An AI-driven accurate demand positioning method and system

Abstract

The present application provides an AI-driven precise demand positioning method and system, which relates to the field of big data analysis technology. The method activates data collection permission channels; under sampling boundary constraints, uses real-time user demand as the retrieval boundary to perform multi-channel data backtracking to obtain user panoramic data; performs demand trend analysis based on channel relevance, and outputs short-term and long-term demand panoramic insight results; recommends resource positioning and matching based on short-term panoramic insight results and real-time user demand; recommends resource positioning and expansion based on long-term panoramic insight results and real-time user demand; and performs hierarchical visualization lazy loading on the obtained recommended resource set and multi-level expansion resource set. The present application solves the technical problems of the prior art that the user demand cannot be accurately grasped and the recommended resource matching is low due to the single data source and limited demand analysis dimension, and improves the accuracy of demand positioning and the personalized matching and coverage of recommended resources.

Description

Technical Field

[0001] The present application relates to the field of big data analysis technology, and specifically to an AI-driven precise demand positioning method and system. Background Art

[0002] With the widespread application of intelligent recommendation systems and personalized services, accurate user demand positioning has become the key to improving user stickiness and conversion rates. Traditional user demand positioning mainly relies on user historical behavior data, keyword search and rule-based recommendation strategies. Some systems analyze users' past operation records through simple label classification or collaborative filtering methods to predict content that may be of interest. However, these methods are usually limited to a single data source and can only analyze based on users' explicit behaviors (such as browsing, clicking, purchasing, etc.), ignoring the deeper demand background and cross-channel user behavior, and it is difficult to adapt to the dynamic changes in user needs, which leads to a low matching degree of recommended content and cannot meet users' expectations for personalized services. Summary of the invention

[0003] The present application provides an AI-driven precise demand positioning method and system, which solves the technical problem that the prior art cannot accurately grasp user needs due to the single user data source and limited demand analysis dimensions, thereby affecting the matching degree of recommended resources, and achieves the technical effect of improving the accuracy of user demand positioning and enhancing the personalized matching degree and coverage of recommended resources.

[0004] In view of the above problems, on the one hand, the present application provides an AI-driven precise demand positioning method, which includes: after the target user inputs the real-time user demand in the recommendation interface, the data collection permission channel is activated through the permission response; under the sampling boundary constraint of the data collection permission channel, the real-time user demand is used as the retrieval boundary constraint, and the crawler technology is used to perform multi-channel data backtracking to obtain user panoramic data; based on the channel correlation, the user panoramic data is subjected to demand trend analysis, and the demand panoramic insight results are output, wherein the demand panoramic insight results include short-term panoramic insight results and long-term panoramic insight results; the short-term panoramic insight results and the real-time user demand are used as the first user portrait to perform recommended resource positioning matching, and a recommended resource set is output; the long-term panoramic insight results and the real-time user demand are used as the second user portrait to perform recommended resource positioning expansion to obtain a multi-level expanded resource set; the recommended resource set and the multi-level expanded resource set are hierarchically visualized lazy loading in the recommendation interface.

[0005] On the other hand, the present application also provides an AI-driven precise demand positioning system, the system comprising: a response activation unit, which is used to activate the data collection permission channel through permission response after the target user inputs the real-time user demand in the recommendation interface; a panoramic data collection unit, which is used to use the real-time user demand as the retrieval boundary constraint under the sampling boundary constraint of the data collection permission channel, and adopt crawler technology to perform multi-channel data backtracking to obtain user panoramic data; a demand trend analysis unit, which is used to perform demand trend analysis on the user panoramic data based on channel relevance, and output demand panoramic insight results, wherein the demand panoramic insight results include short-term panoramic insight results and long-term panoramic insight results; a recommended resource matching unit, which is used to match the short-term panoramic insight results and the real-time user demand as the first user portrait for recommended resource positioning and output a recommended resource set; a recommended resource expansion unit, which is used to use the long-term panoramic insight results and the real-time user demand as the second user portrait for recommended resource positioning expansion to obtain a multi-level expanded resource set; a visualization unit, which is used to perform hierarchical visualization lazy loading of the recommended resource set and the multi-level expanded resource set in the recommendation interface.

