User recommendation method and system based on big data
By classifying and analyzing the user's initial behavior data, combining user basic information, and using multimodal fusion method to generate recommendation instructions, the shortcomings of the user recommendation system in the existing technology in terms of timeliness and personalization are solved, and more accurate and timely personalized recommendations improve user experience and system efficiency.
Patent Information
- Application Number
- CN202411236177.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-04
- Publication Date
- 2024-11-29
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
Existing user recommendation systems are difficult to respond quickly when processing big data, and the timeliness and personalization of recommendations are insufficient. Especially when processing massive user data and high-dimensional feature, the recommendation accuracy of new users or data sparse conditions is not high. Comprehensive analysis of diversified behavioral data and multimodal information fusion are not effective.
By obtaining the user's initial behavior data and basic information, type classification and data cleaning are performed, and the recommended features corresponding to the classified behavior data are matched. A multimodal fusion method is used to generate recommendation instructions based on the behavior type weight, and based on the initial behavior data is Optimize recommended content and adjust the recommendation list with user basic information.
Achieve more accurate and timely personalized recommendations, improve user satisfaction and relevance of recommended content, adapt to changes in user interests, maintain the freshness and attractiveness of recommended content, significantly improve user experience, increase user stickiness, and be equipped with Good scalability.
Smart Images

Figure CN119046533A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a user recommendation method and system based on big data. Background Art
[0002] In the information age, Internet platforms have a huge user base and complex user behavior data. Accurate recommendations based on these data have become an important means to increase user stickiness, improve user satisfaction and promote transaction success. User recommendation systems can use users' historical behavior data, personal information and other contextual information, and use data mining methods to provide users with personalized content recommendations to improve user experience and promote commercial realization.
[0003] In the prior art, user recommendation methods usually involve collaborative filtering, content recommendation, and hybrid recommendation algorithms, which predict items that users may be interested in by analyzing users' historical behavior data and personal information. These methods have been widely used in many e-commerce and social media platforms. However, due to the diversity and dynamic changes of users' interests and behavior patterns, existing recommendation systems still have limitations in accuracy and timeliness, especially when dealing with massive user data and high-dimensional features. It is difficult to respond quickly.
[0004] The main problems existing recommendation systems have when processing big data include being unable to handle the rapid growth and diversity of data, the timeliness and personalization of recommendations need to be improved, and the accuracy of recommendations for new users or when data is sparse. In addition, the comprehensive analysis of diverse behavioral data and the effective fusion of multimodal information are also problems that have not been well solved in existing technologies. These technical defects limit the performance of recommendation systems in practical applications. Therefore, there is an urgent need for a user recommendation method that can make full use of big data to solve these problems and provide more accurate and timely personalized recommendations. Summary of the invention
[0005] The present invention provides a user recommendation method and system based on big data, which are used to solve the above-mentioned technical problems.
[0006] A first aspect of the present invention provides a user recommendation method based on big data, and the user recommendation method based on big data comprises: Acquire the user's initial behavior data and corresponding user basic information, and classify the initial behavior data into types to obtain classified behavior data; wherein the user's initial behavior data at least includes browsing history, purchase history, click-through rate and search history; Matching the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data; Based on the preset behavior type weights, multimodal fusion is performed on each recommendation feature data to obtain a fusion result, and based on the fusion result, a recommendation instruction corresponding to the user's initial behavior data is generated; Based on the recommendation instructions corresponding to the initial behavior data and the basic information of the user, the preset user recommendation content is optimized to generate a recommendation adjustment result, and based on the recommendation adjustment result, the user's recommendation list is adjusted.
[0007] Optionally, in a first implementation of the first aspect of the present invention, the acquiring of the user's initial behavior data and the corresponding user basic information, and classifying the initial behavior data by type to obtain the classified behavior data includes: Collecting the user's initial behavior data and user basic information from multiple data sources; wherein the multiple data sources include at least website logs and mobile applications; the user basic information includes at least the user's age, gender, geographic location and interest preferences; Performing data cleaning and preprocessing on the collected initial behavior data to obtain preprocessed initial behavior data; Based on the preset data mining technology, the preprocessed initial behavior data is classified into types to obtain classified behavior data.
[0008] Optionally, in a second implementation of the first aspect of the present invention, the preprocessed initial behavior data is classified into types based on a preset data mining technology to obtain classified behavior data, including: Using a preset classification algorithm to classify the preprocessed initial behavior data according to predefined behavior types; Performing statistical analysis on the classified behavior data to extract corresponding behavior feature data; The extracted behavior feature data is subjected to feature vectorization processing to obtain classified behavior data.
[0009] Optionally, in a third implementation of the first aspect of the present invention, matching the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data includes: Inputting the classified behavior data into a preset database to obtain a first character set; Based on a preset character segmentation rule, the first character set is segmented to obtain a plurality of character intervals; wherein the character intervals are composed of a plurality of different English characters; Get the last character of each character interval as the identification character of each behavior type; For each identification character corresponding to the behavior type, the initial recommended feature corresponding to the identification character is retrieved from a database; wherein the database stores in advance a matching rule based on matching the identification character with the corresponding initial recommended feature; The similarity between the classified behavior data and the initial recommendation features is analyzed, and the recommendation features that best match the user's initial behavior data are selected to obtain each recommendation feature data.
[0010] Optionally, in a fourth implementation of the first aspect of the present invention, the character segmentation is performed on the first character set based on a preset character segmentation rule to obtain multiple character intervals, including: According to the preset character segmentation rules, define the standard and method of character segmentation; wherein the preset character segmentation rules include character length, character type, character frequency and character position; Using a machine learning-based character recognition algorithm, the characters in the first character set are scanned and recognized to obtain character patterns, and a corresponding segmentation feature value is calculated for each character pattern; wherein the segmentation feature value includes at least a relationship between the character and adjacent characters, context information of the character, and frequency statistics of the character; Perform in-depth analysis on the recognized character patterns, match them with the preset character segmentation rules, and determine the segmentation points for character segmentation; Based on the segmentation feature value and each determined segmentation point of the character segmentation, the first character set is dynamically segmented to generate a plurality of character intervals.
[0011] A second aspect of the present invention provides a user recommendation system based on big data, and the user recommendation system based on big data includes: A classification module is used to obtain the user's initial behavior data and the corresponding user basic information, and classify the initial behavior data into types to obtain the classified behavior data; wherein the user's initial behavior data at least includes browsing history, purchase history, click-through rate and search history; A matching module, used to match the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data; A fusion module, used to perform multimodal fusion on each recommendation feature data based on a preset behavior type weight to obtain a fusion result, and generate a recommendation instruction corresponding to the user's initial behavior data based on the fusion result; The adjustment module is used to optimize the preset user recommendation content based on the recommendation instructions corresponding to the initial behavior data and the basic information of the user, generate a recommendation adjustment result, and adjust the user's recommendation list based on the recommendation adjustment result.
