The specific steps include:

[0070] S201 extracts resource feature data from a resource aggregation platform;

[0071] Get the latest resource information from the internal database or API interface of the resource aggregation platform to ensure that the resource feature data is up to date. This step requires periodic or real-time capture of the latest resource information from the platform database. Ensure that the extracted data is complete and correct, and avoid missing key fields or introducing erroneous information. Unify the format of resource data from different sources to facilitate subsequent analysis and processing.

[0072] S202 records the user behavior data generated by the user in the resource aggregation platform;

[0073] Capture and store every user interaction on the platform through front-end tracking code and back-end logging. This includes but is not limited to: setting event listeners in the web page or application interface to capture user clicks, scrolling, form submissions, etc. Use session management technology to continuously track the user's browsing path until they leave the platform. Record page loading time and response speed to evaluate user experience.

[0074] S203: After the user completes the transaction, the user behavior data and the corresponding resource feature data are linked to form transaction data;

[0075] When a transaction is completed, the user's behavior throughout the purchase process is combined with the characteristics of the purchased resources to form a complete transaction record. This process involves matching user behavior data with specific resource characteristic data through a unique transaction ID or other identifier. Create a comprehensive data set containing user behavior, resource characteristics, and transaction details for subsequent analysis. Identify and address existing data inconsistencies to ensure the accuracy of transaction data.

[0076] S204 uses a message queue to transmit user behavior data and transaction data to a host computer in real time;

[0077] In order to ensure the efficiency and reliability of data transmission, message queues (such as Apache Kafka, RabbitMQ, etc.) are used to realize the transmission of real-time data streams. Message queues are configured to support a large number of concurrent write and read operations to meet real-time requirements. This implements retry logic and persistent storage to prevent data loss.

[0078] S205 pre-processes the resource feature data, user behavior data and transaction data.

[0079] Before data analysis, the collected data is cleaned, converted, and standardized to ensure its quality and consistency. The main tasks include: removing duplicates, fixing missing values, and correcting erroneous entries. Numerical data is standardized so that data of different scales can be compared fairly.

[0080] S102 constructs a user portrait type based on the user behavior data by using a first clustering algorithm, wherein the user portrait features include the user's industry background, browsing preference, click frequency, and demand type;

[0081] The specific steps include:

[0082] S301 standardizes the user behavior data to obtain standard user data;

[0083] Since user behavior data comes from different sources and has different dimensions, it needs to be standardized first to ensure fairness and accuracy in the subsequent clustering process. This process involves the following aspects: For continuous variables (such as the number of clicks), apply Z-score standardization or Min-Max scaling method to convert all values ​​to the same scale range. For non-numeric variables (such as industry background), use one-hot encoding or label encoding to convert them into numerical form so that machine learning algorithms can process them. For data points with missing values, mean filling, median filling or more complex interpolation methods can be used to supplement them.

[0084] S302 clusters the standard user data using a K-means algorithm to obtain a first clustering result;

[0085] K-means was chosen as the initial clustering algorithm because it is simple, efficient, and applicable to large-scale data sets. The key to this step is to select the optimal k value through the Elbow Method, Silhouette Score, or other evaluation indicators. Initialize the centroid position randomly or using optimization methods such as K-means++ to accelerate convergence and improve clustering results. Repeatedly adjust the cluster to which each sample belongs until the centroid no longer moves significantly or the preset maximum number of iterations is reached. Calculate internal evaluation indicators (such as the sum of intra-cluster distances and silhouette coefficients) and external evaluation indicators (such as comparison with known categories) to evaluate the quality of the clustering results.

[0086] S303 defines different types of user portraits based on the first clustering result.

[0087] Based on the clustering results generated by the K-means algorithm, the common characteristics of users in each cluster are further analyzed, and different types of user portraits are defined accordingly. This step mainly includes: in-depth analysis of the characteristic distribution of users in each cluster, and finding the most representative attribute combination, such as "high-frequency technology enthusiasts" and "price-sensitive buyers". Give each cluster an easy-to-understand name and describe its typical characteristics and behavior patterns in detail to facilitate the business team's understanding and application.