[0006] One or more technical solutions provided in this application have at least the following beneficial effects:

[0007] After the target user inputs the real-time user demand in the recommendation interface, the data collection permission channel is activated through the permission response. Based on the user's active demand input, the data collection permission can be dynamically opened to ensure the flexibility and compliance of data acquisition, and provide a clear boundary and basis for subsequent data collection and analysis. Under the permission constraint, with user demand as the retrieval boundary, crawler technology is used to trace back multi-channel data, breaking the limitations of the traditional single data source. Through cross-channel data tracing, more complete user panoramic data is obtained to enhance the depth and breadth of demand analysis. Based on channel relevance, the demand trend of user panoramic data is analyzed, and short-term and long-term panoramic insight results are output. This step forms a two-layer demand insight through intelligent analysis of short-term demand trends and long-term interest evolution, providing a key basis for accurately positioning user needs, so as to achieve accurate matching and demand expansion of recommended resources. The short-term panoramic insight results are combined with real-time needs to form the first user portrait, and the recommended resource positioning matching is performed to ensure that the recommended content is highly relevant to the user's immediate needs and improve the accuracy and immediacy of the recommendation. The long-term panoramic insight results are combined with real-time needs to form the second user portrait, and the recommended resources are expanded to improve the coverage of recommended resources. Finally, the display of recommended content is optimized through hierarchical lazy loading to ensure that users can quickly obtain resources that are highly matched to their needs while reducing the waste of computing resources.

[0008] To sum up, this application not only improves the real-time perception capability of user needs, but also realizes the accuracy and scalability of personalized recommendations through multi-level matching strategies, while optimizing the interactive experience and computing efficiency of the recommendation process, thereby significantly improving the accuracy of user demand positioning, improving the personalized matching and coverage of recommended resources, and enhancing user experience and satisfaction, providing technical support for corporate business decisions and resource planning.

[0009] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 A flowchart of an AI-driven precise demand location method provided in an embodiment of the present application.

[0011] Figure 2 A flowchart of performing demand trend analysis on user panoramic data and outputting demand panoramic insight results in an AI-driven precise demand location method provided in an embodiment of the present application.

[0012] Figure 3 A structural diagram of an AI-driven precise demand positioning system provided in an embodiment of the present application.

[0013] Explanation of the reference numerals: response activation unit 10 , panoramic data collection unit 20 , demand trend analysis unit 30 , recommended resource matching unit 40 , recommended resource expansion unit 50 , visualization unit 60 . DETAILED DESCRIPTION

[0014] The embodiments of the present application provide an AI-driven precise demand positioning method and system to solve the technical problem in the prior art that, due to the single user data source and limited demand analysis dimensions, user needs cannot be accurately grasped, thereby affecting the matching degree of recommended resources. The technical effect of improving the accuracy of user demand positioning and enhancing the personalized matching degree and coverage of recommended resources is achieved.

[0015] Embodiment 1, as Figure 1 As shown, the embodiment of the present application provides an AI-driven accurate demand positioning method, the method comprising:

[0016] Step S1: After the target user inputs the real-time user demand in the recommendation interface, the data collection permission channel is activated through the permission response.

[0017] Specifically, the recommendation interface is an interactive interface that displays recommended content (such as products, services, information, etc.) to users. It is used to receive user needs and display recommendation results, such as the product recommendation page of an e-commerce platform, the song recommendation page of a music platform, etc. When the target user enters real-time user needs in the recommendation interface, the permission response mechanism is triggered to request the necessary data collection permissions from the target user. After the user authorizes, the corresponding data collection permission channel is activated to prepare for subsequent data collection. For example, on a news information platform, when the user enters "technology news", the system will first check the user's privacy settings and other related permissions. If the user allows the collection of relevant data, data collection permission channels such as user browsing history data collection channels and specific content sources related to technology news (data interfaces of certain technology blogs permitted by the cooperation agreement, etc.) are activated.

[0018] Activating data collection permission channels through permission response ensures the targetedness and legality of data collection, and provides basic conditions for subsequent accurate data collection and analysis.

[0019] Step S2: Under the sampling boundary constraints of the data collection authority channel, taking the real-time user demand as the retrieval boundary constraint, crawler technology is used to perform multi-channel data backtracking to obtain user panoramic data.

[0020] Specifically, sampling boundary constraints are the scope and restrictions of data collection, ensuring that the collected data is in compliance with the requirements and valid, such as only accessing shopping records in the past six months, rather than all historical data. Retrieval boundary constraints are the scope and direction restrictions of data retrieval, which are determined based on real-time user needs. They ensure that only data related to real-time user needs is collected when crawling data, avoiding interference from irrelevant information and improving the efficiency and accuracy of data retrieval. User panoramic data includes user behavior data (such as browsing, clicking, purchasing and other behavior records), preference data (such as favorite colors, styles, etc.), social data (such as friend relationships, social platform interactions, etc.) and historical demand records (such as content searched in the past, products purchased, etc.).