[0012] The third aspect of the present invention provides a user recommendation device based on big data, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the user recommendation device based on big data executes the above-mentioned user recommendation method based on big data.
[0013] A fourth aspect of the present invention provides a computer-readable storage medium, in which instructions are stored, and when the computer-readable storage medium is run on a computer, the computer executes the above-mentioned user recommendation method based on big data.
[0014] In the technical solution provided by the present invention, the beneficial effects are as follows: the present invention provides a user recommendation method and system based on big data, by obtaining the initial behavior data of the user and the corresponding basic information of the user, and classifying the initial behavior data into types to obtain the classified behavior data; matching the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data; based on the preset behavior type weight, multimodal fusion is performed on each recommendation feature data to obtain a fusion result, and based on the fusion result, a recommendation instruction corresponding to the initial behavior data of the user is generated; based on the recommendation instruction corresponding to the initial behavior data and the basic information of the user, the preset user recommendation content is optimized to generate a recommendation adjustment result, and based on the recommendation adjustment result, the user's recommendation list is adjusted. By classifying and analyzing the initial behavior data of the user and combining the basic information of the user, the present invention can provide more personalized recommendations, which helps to increase user satisfaction and improve the relevance of the recommended content. The multimodal fusion method is used to comprehensively consider various behavior type weights, so that the recommendation system can more accurately capture the user's true intentions and preferences, thereby generating more accurate recommendation instructions. Based on the big data processing and analysis capabilities, the matching and fusion of recommendation feature data can be completed in a short time, realizing a rapid response to user needs and improving the operating efficiency of the entire recommendation system. According to the recommendation instructions and basic user information, the recommended content is continuously optimized. The present invention can adapt to changes in user interests and adjust the recommendation list in time to keep the recommended content fresh and attractive. By regularly adjusting the recommendation list to keep the content novel and personalized, the present invention can significantly improve the user experience and increase the user's stickiness to the platform. The present invention is suitable for processing large-scale user data, can easily cope with the processing needs of massive data, has good scalability, and is suitable for an expanding user group. BRIEF DESCRIPTION OF THE DRAWINGS
[0015] Figure 1 A schematic diagram of an embodiment of a user recommendation method based on big data in an embodiment of the present invention; Figure 2Schematic diagram of an embodiment of a user recommendation system based on big data in an embodiment of the present invention. DETAILED DESCRIPTION
[0016] Embodiments of the present invention provide a method and system for user recommendation based on big data. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.
[0017] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the user recommendation method based on big data in an embodiment of the present invention includes: Step 101: obtaining the initial behavior data of the user and the corresponding basic information of the user, and classifying the initial behavior data by type to obtain the classified behavior data; wherein the initial behavior data of the user at least includes browsing history, purchase history, click-through rate and search history; It is understandable that the execution subject of the present invention may be a user recommendation system based on big data, or may be a terminal or a server, which is not specifically limited here. The embodiment of the present invention is described by taking a server as the execution subject as an example.
[0018] Specifically, step 101 is described in detail as follows: Get the user's initial behavior data: Data source collection: Collect users’ initial behavior data through Internet applications, e-commerce platforms, social media and other digital user interaction channels. Common data acquisition methods include server log records, client tracking codes (such as browser cookies, mobile application SDKs) and third-party data cooperation.
[0019] Behavioral data collection: Browsing history: records users’ browsing activities on web pages, including URLs of visited pages, dwell time, scroll depth, etc.
[0020] Purchase records: Collect users’ purchase behavior data on e-commerce platforms, including purchased items, purchase time, consumption amount, order status, etc.
[0021] Click-through rate: Tracks the user's behavioral data of clicking on various interactive elements, such as ads, buttons, links, etc., recording the click location, time and frequency.
[0022] Search history: records the user's search request history within the site or using search engines, including search keywords, search time, result clicks, etc.
[0023] Get basic user information: User information collection: Collect basic user information from user registration information, login activities, third-party authentication services (such as OAuth), etc. Basic information includes but is not limited to user ID, name, age, gender, region, interests, etc.
[0024] Data Integration: Integrate user behavior data obtained from different sources with user basic information, and associate and match them with the user's unique identifier (such as user ID) to ensure that all data can be uniformly attributed to a specific user.
[0025] Store and manage integrated user data through database management systems (such as MySQL, PostgreSQL) or data warehouses (such as Amazon Redshift, Google BigQuery).
[0026] Behavioral data type classification: Classification algorithm design: Use pre-designed classification rules or machine learning algorithms to classify the initial behavior data. Classification rules can be based on data fields or labels (such as URL structure, click element type, etc.), or can be automatically classified by training classification models (such as decision trees, SVMs, neural networks, etc.).
[0027] Data processing: Pre-process the collected behavioral data such as browsing history, purchase history, click-through rate and search history, such as data cleaning, format conversion, unified timestamp, etc.
[0028] Apply classification rules and algorithms to classify behavioral data and mark its type. For example, browsing records are classified into different page visit types (such as product pages, homepages, article pages, etc.), and purchase records are classified into different consumer categories (such as electronic products, daily necessities, clothing, etc.).
[0029] Storage and Management: The classified behavior data is stored in the corresponding database or data warehouse according to the predetermined data model design. In addition to the original data, each data record must also be accompanied by its classification label to facilitate subsequent statistical analysis and model training.
[0030] Use distributed databases (such as HBase, Cassandra) or big data platforms (such as Hadoop, Spark) to process large-scale user behavior data and ensure the efficiency and scalability of the classification process.
[0031] Example description: Example background: Assume that an e-commerce platform needs to classify user behavior data to improve the service of the personalized recommendation system. The initial behavior data of a user is obtained: browsing history includes visiting the homepage, viewing multiple product detail pages, and visiting the shopping cart page; purchase history includes purchasing two electronic products; click-through rate records include clicking on multiple advertisements and recommended links; search history includes searching for smartphones and Bluetooth headsets.
[0032] These data are associated and integrated with the user's basic information (for example, user ID is 12345, gender is male, age is 30, and region is Shanghai).