[0088] S103 constructs a resource portrait based on the resource feature data by using a second clustering algorithm, wherein the resource portrait features include the industry type of the resource, the resource type, and potential partners of the resource;

[0089] The specific steps include:

[0090] S401 obtains resource profile feature data related to the resource profile;

[0091] First, we need to ensure that comprehensive and high-quality resource feature data is collected, which will serve as the basis for subsequent cluster analysis. This process involves obtaining resource information from multiple internal and external data sources, including but not limited to databases, API interfaces, third-party service platforms, etc. Then, we remove duplicates, fix missing values, and correct erroneous entries to ensure the completeness and accuracy of the data.

[0092] S402 clusters the resource portrait feature data using a hierarchical clustering algorithm to obtain a second clustering result;

[0093] Hierarchical Clustering is selected as the main clustering tool because it can handle clusters of arbitrary shapes and does not require the number of clusters to be specified in advance. The specific operation is to select an appropriate distance measurement method (such as Euclidean distance, Manhattan distance, cosine similarity) to measure the similarity between resources. Determine how to calculate the distance between clusters. Common link criteria include single link, complete link, average link, etc. By recursively merging the most similar resources, a hierarchical clustering tree (Dendrogram) is gradually formed. According to business needs or through evaluation indicators (such as silhouette coefficient, Calinski-Harabasz index), the appropriate level is selected for truncation to obtain the final clustering result. Use charts or graphical tools to intuitively display the clustering tree and the final clustering results to help decision makers understand the relationship between resources.

[0094] S403 analyzes the common features of the resources in each cluster in the second clustering result to obtain different types of resource portraits.

[0095] Based on the clustering results generated by the hierarchical clustering algorithm, further analyze the common characteristics of resources in each cluster, and define different types of resource portraits accordingly. This step includes in-depth analysis of the characteristic distribution of resources in each cluster to find the most representative attribute combinations, such as "high-value technical resources" and "emerging market potential stocks". Give each cluster an easy-to-understand name and describe its typical characteristics and behavior patterns in detail to facilitate the business team's understanding and application. Use the Apriori algorithm or other association rule mining techniques to find frequently occurring combination patterns, which can serve as an important part of resource portraits.

[0096] S104 extracts a transaction feature group based on the transaction data corresponding to a certain type of user portrait;

[0097] The specific steps include:

[0098] S501 extracts resource feature data from transaction data;

[0099] First, it is necessary to obtain detailed resource feature information from the existing transaction database, which will serve as the basis for subsequent filtering and analysis.

[0100] Make sure you have all relevant transaction records, and each record should contain a complete description of the resource (such as industry type, resource type, potential partners, etc.). Select the fields that best reflect the characteristics of the resource based on business needs, such as the value of the resource, the industry it belongs to, the cooperation model, etc.

[0101] S502: filtering out relevant transaction data from resource feature data according to user portrait type;

[0102] Next, by matching the user profile type, the transaction records related to the specific user group are filtered out from the resource feature data. The key to this step is to use the pre-built user profile model to determine which transaction records meet the characteristics of a specific user profile. Then set reasonable screening conditions, such as time range, geographic area, transaction amount, etc., to further narrow the data set. If applicable, association rule mining technology can be used to discover the implicit relationship between user profiles and transaction data to improve the screening accuracy. Ensure the consistency and accuracy of the screening results to avoid missing important information or introducing bias.

[0103] S503 normalizes the relevant transaction data to obtain processed transaction data;

[0104] In order to eliminate the impact of dimensional differences on the analysis results, the filtered transaction data needs to be normalized.

[0105] Select an appropriate normalization method based on the data characteristics, such as Min-Max scaling, Z-score standardization, etc. Apply the selected method to normalize all numerical features to ensure that they are within the same scale range.

[0106] S504 extracts a transaction feature group based on the processed transaction data.