[0021] Under the activated data collection permission channels, crawler technology is used to perform multi-channel data backtracking based on sampling boundary constraints and retrieval boundary constraints. For example, on e-commerce platforms, the sampling boundary constraint is to collect user data within the past three months, and the retrieval boundary constraint is the "sports equipment" demand input by the user. Crawler technology will capture relevant user behavior data (such as which sports brands have been browsed), preference data (such as preferred price range), social data (such as whether to follow sports experts) and historical demand records (such as sports equipment purchased before) from the user's browsing history, purchase history, favorites and other channels according to the set constraints, so as to obtain panoramic user data.

[0022] Through multi-channel data crawling based on sampling boundary constraints and retrieval boundary constraints, we can obtain comprehensive user data that is highly relevant to user needs, providing a rich source of data for accurate analysis of user needs and avoiding the collection of useless or irrelevant data.

[0023] Step S3: Based on channel relevance, the user panoramic data is subjected to demand trend analysis, and demand panoramic insight results are output, wherein the demand panoramic insight results include short-term panoramic insight results and long-term panoramic insight results.

[0024] Specifically, channel relevance refers to the correlation and mutual influence between different data channels, such as the likes on social media may affect the purchase behavior on shopping websites. The demand panorama insight results are the results obtained after a comprehensive analysis of user needs, including short-term panorama insight results (recent demand trends and characteristics) and long-term panorama insight results (demand evolution and potential demand over a longer period of time).

[0025] Based on channel relevance, we analyze demand trends on user panoramic data. For example, by analyzing the popularity of users’ discussions on a certain type of product on social media (such as the recent frequent mention of thin and light laptops) and their purchasing behavior on e-commerce platforms (such as viewing laptop pages multiple times in the past month), we can combine machine learning models to predict that users may have a need to buy laptops in the short term, and analyze their long-term demand trends (such as users may have a sustained interest in technology products and may have a need for related accessories in the future). The final output includes demand panoramic insight results for both short-term and long-term panoramic insight results.

[0026] By deeply exploring the changing trends of user needs from the perspective of the relationship between different channels, we can obtain short-term and long-term panoramic insights, which will help us to more comprehensively understand the characteristics of user needs at different time scales and provide a more accurate basis for the subsequent recommendation of resource positioning.

[0027] Step S4: Use the short-term panoramic insight results and real-time user needs as the first user profile to perform recommended resource positioning and matching, and output a recommended resource set.

[0028] Specifically, the short-term panoramic insight results obtained in step S3 are combined with real-time user needs to construct a first user portrait. For example, on a travel booking platform, the short-term panoramic insight results show that the user has a certain interest in seaside vacations in the near future, and the real-time user demand is "travel within three days", then the first user portrait constructed is a user image who has a recent interest in seaside vacations and whose demand is to travel within three days. Then, a recommendation algorithm (such as a content-based recommendation algorithm or a collaborative filtering algorithm) is used to locate and match recommended resources in the resource library, screen out resources that are highly matched with user needs, and generate and output a recommended resource set. Taking the content-based recommendation algorithm as an example, the algorithm will look for tourism products related to seaside vacations and suitable for three-day travel (such as seaside hotels, local tourist attractions, etc.), and these matching tourism products will be combined into a recommended resource set.

[0029] Combine the short-term needs obtained from the previous analysis with the real-time needs, accurately locate the recommended resources that match the user's current needs, and improve the matching degree and immediacy of the recommended resources.

[0030] Step S5: Using the long-term panoramic insight results and real-time user needs as the second user profile to perform recommended resource location expansion to obtain a multi-level expanded resource set.

[0031] Specifically, a second user portrait is constructed based on the long-term panoramic insight results and real-time user needs obtained in step S3. For example, on a music platform, the long-term panoramic insight results show that users have a long-term preference for classical music, and the real-time user demand is "relaxing music", then the second user portrait is a user image who has long liked classical music and whose current demand is relaxing music. Then, the recommended resources are located and expanded in the resource library through a recommendation algorithm (such as a recommendation algorithm based on a knowledge graph), and on the basis of matching the recommended resources, the recommendation scope is further expanded to introduce resources related to the user's potential needs. For example, based on the knowledge graph, in addition to recommending classic relaxation tracks in classical music, some light music and meditation music with similar styles to classical music will also be recommended, and these expanded music resources will be combined into a multi-level expanded resource set.

[0032] By building a second user portrait based on long-term panoramic insights and real-time needs, and expanding the recommended resource positioning, it is possible to consider the long-term needs of users on the basis of meeting their short-term needs, expand the diversity of recommended resources, and thus provide richer and more comprehensive recommended resources to meet the potential needs of users.