[0033] Using preset classification rules, for example, according to product categories and access paths, browsing records are classified into "home page browsing", "product page browsing", and "shopping cart page browsing"; purchase records are classified into "consumer electronics"; click-through rates are classified into "advertising clicks" and "recommended link clicks"; and search histories are classified into "smartphone searches" and "Bluetooth headset searches".
[0034] The classified behavior data is stored in a distributed database and labeled to facilitate subsequent recommendation model training and online service.
[0035] Step 102: matching recommendation features corresponding to the classified behavior data in a preset database to obtain various recommendation feature data; Specifically, step 102 is described in detail as follows: In step 101, the type classification of the user's initial behavior data has been completed. Next, it is necessary to match the recommendation features corresponding to the classified behavior data in the preset database to obtain each recommendation feature data. The specific implementation process is as follows: Construction and management of preset database: The preset database contains detailed information about the recommendation features and the mapping relationship between these features and user behaviors.
[0036] The design of the preset database needs to be reasonably structured, and usually a relational database (such as MySQL, PostgreSQL) or a NoSQL database (such as MongoDB, Cassandra) is used for storage and management.
[0037] In the initial stage of database establishment, a recommended feature table needs to be constructed, which contains fields such as feature ID, feature name, feature description, and user behavior type associated with the feature.
[0038] For example, the feature table contains the following fields: feature ID, feature name, description, associated behavior type, etc.; specific content such as: feature ID 1, the corresponding feature name is "technology and fashion interest users", the description is "users browsing new technology product pages or purchasing technology products", and the associated behavior type is "technology product browsing records, technology product purchase records".
[0039] The mapping relationship between behavior data and recommendation features: Design and store the mapping relationship between behavior data and recommendation features, and associate each classified behavior data with the corresponding recommendation feature.
[0040] The mapping relationship can be set through preset rules or trained models, which are derived from big data analysis and domain expert knowledge.
[0041] For example, the behavior data of users clicking on advertisements related to "smartphones" can be mapped to the recommendation feature of "users interested in electronic products"; the behavior of users searching for "baby toys" can be mapped to the recommendation feature of "parenting family users".
[0042] Match the classified behavior data: The classified behavior data obtained in step 101 is extracted and matched in turn in the recommendation feature table in the preset database.
[0043] The query process is implemented through SQL statements or database query APIs, and the corresponding recommendation features are located based on the classified behavior data type and content.
[0044] Database queries require efficient indexing and optimization, such as creating indexes for feature ID and behavior type fields in the feature table to speed up queries.
[0045] Get recommended feature data: After the matching is completed, the recommendation features corresponding to the classified behavior data are extracted from the database to form a recommendation feature data set.
[0046] The recommendation feature dataset includes detailed information of all recommendation features that match user behaviors and may contain multiple feature items.
[0047] For further processing and use, these data can be organized into appropriate data structures (such as dictionaries, lists, etc.) for subsequent recommendation algorithm calls.
[0048] For example, feature ID 1 ("users interested in technology and fashion") and feature ID 3 ("users interested in electronic products") are obtained through query and jointly formed into a recommended feature dataset.
[0049] Example description: Example background: Assume that an e-commerce platform needs to generate recommendation features based on user behavior data for personalized push notifications. User A browsed multiple pages of electronic technology products, searched for the keyword "smartphone", and clicked on advertisements related to "5G mobile phone" several times.
[0050] These behavioral data of user A are classified into "electronic product browsing records", "smartphone search records", and "5G-related advertising click records".
[0051] In the preset database, the mapping relationship between the behavior data and the feature information is matched, and the recommended feature data is queried from the database. For example, the recommended features obtained by mapping include feature ID 1 (described as "users with a high interest in technology products") and feature ID 4 (described as "users who pay special attention to smartphones").
[0052] The system integrates and returns these recommendation feature data for use in subsequent recommendation models.
[0053] Storing recommended feature data: The obtained recommendation feature data set is stored in a temporary data storage module (such as the memory database Redis) or a persistent storage module for subsequent recommendation strategy processing.
[0054] Data storage needs to consider access and call efficiency to ensure fast query and operation in subsequent real-time recommendation processes.
[0055] Step 103: Based on the preset behavior type weights, perform multimodal fusion on each recommendation feature data to obtain a fusion result, and based on the fusion result, generate a recommendation instruction corresponding to the user's initial behavior data; Specifically, step 103 is described in detail as follows: In step 102, each recommendation feature data has been obtained. Next, it is necessary to perform multimodal fusion on these recommendation feature data based on the preset behavior type weights to obtain a fusion result, and generate a recommendation instruction corresponding to the user's initial behavior data based on the fusion result. The specific implementation process is as follows: Obtaining the preset behavior type weight: The preset behavior type weight is a weight value pre-set by the system based on the importance of the user behavior and its potential impact on the recommendation results.
[0056] The weight setting can be obtained through expert experience, historical data analysis or machine learning model training. For example, the weight of browsing history may be 0.3, the weight of purchase history may be 0.5, the weight of click-through rate may be 0.1, and the weight of search history may be 0.1.
[0057] These weights need to be stored in a system configuration file or database so that they can be called when performing multimodal fusion.
[0058] Recommended feature data processing: The recommended feature data obtained in step 102 includes a detailed description of each feature and the corresponding classified behavior data. Ensure that these data have been standardized for fusion calculation.
[0059] All recommended feature data must be converted into an appropriate calculation form based on its original form (such as numerical type, categorical type, etc.), for example, through normalization, standardization and other methods.
[0060] Multimodal fusion model design: The multimodal fusion model is an algorithm framework for comprehensive processing of multi-source and multi-type data. Commonly used models include weighted sum model, linear regression model, neural network, etc.
[0061] The present invention mainly adopts the weighted summation method for preliminary fusion, and can be combined with more complex algorithms for optimization.
[0062] The specific steps are as follows: For each recommended feature data, multiply it by the corresponding behavior type weight. Assume that recommended feature A is generated by browsing records and has a weight of w1, and feature B is generated by purchase records and has a weight of w2. Calculate the product of each feature data and its weight.
[0063] Sum up all the product results to get the fused result vector.
[0064] Calculate the fusion result: All recommended feature data are processed comprehensively through weighted calculation: Determine the weighted value of each recommended feature data item: Weighted value = recommended feature data * preset behavior type weight.
[0065] Sum up all weighted values: fusion result = weighted value collection.
[0066] For example: Suppose there are three recommended feature data, namely feature A (weight 0.3), feature B (weight 0.5), and feature C (weight 0.2).
[0067] The value obtained by weighted calculation is: 0.3 * Feature A + 0.5 * Feature B + 0.2 * Feature C.