[0107] Finally, by conducting an in-depth analysis of the normalized transaction data, we can extract the key feature groups that can represent the transaction behavior patterns of this type of user. This step involves determining which features have the greatest impact on transaction results based on business logic and domain knowledge. Results can be obtained through statistical analysis, expert judgment, or machine learning models. If there are too many features, consider using methods such as PCA (principal component analysis) and Lasso regression to reduce dimensions and improve the efficiency of subsequent analysis.

[0108] S105 matches the features of the transaction feature group with the resource portrait to obtain a recommended resource portrait;

[0109] The specific steps include:

[0110] S601 creates a matching matrix, where rows represent users, columns represent resources, and each cell contains user behavior data on resources;

[0111] First, you need to create a user-resource interaction matrix to record the interaction between users and resources. This includes constructing a two-dimensional matrix where each row corresponds to a user, each column corresponds to a resource, and each cell in the matrix represents the user's specific behavior data for the resource (such as ratings, clicks, purchase frequency, etc.). For cells without interaction records, default values ​​(such as 0 or the average value) can be used to fill them, or more complex interpolation methods can be used to estimate missing values. Considering that the user-resource matrix is ​​highly sparse in most cases, matrix decomposition techniques (such as SVD, ALS) can be used to reduce the impact of sparsity.

[0112] S602 selects cosine similarity to calculate the similarity between users;

[0113] Next, we use cosine similarity to calculate the similarity between users to identify groups of users with similar preferences. We ensure that all user vectors are on the same scale to avoid deviations caused by large numerical differences. Based on the user-resource interaction matrix, we calculate the cosine similarity between all users to form a user similarity matrix.

[0114] S603, finding a group of users most similar to the target user according to the similarity score;

[0115] Define a similarity threshold and filter out users with scores above the threshold as candidate similar users. Select the top K users closest to the target user from the candidate similar users as the target user's "neighbors". According to actual business needs, assign different weights to different similarity scores to enhance the relevance of recommendation results.

[0116] S604 calculates resource similarity based on the features of the transaction feature group corresponding to the group of users and the resource portrait;

[0117] Combine the transaction feature groups of the target user's neighbors with the existing resource portraits to calculate the similarity between resources. This step includes extracting key features (such as industry type, resource type, potential partners, etc.) from the transaction feature group and comparing them with the features in the resource portrait. Use a suitable distance metric (such as Euclidean distance, Manhattan distance, cosine similarity, etc.) to calculate the similarity between resources. If there are multiple similarity indicators, the final resource similarity score can be obtained through weighted average or other comprehensive scoring methods.

[0118] S605 sorts the candidate resources according to resource similarity;

[0119] According to the calculated resource similarity scores, candidate resources are sorted for subsequent recommendation. This step includes sorting candidate resources from high to low according to resource similarity scores. Duplicate resource items are removed to ensure the diversity of the recommendation list. The upper limit of the number of recommended resources is set according to business needs, for example, only the first N resources are retained.

[0120] S606 selects the first N resources from the sorted resources as resource portraits recommended to the user, and obtains a resource list.

[0121] Finally, select the top N most relevant resources from the sorted resources as the resource profiles recommended to the user, and generate the final resource recommendation list. This step includes: organizing the top N sorted resources into a recommendation list, ensuring that the list is clear and easy to read. Provide a brief reasoning for each recommended resource to help users understand why these resources are recommended. Personalize the recommended content based on user preferences and historical behavior to improve user experience.

[0122] Through the above steps, the characteristics of the transaction feature group can be effectively matched with the resource profile, thereby providing users with personalized and efficient resource recommendations. This method not only improves the accuracy of the recommendation system, but also enhances user participation and satisfaction.

[0123] S106 pushes corresponding resource data to the customer based on the recommended resource profile;

[0124] The specific steps include:

[0125] S701 obtains a recommended resource list;

[0126] First, starting from the recommended resource portrait obtained in the previous step (S105), extract a specific resource data list. The key to this step is to verify whether the recommended resource list has been sorted and filtered according to the predetermined criteria. Ensure that the latest recommendation results are synchronized with the resource database in the system to obtain the most accurate resource information. Perform necessary formatting on the resource data to meet the requirements of the subsequent push platform.