[0068] The weighted sum obtained is the user's multimodal fusion result.
[0069] Generate recommended instructions: Generate recommendation instructions based on the fusion results. Recommendation instructions are instructions for the system to provide users with personalized recommendation content based on the fusion results.
[0070] The recommendation instruction includes information such as the recommended content list, recommendation strategy, and recommended path. When generating the recommendation instruction, the recommended content needs to be sorted and filtered according to the weight value and feature value in the fusion result.
[0071] For example, if the fusion results show that the user is most interested in "electronic products", the system will generate instructions to recommend electronic products and refine the specific product list.
[0072] Example description: Example background: Suppose an e-commerce platform generates personalized recommendation content for users. User behavior data includes: browsing smartphone pages (weight 0.3), purchasing headphones (weight 0.5), clicking on technology news ads (weight 0.1), and searching for Bluetooth speakers (weight 0.1).
[0073] The weighted values are calculated according to the preset behavior type weights: 0.3 * browsing smartphone page feature value + 0.5 * buying headphones feature value + 0.1 * clicking on technology news advertisement feature value + 0.1 * searching for Bluetooth speakers feature value.
[0074] Sum up the weighted values to get the fused result vector.
[0075] Generate recommendation instructions based on the fusion results, such as recommending a list of smartphones and accessories.
[0076] Recommended instruction delivery and execution: The generated recommendation instructions are passed to the user recommendation system, and the recommendation system dynamically generates recommendation pages or pushes information based on the instruction content.
[0077] Usually, recommendation instructions are passed to the front-end display module through API calls or message queues and executed according to the established recommendation strategy.
[0078] Step 104: Based on the recommendation instructions corresponding to the initial behavior data and the basic information of the user, the preset user recommendation content is optimized to generate a recommendation adjustment result, and based on the recommendation adjustment result, the user's recommendation list is adjusted.
[0079] Specifically, step 104 is described in detail as follows: In step 103, recommendation instructions have been generated based on the initial behavior data. Next, based on these recommendation instructions and the user's basic information, the preset user recommendation content needs to be optimized, a recommendation adjustment result is generated, and the user's recommendation list is adjusted based on the recommendation adjustment result. The specific implementation process is as follows: Get recommended instructions and basic user information: The generated recommendation instructions are obtained from step 103, and these instructions include personalized recommendation content and strategies for the user.
[0080] Read the user's basic information from the user database or user portrait system. Basic information includes the user's basic attributes (such as age, gender, place of residence, etc.), interest preferences, historical interaction records, etc.
[0081] Ensure that the data format of the recommended instructions is consistent with the user's basic information to facilitate subsequent processing.
[0082] Preparation of preset recommended content library: Build and manage a library of preset recommended content, which contains a variety of recommended content items.
[0083] The recommended content library can be stored in a relational database (such as PostgreSQL) or a non-relational database (such as MongoDB), supporting efficient query and update operations.
[0084] Each content item contains detailed information such as content ID, content type, content description, related tags, applicable user groups, etc. For example, the electronic product with content ID 101 is of type smartphone, described as "the latest 5G smartphone", tagged as "technology, electronic products", and applicable user groups include "18-35 years old, male, technology enthusiasts".
[0085] Recommended content matches user characteristics: Based on the user features extracted from the user's basic information, the preset recommended content library is screened and matched. Matching algorithms or rules (such as matching based on content similarity, collaborative filtering, etc.) are used to find content items that are highly matched with the user's basic information.
[0086] Matching rules can include cold start strategies. When a user's historical behavior data is relatively small or the user is using the app for the first time, content recommendations are made based on the user's basic attributes such as age, gender, and interest tags.
[0087] Optimize recommended content: Use recommendation optimization algorithms to optimize and adjust the initial recommended content. Optimization algorithms can include weighted scoring, multi-armed bandit algorithms, reinforcement learning, etc.
[0088] The specific process is as follows: Assign a score to the recommended content item. The score is calculated based on user behavior data (such as log information such as clicks, browsing, and purchases) and basic user information (such as interest tags, geographic location, etc.).
[0089] According to the set strategies (such as exploration and exploitation balance strategy, diversity enhancement strategy, etc.), the sorting and screening of content items are adjusted to generate optimized recommendation adjustment results.
[0090] For example, the reinforcement learning model will continuously adjust the scoring strategy through real-time feedback (such as user clicks and dwell time) to ensure dynamic optimization of recommendations.
[0091] Generate recommended adjustment results: Based on the optimization algorithm, a recommendation adjustment result is generated, which includes the optimized recommended content items and their weights or scores. The recommendation adjustment result reflects the final recommended content and the best ranking.
[0092] The adjustment results can be organized into a structured data format (such as JSON, XML) for easy transmission and processing.
[0093] Adjust the user recommendation list: According to the recommendation adjustment results, the user's recommendation list is dynamically adjusted. The recommendation list is the sorting and display format of the recommended content that users see in the application.
[0094] The processing process includes the generation, update and adjustment of the recommendation list to ensure that the recommended content is highly consistent with the user's actual interests and improve the user experience.
[0095] Specific adjustment steps: Read the existing recommendation list and compare the content items and scores in the recommendation adjustment results.
[0096] Adjust the order of content items in the list, add new recommended items, and remove low-rated items.
[0097] Update the user's recommendation list through the interface or API to ensure that the content displayed on the front end of the recommendation system remains up to date.
[0098] Example description: Example background: An e-commerce platform needs to optimize recommended content based on recommendation instructions and user basic information. Get the recommendation instructions to recommend high-tech products. At the same time, extract the user's favorite tags of "smartphones" and "digital accessories" from the user's basic information.
[0099] Filter out highly matching content items from the content library, such as the latest smartphones, wireless headphones, smart watches, etc.
[0100] The multi-arm gambling algorithm is used to optimize the ranking of recommended content, and the results show that smartphones have the highest priority.
[0101] Generate recommended adjustment results, including smartphones (score 9.5), wireless headphones (score 8.0), and smart watches (score 7.0).
[0102] Adjust the user's recommendation list to ensure that smartphones appear first in the list of recommendations, followed by wireless headphones and smart watches.
[0103] Real-time update and monitoring of recommendation lists: Deploy real-time data processing modules to ensure that the recommendation system can dynamically respond to user behavior and feedback and continuously update the recommendation list.
[0104] Monitor recommendation results in real time, collect user interaction data (such as clicks, browsing, and retention) for the next round of recommendation optimization, and gradually improve the intelligence of the recommendation system.
[0105] Another embodiment of the user recommendation method based on big data in the embodiment of the present invention includes: obtaining the initial behavior data of the user and the corresponding basic information of the user, and classifying the initial behavior data by type to obtain the classified behavior data, including: Collecting the user's initial behavior data and user basic information from multiple data sources; wherein the multiple data sources include at least website logs and mobile applications; the user basic information includes at least the user's age, gender, geographic location and interest preferences; Performing data cleaning and preprocessing on the collected initial behavior data to obtain preprocessed initial behavior data; Based on the preset data mining technology, the preprocessed initial behavior data is classified into types to obtain classified behavior data.
[0106] Specifically, another embodiment of the user recommendation method based on big data in the embodiment of the present invention is described in detail as follows: The process of obtaining the initial behavior data of the user and the corresponding basic information of the user, and classifying the initial behavior data by type to obtain the classified behavior data includes the following specific steps: Collect users' initial behavior data and basic information from multiple data sources: Data source design: Website logs: The web server's log files record user access records, click behaviors, and other detailed information. These logs usually contain the URLs visited by users, IP addresses, timestamps, request methods (GET / POST), user agent strings, etc.
[0107] Mobile applications: Collect user usage behavior data through the SDK integrated in the mobile application or the built-in log function, including user application launch, page browsing, button clicks, etc., and may even include advanced information such as screen scrolling depth, touch events, etc.
[0108] Other data sources: can be expanded to social media platforms, third-party data providers, etc.
[0109] Collection of basic user information: User basic information is mainly obtained from user registration forms, third-party authorized logins, etc. The information is stored in the user management database and includes fields such as user ID, age, gender, geographic location, interest preferences, etc.
[0110] For example, request users to fill in basic information through the user registration page; use the OAuth protocol to access third-party social media accounts to obtain basic information.
[0111] Clean and preprocess the collected initial behavior data: Data cleaning: Remove redundant data: Delete duplicate records and retain unique behavior data.
[0112] Data format standardization: unify the formats of various data sources and convert data from different sources into a unified storage format.
[0113] Abnormal data processing: Detect and delete or correct abnormal data points (such as invalid timestamps, illegal IP addresses, etc.).
[0114] Data preprocessing: Time synchronization: Synchronize data from different sources based on timestamps to ensure data consistency.
[0115] Data conversion: Perform necessary conversions on raw data, such as parsing browsed URLs into structured data (such as page categories, product IDs, etc.).
[0116] Feature engineering: Extract important feature variables, such as extracting page dwell time from browsing history, extracting click frequency from click behavior, etc.
[0117] Based on the preset data mining technology, the preprocessed initial behavior data is classified into types: Preset data mining techniques: Classify the preprocessed data using classification algorithms such as decision tree, support vector machine, logistic regression, etc.
[0118] Machine learning can be used for training to build models that can classify user behavior data.
[0119] Specific implementation method: Build data labels: Label behavior labels according to preset rules, for example, define the behavior of "browsing smartphone product pages" as the "electronic product interest" label.
[0120] Algorithm implementation: Input the behavior data into the trained classification model and classify each piece of behavior data.
[0121] Classification processing: The classification results are labeled and stored in the database to ensure the consistency and completeness of data types.
[0122] Example description: Assume that user A has multiple behavior records in the access data of an e-commerce platform.
[0123] These records are converted into structured data after data cleaning and preprocessing.
[0124] Use the trained decision tree model to classify user A's behavior data (such as browsing history, purchase history, etc.).
[0125] The final classification results may include behavioral categories such as "smartphone browsing" and "headphone purchasing".
[0126] Example process: Data collection: When a user visits an e-commerce website, the server log records the user's various behaviors on the page, including browsing product A, clicking on advertisement B, etc.
[0127] The user uses the mobile app to record his purchase behavior of product C and sends this data to the server through the mobile SDK.
[0128] Data cleaning and preprocessing: After receiving the user's access log and application data, the server performs data cleaning and processes duplicate records and invalid data.
[0129] The preprocessed data is converted into structured information, such as extracting features such as access time, page type, and purchase items.
[0130] Type classification: The preprocessed data is input into the trained classification model, and the classification model identifies the user's behavior type.
[0131] Classification results, such as “browsing electronic products,” “clicking on technology news ads,” and “purchasing digital accessories,” are stored in the behavior database for subsequent use.
[0132] Through the above detailed steps, the embodiment of the present invention can collect the user's initial behavior data and user basic information from multiple data sources, perform data cleaning and preprocessing on the initial behavior data, and then classify it based on the preset data mining technology to obtain the classified behavior data. These steps ensure the accuracy and refined processing of the behavior data, and provide a reliable data foundation for the subsequent personalized recommendation algorithm. For example, on an e-commerce platform, these steps can be used to efficiently analyze user behavior and improve the accuracy of the recommendation system and user satisfaction. In various online services, this method can also be used to conduct a comprehensive analysis of user interests to achieve differentiated services and precision marketing.
[0133] Another embodiment of the user recommendation method based on big data in the embodiment of the present invention includes: based on the preset data mining technology, the pre-processed initial behavior data is classified into types to obtain the classified behavior data, including: Using a preset classification algorithm to classify the preprocessed initial behavior data according to predefined behavior types; Performing statistical analysis on the classified behavior data to extract corresponding behavior feature data; The extracted behavior feature data is subjected to feature vectorization processing to obtain classified behavior data.
[0134] Specifically, the user recommendation method based on big data in the embodiment of the present invention is further described in detail as follows: The process of classifying the preprocessed initial behavior data into types based on the preset data mining technology to obtain the classified behavior data includes the following specific steps: The pre-processed initial behavior data is classified according to the pre-defined behavior types using a preset classification algorithm: Classification algorithm selection: Commonly used classification algorithms include decision trees, random forests, support vector machines (SVMs), logistic regression, and naive Bayes.
[0135] You can choose a suitable classification algorithm based on the data characteristics and application scenarios, or you can combine multiple algorithms to build a hybrid model to improve classification accuracy. For example, you can use ensemble learning methods (such as Bagging and Boosting) to improve the generalization ability of the model.
[0136] Classification model training: Leverage historical user behavior datasets to label training data according to predefined behavior types (e.g., browsing, clicking, purchasing, searching).
[0137] The classification model is trained, and the model parameters are optimized through cross-validation, grid search and other methods to select the best model.
[0138] The trained model is stored in the model warehouse for real-time classification applications.
[0139] Classification process: The preprocessed initial behavior data is input into the trained classification model for prediction, and the model outputs the classification result of each behavior data item.