[0127] S702 further customizes the resource list according to the customer's preference to obtain a customized list;

[0128] Considering that each customer has unique preferences and needs, it is crucial to personalize the recommended resource list. This process includes in-depth research on customers' historical purchase records, browsing habits, and other interactive behaviors to better understand their preferences. Use machine learning models or rule engines to optimize the resource list based on customers' specific attributes (such as age, gender, geographic location, etc.) and interest tags. Assign different weights to different types of resources so that those that are more attractive to target customers are ranked first. Make sure that the final customized list is neither too lengthy nor too long, but covers multiple types of resources to avoid single recommendations.

[0129] S703 sends the customized list to the customer according to the corresponding push platform.

[0130] Finally, choose the appropriate push channel and efficiently convey the customized resource list to customers.

[0131] This involves

[0132] Determine the best push platform (e.g., email, mobile app notifications, social network messages, etc.) based on the customer's usage habits and technical environment. Adjust the format and presentation of push content based on the characteristics of different platforms to ensure the effectiveness and attractiveness of information delivery. Consider the customer's active time and time zone differences, choose the most appropriate time to send push notifications, and increase the open rate and interaction rate. Establish a feedback loop mechanism to collect customer responses to push content and continuously optimize push strategies and content quality.

[0133] S107 scores the resource data based on feedback from all customers.

[0134] The specific steps include:

[0135] S801 collects customer feedback information in the resource aggregation platform;

[0136] First, you need to collect customer feedback from multiple channels to ensure that you cover as many user groups as possible. The key to this step is to use a variety of methods to obtain customer feedback, including but not limited to online evaluation systems, questionnaires, customer service records, social media comments, email feedback, etc.

[0137] Deploy automated tools (such as API interfaces, Webhooks) to connect with various feedback channels to capture the latest customer reviews in real time. Design reasonable incentive mechanisms, such as point rewards, discount coupons, etc., to encourage more users to provide valuable feedback. Provide anonymous submission options to increase user participation, especially for those who do not want to disclose their identities.

[0138] S802 removes invalid feedback records from customer feedback information;

[0139] In order to ensure the accuracy and fairness of the ratings, the collected feedback information must be strictly screened to remove invalid or inappropriate records. Identify and remove duplicate, blank or obviously erroneous feedback records. Use statistical methods or machine learning models to detect and exclude extreme values ​​or outliers to prevent individual malicious reviews from affecting the overall rating. Ensure that each feedback comes from a real user through cross-validation (such as IP address checking and account behavior analysis) to avoid false reviews.

[0140] S803 calculates the resource data score according to each score and the corresponding weight in the customer feedback information.

[0141] Finally, based on the filtered valid feedback information, calculate the comprehensive score of resource data. This process involves determining which aspects are the most important in customer feedback, such as product quality, service attitude, price rationality, etc., and setting a specific scoring range for each dimension (such as 1-5 stars). Assign corresponding weights according to the importance of different scoring dimensions to ensure that the final score can fully reflect the overall performance of the resource. The weight can be determined based on business needs, market research results or historical data analysis. Finally, the comprehensive resource data score is calculated based on the weight and the score of each indicator. For scores of different scales (such as percentages for some dimensions and five-stars for other dimensions), apply standardization methods (such as Z-score standardization or Min-Max scaling) to keep them within the same scale range. As new feedback continues to flow in, update resource scores regularly to maintain the timeliness and accuracy of the scores.

[0142] Second embodiment

[0143] The present invention also provides a multi-dimensional user behavior feature mining and recommendation system for precise business cooperation, including a data collection module, a user portrait generation module, a resource portrait generation module, a transaction feature generation module, a recommended resource portrait generation module, a push module and a scoring module; the data collection module is used to collect user behavior data, resource feature data and transaction data in real time, the user behavior data includes user identity data, user search data, browsing data, and click data; the resource feature data is extracted from resource data, and the resource feature data includes industry data, cooperation model data, resource type data, and resource demand data; the user portrait generation module is used to construct a user portrait based on the user behavior data through a first clustering algorithm. The user portrait features include the user's industry background, browsing preferences, click frequency and demand type; the resource portrait generation module is used to construct a resource portrait based on the resource feature data through a second clustering algorithm, and the resource portrait features include the industry type of the resource, the resource type, and the resource potential partner; the transaction feature generation module is used to extract a transaction feature group based on the transaction data corresponding to a certain type of user portrait; the recommended resource portrait generation module is used to match the features of the transaction feature group with the resource portrait to obtain a recommended resource portrait; the push module is used to push the corresponding resource data to the customer based on the recommended resource portrait; the scoring module is used to score the resource data based on the feedback from all customers.