[0140] For example, a user's purchase record is identified as "purchase behavior" after being classified by the model; a user's search record is identified as "search behavior".
[0141] Perform statistical analysis on the classified behavior data to extract the corresponding behavior feature data: Statistical analysis methods: The classified behavioral data is statistically analyzed through methods such as aggregate query, frequency statistics, and time window analysis.
[0142] Descriptive statistics: Calculate various statistics of data, such as mean, median, standard deviation, frequency, proportion, etc.
[0143] Cross-dimensional statistics: multi-dimensional cross-analysis based on different dimensions of data (such as time, location, behavior type, etc.).
[0144] Feature extraction: Extract key behavioral features from the results of statistical analysis, such as behavior frequency, behavior duration, behavior distribution, etc.
[0145] For example, features such as "average page browsing time of users", "number of daily page views of users", and "main categories browsed by users" can be extracted from users' browsing behavior data.
[0146] Example description: Assume that user A's behavior data in the past month includes: browsing electronic products 20 times, purchasing 3 times, and clicking on advertisements 5 times.
[0147] After statistical analysis, the following behavioral characteristic data can be extracted: "average number of views per week", "purchase frequency", "ad click frequency", etc.
[0148] Perform feature vectorization on the extracted behavior feature data to obtain classified behavior data: Feature vectorization method: The extracted behavioral feature data is standardized and converted into a unified vector form.
[0149] Vectorization processing includes normalization, standardization, binarization, regularization, etc., which makes the data have a unified scale and format, facilitating subsequent machine learning model processing.
[0150] Specific steps of feature vectorization: Standardization: Standardize the feature data so that its distribution conforms to the standard normal distribution (mean is 0 and standard deviation is 1).
[0151] Normalization: Map feature data to the [0,1] interval, and use linear scaling to make different features have the same scale. For example, the value range of a behavior frequency feature is 0 to 100, which is mapped to the range of 0 to 1 through normalization.
[0152] Create feature vector: Combine the standardized and normalized feature data into a vector as the feature vector of the classified behavior data. After each user's behavior feature data is vectorized, it forms a feature point in a high-dimensional space.
[0153] Example description: Assume that for user A, the extracted behavioral features include "average number of views per week (20 times)", "purchase frequency (3 times)", and "ad click frequency (5 times)".
[0154] These features are standardized, for example, the standardized feature vector is [1.2, -0.8, 0.5].
[0155] The final feature vector represents the position of user A in the behavior space, which is convenient for subsequent recommendation algorithm processing.
[0156] Through the above detailed steps, the present invention can classify the preprocessed initial behavior data based on the preset data mining technology, extract and vectorize the behavior features, and obtain the classified behavior data. These steps ensure the systematic and refined data processing, and provide high-quality input data for the subsequent personalized recommendation algorithm. For example, on an e-commerce platform, this method can effectively identify users' shopping preferences and browsing habits, realize personalized product recommendations, and improve user satisfaction and purchase conversion rate. On a content recommendation platform, these steps can accurately analyze users' interests and behavior patterns, provide highly customized content recommendations, and enhance user experience and platform stickiness.
[0157] Another embodiment of the user recommendation method based on big data in the embodiment of the present invention includes: The matching of recommendation features corresponding to the classified behavior data in a preset database to obtain various recommendation feature data includes: Inputting the classified behavior data into a preset database to obtain a first character set; Based on a preset character segmentation rule, the first character set is segmented to obtain a plurality of character intervals; wherein the character intervals are composed of a plurality of different English characters; Get the last character of each character interval as the identification character of each behavior type; For each identification character corresponding to the behavior type, the initial recommended feature corresponding to the identification character is retrieved from a database; wherein the database stores in advance a matching rule based on matching the identification character with the corresponding initial recommended feature; The similarity between the classified behavior data and the initial recommendation features is analyzed, and the recommendation features that best match the user's initial behavior data are selected to obtain each recommendation feature data.
[0158] Specifically, another embodiment of the user recommendation method based on big data in the embodiment of the present invention is described in detail as follows: The step of matching the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data includes the following specific contents: Input the classified behavior data into the preset database to obtain the first character set: Behavior data characterization: The classified behavior data is characterized, and each behavior type and corresponding information is converted into a string form.
[0159] An example of characterization of behavioral data: Browsing records are encoded in a specific format, for example, "BRW-20231001-PROD123", where "BRW" represents the browsing behavior type, "20231001" is the date, and "PROD123" is the product ID.
[0160] Enter the database: The characterized classification behavior data is input into a preset database and stored as a character set.
[0161] For example, multiple behavior records of a user are converted into a set of character sets: [“BRW-20231001-PROD123”, “CLK-20231002-AD456”, “PUR-20231003-PROD123”].
[0162] Based on a preset character segmentation rule, the first character set is segmented to obtain a plurality of character intervals: Character segmentation rule settings: The preset character segmentation rules usually use separators (such as "-", "_", etc.) to ensure that the string is divided into multiple parts according to a predefined format.
[0163] Rule example: In the character format of "behavior type-time-details", "-" is used as a separator.
[0164] Perform character segmentation: Each character string of the first character set is segmented using a separator to obtain multiple character intervals.
[0165] For example, the string "BRW-20231001-PROD123" is split into three intervals: ["BRW", "20231001", "PROD123"].
[0166] Get the last character of each character range as the identification character of each behavior type: Definition of identification characters: The last character is extracted from each segmented character interval as an identification character of the behavior type.
[0167] Extract the identifying characters: Characters are directly taken from the behavior type interval. For example, "BRW" represents browsing behavior, and its last character is "W".
[0168] After the identification characters are extracted, for example, an identification character set is obtained: ["W", "K", "R"], which respectively correspond to the behavior types such as "browse", "click", and "purchase".
[0169] For each identification character corresponding to the behavior type, the initial recommended features corresponding to the identification character are retrieved from the database: Database structure and matching rules: In the preset database, recommendation feature matching rules based on identification characters are stored in advance, for example, the character "W" matches the "browsing behavior recommendation feature".
[0170] Matching rule example: The recommendation feature corresponding to the character "W" is "similar products that users often browse", the feature corresponding to "K" is "ad categories that users often click", and the feature corresponding to "R" is "product categories that users often buy".
[0171] Retrieve initial recommended features: Based on the identification characters, the corresponding initial recommendation features are retrieved from the database using the preset matching rules.
[0172] Retrieval example: For the identification character "W", the query obtains the recommended features of similar products that users often browse.
[0173] Analyze the similarity between the classified behavior data and the initial recommendation features, select the recommendation features that best match the user's initial behavior data, and obtain the data for each recommendation feature: Similarity analysis method: The similarity between user behavior data and initial recommendation features is analyzed using methods such as content similarity calculation, collaborative filtering algorithm, and similarity measurement based on vector space model.
[0174] Commonly used similarity calculation methods include cosine similarity, Jaccard similarity, Euclidean distance, etc.
[0175] Select the best matching recommendation feature: Based on the similarity analysis results, select the initial recommendation features that best match the user's initial behavior data.
[0176] Selection criteria: Usually set a similarity threshold or select the top N recommended features with the highest similarity.
[0177] Example: Assume that the calculated similarity example is "Similarity = 0.85" and the recommended feature "Smartphone Similar Products" is a better match than the recommended feature "Technology Accessories Interest" with "Similarity = 0.75".
[0178] Example process: Behavior data characterization and input: A user visited an e-commerce website, browsed the product "e-book reader", clicked on the advertisement "smart watch", and purchased a "Bluetooth headset".
[0179] These behavioral data are entered into the database in character form to form a character set: [“BRW-20231001-EB123”, “CLK-20231002-AD456”, “PUR-20231003-BT789”].
[0180] Character segmentation and identification character extraction: Split the character set according to “-”, extract the last character in the interval, and form the identification character set: [“W”, “K”, “R”].
[0181] Retrieve initial recommended features: Based on the identification character "W", search for "similar recommendations for electronic products", based on "R", search for "recommendations for digital accessories", etc.
[0182] Similarity calculation and recommendation feature selection: The similarity between the user's browsing of "e-book reader" and the recommended "similar electronic product recommendations" is calculated through the matrix decomposition model.
[0183] Select the recommended features with the highest similarity and generate a recommended feature set for personalized recommendation.
[0184] Through the above detailed steps, the present invention can match the recommendation features corresponding to the classified behavior data in a preset database to obtain various recommendation feature data. These steps ensure the systematic and refined data processing, and significantly improve the accuracy of the recommendation system and user experience. For example, on an e-commerce platform, this method can identify the user's specific interests, recommend more accurate products, and improve purchase conversion rates and user satisfaction. On a content recommendation platform, by accurately matching user interests, more appropriate content can be recommended to enhance user stickiness and interactivity.
[0185] Another embodiment of the user recommendation method based on big data in the embodiment of the present invention includes: performing character segmentation on the first character set based on a preset character segmentation rule to obtain multiple character intervals, including: According to the preset character segmentation rules, define the standard and method of character segmentation; wherein the preset character segmentation rules include character length, character type, character frequency and character position; Using a machine learning-based character recognition algorithm, the characters in the first character set are scanned and recognized to obtain character patterns, and a corresponding segmentation feature value is calculated for each character pattern; wherein the segmentation feature value includes at least a relationship between the character and adjacent characters, context information of the character, and frequency statistics of the character; Perform in-depth analysis on the recognized character patterns, match them with the preset character segmentation rules, and determine the segmentation points for character segmentation; Based on the segmentation feature value and each determined segmentation point of the character segmentation, the first character set is dynamically segmented to generate a plurality of character intervals.
[0186] Specifically, another embodiment of the user recommendation method based on big data in the embodiment of the present invention is described in detail as follows: The process of performing character segmentation on the first character set based on a preset character segmentation rule to obtain a plurality of character intervals includes the following specific steps: According to the preset character segmentation rules, define the standard and method of character segmentation: Character segmentation rule settings: Character length: Set the character length limit for segmentation, such as fixed length or variable length segmentation.
[0187] Character Type: Identify and define different types of characters, such as letters, numbers, special symbols, etc., which are used as the basis for segmentation.
[0188] Character frequency: Count the frequency of occurrence of characters. Characters with higher frequencies may be used as segmentation points.
[0189] Character position: Considering the position of the character in the string, the characters at the beginning, middle, and end may be assigned different segmentation weights.
[0190] Example of segmentation rules: For example, if the string is in the form of "behavior type-time-details", preliminary segmentation may be performed based on the separator "-".
[0191] A character recognition algorithm based on machine learning is used to scan and recognize the characters in the first character set to obtain character patterns, and the corresponding segmentation feature value is calculated for each character pattern: Character recognition algorithm: Use machine learning algorithms such as convolutional neural networks (CNN) or long short-term memory networks (LSTM) for character recognition.
[0192] During the training phase, the model learns the relationships and features between characters through a large amount of sample data so that character patterns can be accurately recognized in practical applications.
[0193] Scan and Identify: The characters in the first character set are scanned in all directions to identify the character type, position and frequency one by one.
[0194] Use the trained model to perform real-time recognition of character data; for example, recognize that "B" represents the behavior type, "2" represents the beginning of the year, and "-" is a separator.
[0195] Eigenvalue calculation: Calculate multiple segmentation feature values, including the relationship between characters and adjacent characters, the context information of characters, the frequency of character occurrence, etc.
[0196] For example, in the string "BRW-20231001-PROD123", the adjacency relationship between "B" and "R" is identified, and the analysis context is "behavior type".
[0197] Perform in-depth analysis on the recognized character patterns, match them with the preset character segmentation rules, and determine the segmentation points for character segmentation: Character pattern in-depth analysis: Use deep learning models to analyze character patterns step by step to determine adjacent character relationships, contextual background, and frequency information.
[0198] Detect key features in the pattern, such as high-frequency separator "-", common date format "YYYYMMDD", etc.
[0199] Matching preset rules: The recognized character patterns are matched one by one with the preset character segmentation rules to determine the accurate segmentation points.
[0200] For example, if “-” is detected as a separator and satisfies the preset segmentation rule, the position can be set as the segmentation point.
[0201] Based on the segmentation feature value and each determined character segmentation point, the first character set is dynamically segmented to generate multiple character intervals: Dynamic segmentation algorithm: Based on the determined segmentation points, a dynamic segmentation algorithm is used to segment the character set to generate character intervals of different ranges.
[0202] Dynamic segmentation takes into account the relationship between characters and contextual information to ensure segmentation accuracy and avoid information loss or mis-segmentation.
[0203] Generate character range: The characters between each split point are a character interval, which maintains the logical order of the original string.
[0204] For example, three character intervals are generated from "BRW-20231001-PROD123": ["BRW", "20231001", "PROD123"].
[0205] Example description: Data collection and characterization: The generated behavior data strings are "BRW-20231001-PROD123", "CLK-20231002-AD456", "PUR-20231003-BT789".