[0144] In this embodiment, the data collection module collects various types of data in real time. It not only covers the various activities of users on the platform, such as searching keywords, browsing pages, clicking links, etc., but also includes the user's identity information (such as enterprise size, industry, etc.) and detailed attributes of resources (such as industry category, cooperation model, type and demand).

[0145] The user portrait generation module uses the first clustering algorithm to build detailed user portraits based on the collected user behavior data. This enables the system to gain in-depth understanding of each user's background information, preferred areas, interaction frequency, and their specific demand patterns.

[0146] The resource portrait generation module uses the second clustering algorithm to classify and describe various resources provided on the platform. The resource portrait is a high-level summary of the characteristics of the resource itself, including but not limited to the industry to which it belongs, the cooperation methods provided, the types of resources, and potential partners. This step helps to establish a clear and representative resource image, which facilitates accurate positioning in the subsequent matching process.

[0147] The transaction feature generation module focuses on analyzing the transaction patterns hidden behind specific types of user portraits. By extracting the common elements in these successful cases, a set of key factors that can reflect the success of the transaction is formed.

[0148] ——That is, the “transaction feature group”. The work of this module is directly related to the effectiveness and accuracy of the recommendation effect.

[0149] The recommended resource profile generation module compares the above transaction feature group with the existing resource profiles to find the resource combination that is most likely to attract the target user's interest. This process emphasizes the high degree of fit between transaction features and resource characteristics, ensuring that the recommended resources meet both user expectations and their actual needs.

[0150] As a bridge between the system and end users, the push module is tasked with displaying selected high-quality resources to customers in a timely and accurate manner. In order to achieve the best user experience, the push content must not only fit personal preferences, but also take into account factors such as timeliness and relevance.

[0151] The scoring module evaluates the pushed resources based on the feedback from all customers. This mechanism encourages positive interactions and provides the system with opportunities for continuous learning and optimization, thus ensuring a high level of service quality in the long term.

[0152] In summary, this system forms a closed-loop learning and optimization process through comprehensive data collection, detailed portrait construction, scientific feature extraction, precise resource matching, timely content push, and effective feedback evaluation, which can make efficient recommendations and improve work efficiency.

[0153] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A multi-dimensional user behavior feature mining and recommendation method for precise business cooperation, It is characterized in that Including: real-time collection of user behavior data, resource feature data and transaction data, the user behavior data includes user identity data, user search data, browsing data, click data; the resource feature data is extracted from resource data, the resource feature data includes industry data, cooperation model data, resource type data, resource demand data; Based on the user behavior data, a user portrait type is constructed by a first clustering algorithm, wherein the user portrait features include the user's industry background, browsing preference, click frequency, and demand type; Based on the resource feature data, a resource portrait is constructed by a second clustering algorithm, wherein the resource portrait features include the industry type of the resource, the resource type, and potential partners of the resource; Extract transaction feature groups based on transaction data corresponding to a certain type of user portrait; Match the features of the transaction feature group with the resource profile to obtain a recommended resource profile; Push the corresponding resource data to customers based on the recommended resource profile; Resource data is scored based on feedback from all customers.

2. A multi-dimensional user behavior feature mining and recommendation method for precise business cooperation as claimed in claim 1, characterized in that: The specific steps of real-time collection of user behavior data, resource feature data and transaction data include: Extract resource feature data in the resource aggregation platform; Record the user behavior data generated by users on the resource aggregation platform; When a user completes a transaction, the user behavior data and the corresponding resource feature data are linked to form transaction data; Use message queues to transmit user behavior data and transaction data to the host computer in real time; Preprocess resource feature data, user behavior data, and transaction data.