[0206] Character Recognition and Pattern Computation: The CNN model is used to recognize character patterns, identify “W”, “K”, and “R” as the end-type characters, and calculate their context feature values.
[0207] Segmentation rule matching: The "-" is recognized as a separator and matches the preset rules such as character length, character type and frequency.
[0208] Dynamically split and generate intervals: Identify the segmentation points and perform dynamic segmentation, and the result is segmented into multiple character intervals: ["BRW", "20231001", "PROD123"].
[0209] Through the above detailed steps, the present invention performs character segmentation on the preprocessed initial behavior data to obtain multiple character intervals. These steps ensure the accuracy and reliability of character segmentation, which facilitates subsequent feature-based recommendation calculations. For example, on an e-commerce platform, user behavior data from different sources can be effectively parsed and processed to improve the accuracy and effectiveness of personalized recommendations. In a financial transaction system, this method can also be used to efficiently segment and process transaction record data to enhance risk control and customer service capabilities.
[0210] The above describes the user recommendation method based on big data in the embodiment of the present invention. The following describes the user recommendation system based on big data in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a user recommendation system based on big data includes: A classification module is used to obtain the user's initial behavior data and the corresponding user basic information, and classify the initial behavior data into types to obtain the classified behavior data; wherein the user's initial behavior data at least includes browsing history, purchase history, click-through rate and search history; A matching module, used to match the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data; A fusion module, used to perform multimodal fusion on each recommendation feature data based on a preset behavior type weight to obtain a fusion result, and generate a recommendation instruction corresponding to the user's initial behavior data based on the fusion result; The adjustment module is used to optimize the preset user recommendation content based on the recommendation instructions corresponding to the initial behavior data and the basic information of the user, generate a recommendation adjustment result, and adjust the user's recommendation list based on the recommendation adjustment result.
[0211] The present invention also provides a user recommendation device based on big data, which includes a memory and a processor, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the user recommendation method based on big data in the above-mentioned embodiments.
[0212] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the user recommendation method based on big data.
[0213] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0214] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.
[0215] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A user recommendation method based on big data, characterized in that: The user recommendation method based on big data comprises the following steps: Acquire the user's initial behavior data and corresponding user basic information, and classify the initial behavior data into types to obtain classified behavior data; wherein the user's initial behavior data at least includes browsing history, purchase history, click-through rate and search history; Matching the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data; Based on the preset behavior type weights, multimodal fusion is performed on each recommendation feature data to obtain a fusion result, and based on the fusion result, a recommendation instruction corresponding to the user's initial behavior data is generated; Based on the recommendation instructions corresponding to the initial behavior data and the basic information of the user, the preset user recommendation content is optimized to generate a recommendation adjustment result, and based on the recommendation adjustment result, the user's recommendation list is adjusted.
2. The user recommendation method based on big data according to claim 1, characterized in that: The obtaining of the user's initial behavior data and the corresponding user basic information, and classifying the initial behavior data by type to obtain the classified behavior data, includes: Collecting the user's initial behavior data and user basic information from multiple data sources; wherein the multiple data sources include at least website logs and mobile applications; the user basic information includes at least the user's age, gender, geographic location and interest preferences; Performing data cleaning and preprocessing on the collected initial behavior data to obtain preprocessed initial behavior data; Based on the preset data mining technology, the preprocessed initial behavior data is classified into types to obtain classified behavior data.
3. The user recommendation method based on big data according to claim 2, characterized in that: The pre-processed initial behavior data is classified based on the preset data mining technology to obtain the classified behavior data, including: Using a preset classification algorithm to classify the preprocessed initial behavior data according to predefined behavior types; Performing statistical analysis on the classified behavior data to extract corresponding behavior feature data; The extracted behavior feature data is subjected to feature vectorization processing to obtain classified behavior data.
4. The user recommendation method based on big data according to claim 1, characterized in that: The matching of recommendation features corresponding to the classified behavior data in a preset database to obtain various recommendation feature data includes: Inputting the classified behavior data into a preset database to obtain a first character set; Based on a preset character segmentation rule, the first character set is segmented to obtain a plurality of character intervals; wherein the character intervals are composed of a plurality of different English characters; Get the last character of each character interval as the identification character of each behavior type; For each identification character corresponding to the behavior type, the initial recommended feature corresponding to the identification character is retrieved from a database; wherein the database stores in advance a matching rule based on matching the identification character with the corresponding initial recommended feature; The similarity between the classified behavior data and the initial recommendation features is analyzed, and the recommendation features that best match the user's initial behavior data are selected to obtain each recommendation feature data.
5. The user recommendation method based on big data according to claim 4, characterized in that: The first character set is segmented based on a preset character segmentation rule to obtain a plurality of character intervals, including: According to the preset character segmentation rules, define the standard and method of character segmentation; wherein the preset character segmentation rules include character length, character type, character frequency and character position; Using a machine learning-based character recognition algorithm, the characters in the first character set are scanned and recognized to obtain character patterns, and a corresponding segmentation feature value is calculated for each character pattern; wherein the segmentation feature value includes at least a relationship between the character and adjacent characters, context information of the character, and frequency statistics of the character; Perform in-depth analysis on the recognized character patterns, match them with the preset character segmentation rules, and determine the segmentation points for character segmentation; Based on the segmentation feature value and each determined segmentation point of the character segmentation, the first character set is dynamically segmented to generate a plurality of character intervals.
6. A user recommendation system based on big data, characterized in that: The user recommendation system based on big data includes: A classification module is used to obtain the user's initial behavior data and the corresponding user basic information, and classify the initial behavior data into types to obtain the classified behavior data; wherein the user's initial behavior data at least includes browsing history, purchase history, click-through rate and search history; A matching module, used to match the recommendation features corresponding to the classified behavior data in a preset database to obtain each recommendation feature data; A fusion module, used to perform multimodal fusion on each recommendation feature data based on a preset behavior type weight to obtain a fusion result, and generate a recommendation instruction corresponding to the user's initial behavior data based on the fusion result; The adjustment module is used to optimize the preset user recommendation content based on the recommendation instructions corresponding to the initial behavior data and the basic information of the user, generate a recommendation adjustment result, and adjust the user's recommendation list based on the recommendation adjustment result.
7. A user recommendation device based on big data, characterized in that: The user recommendation device based on big data includes: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the big data-based user recommendation device executes the big data-based user recommendation method according to any one of claims 1 to 5.
8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the user recommendation method based on big data as described in any one of claims 1 to 5 is implemented.
Citation Information
Cited By
User behavior analysis method and device for burglary prevention of mobile equipment in exhibition hall
CN119580406A