3. A multi-dimensional user behavior feature mining and recommendation method for precise business cooperation as claimed in claim 2, characterized in that: The specific steps of constructing a user portrait type based on the user behavior data by using the first clustering algorithm include: Standardize user behavior data to obtain standard user data; Use K-means algorithm to cluster the standard user data to obtain the first clustering result; Different types of user portraits are defined based on the first clustering results.

4. A multi-dimensional user behavior feature mining and recommendation method for precise business cooperation as claimed in claim 3, characterized in that: The specific steps of constructing a resource profile based on the resource feature data by using the second clustering algorithm include: Obtain resource profile feature data related to resource profile; Use a hierarchical clustering algorithm to cluster the resource portrait feature data to obtain the second clustering result; The common characteristics of the resources in each cluster in the second clustering result are analyzed to obtain different types of resource portraits.

5. The multi-dimensional user behavior feature mining and recommendation method for precise business cooperation according to claim 4, characterized in that: The specific steps of extracting a transaction feature group based on the transaction data corresponding to a certain type of user portrait include: Extract resource feature data from transaction data; Filter relevant transaction data from resource feature data based on user profile type; Normalizing the relevant transaction data to obtain processed transaction data; Extract transaction feature groups based on processed transaction data.

6. A multi-dimensional user behavior feature mining and recommendation method for precise business cooperation according to claim 5, characterized in that: The specific steps of matching the features of the transaction feature group with the resource portrait to obtain the recommended resource portrait include: Create a matching matrix, where rows represent users, columns represent resources, and each cell contains the user's behavior data on the resource; Select cosine similarity to calculate the similarity between users; Based on the similarity score, find a group of users that are most similar to the target user; Calculating resource similarity based on the features of the transaction feature group corresponding to the group of users and the resource portrait; Sort candidate resources by resource similarity; The first N resources are selected from the sorted resources as resource profiles recommended to the user to obtain a resource list.

7. A multi-dimensional user behavior feature mining and recommendation method for precise business cooperation according to claim 6, characterized in that: The specific steps of pushing the corresponding resource data to the client based on the recommended resource portrait include: Get a list of recommended resources; The resource list is further customized according to the customer's preferences to obtain a customized list; Send customized lists to customers according to the corresponding push platform.

8. The multi-dimensional user behavior feature mining and recommendation method for precise business cooperation according to claim 7, characterized in that: The specific steps of scoring resource data based on feedback from all customers include: Collect customer feedback information in the resource aggregation platform; Remove invalid feedback records from customer feedback information; The resource data score is calculated based on each score and corresponding weight in the customer feedback information.

9. A multi-dimensional user behavior feature mining and recommendation system for precise business cooperation, using a multi-dimensional user behavior feature mining and recommendation method for precise business cooperation as claimed in any one of claims 1 to 8, characterized in that: It includes data collection module, user portrait generation module, resource portrait generation module, transaction feature generation module, recommended resource portrait generation module, push module and scoring module; The data collection module is used to collect user behavior data, resource feature data and transaction data in real time. The user behavior data includes user identity data, user search data, browsing data and click data. The resource feature data is extracted from resource data. The resource feature data includes industry data, cooperation model data, resource type data and resource demand data. The user portrait generation module is used to construct a user portrait type based on the user behavior data through a first clustering algorithm, wherein the user portrait features include the user's industry background, browsing preference, click frequency, and demand type; The resource portrait generation module is used to construct a resource portrait based on the resource feature data through a second clustering algorithm, wherein the resource portrait features include the industry type of the resource, the resource type, and the potential partners of the resource; The transaction feature generation module is used to extract a transaction feature group based on the transaction data corresponding to a certain type of user portrait; The recommended resource portrait generation module is used to match the features of the transaction feature group with the resource portrait to obtain the recommended resource portrait; The push module is used to push the corresponding resource data to the client based on the recommended resource portrait; The scoring module is used to score the resource data based on the feedback from all customers.

Images