[0033] Step S6: The recommended resource set and the multi-level extended resource set are lazily loaded in a hierarchical visual manner on the recommendation interface.

[0034] Specifically, the recommended resource set and the multi-level extended resource set are lazily loaded in the recommendation interface. This process mainly relies on front-end development technologies, such as JavaScript and related lazy loading plug-ins. For example, on the recommendation interface of a news information website, the core resources that highly match the user's current needs are loaded first (such as the technology news from the recommended resource set that best meets the user's needs). When the user scrolls to the bottom of the news list, the JavaScript code detects this scrolling event, triggers the lazy loading mechanism, obtains the next batch of news resources (from the recommended resource set and the multi-level extended resource set) from the server, and displays them to the user in a visual way (such as smooth scrolling to load new news items). Hierarchical visual lazy loading automatically loads more content when the user scrolls to the bottom of the page, achieving a seemingly infinite browsing experience, and loading resources only when needed (when the user scrolls), rather than loading all resources when the page is initially loaded. This not only improves the interface loading efficiency, but also displays recommended resources to users in a hierarchical and progressive manner, so that recommended resources can be better accepted and utilized by users, improving user experience and satisfaction.

[0035] Further, step S6 includes:

[0036] Step S61: analyzing the compatibility of the recommended resource set with the real-time user demand, and constructing a resource association map between the recommended resource set and the real-time user demand according to the analysis result.

[0037] Step S62: Taking the resource association map as a starting point, performing hierarchical resource association analysis on the multi-level extended resource set to obtain a multi-level extended resource map.

[0038] Step S63: Visually loading the recommended resource set on the recommendation interface based on the resource association graph.

[0039] Step S64: According to the interactive behavior of the target user, lazy loading of the multi-level extended resource set is performed on the recommendation interface according to the multi-level extended resource map.

[0040] Specifically, demand adaptability refers to the degree of match between the recommended resources and the user's current needs. For example, if a user searches for the latest smartphone of a certain brand, but the recommended content is information about the old smartphone of the brand, the adaptability is low. First, each resource in the recommended resource set is analyzed in detail to determine the degree of match between each attribute (such as price, function, style, etc.) and the real-time user needs. For example, on a video platform, if the user's demand is to watch comedy videos with a short duration, for each video in the recommended resource set, analyze whether it is a comedy type and whether it is short in duration. Then, based on the analysis results, a resource association graph of the recommended resource set and the real-time user needs is constructed using a knowledge graph construction tool. The resource association graph is a graph that graphically displays the association relationship between the recommended resource set and the real-time user needs. It takes the user demand as the central node, and the resources in the recommended resource set as the peripheral nodes. The lines between the nodes represent the association relationship (such as the function of the resource meets a certain aspect of the user demand). For example, if a video is a comedy and is 10 minutes long, a connection is established between the node representing the video and the nodes representing "comedy" and "short duration" to form a resource association graph. By constructing a resource association graph, the relationship between the recommended resource set and real-time user needs is clearly displayed, which provides a basis for subsequent hierarchical resource association analysis and visual loading, and helps to improve the accuracy of recommended resource display.

[0041] Starting from the constructed resource association graph, the multi-level extended resource set is analyzed hierarchically to explore deeper resource relationships and obtain a multi-level extended resource graph, which shows the internal resources of the multi-level extended resource set and the association relationship with the recommended resource set and real-time user needs. When performing hierarchical association analysis on the multi-level extended resource set, semantic matching (such as BERT, Word2Vec) can be used to analyze the semantic similarity between recommended resources; or user behavior analysis (such as collaborative filtering, user click stream analysis) can be used to establish the association relationship between resources. If many users visit resource B after searching for resource A, the association between the two is automatically established; the knowledge graph construction method can also be used to automatically mine the hierarchical relationship of content based on the existing knowledge system. For example, after learning "machine learning", "deep learning" may be needed as advanced content. Exemplarily, on a news information platform, the recommended resource set is news that initially matches user needs (such as paying attention to technology news), and the multi-level extended resource set may include related technology reviews, technology company dynamics, etc. Analyze the relationship between these extended resources and the recommended resource set as well as user needs. For example, which technology reviews are targeted at technology events in the recommended news, and which technology company dynamics are related to the technology in the news. Based on the analysis results, build a multi-level extended resource map and display the resources in the multi-level extended resource set hierarchically in the map according to the association relationship.

[0042] According to the constructed resource association graph, the loading order and display method of the recommended resource set are determined. For example, in the travel recommendation interface, if the user's demand is to find attractions suitable for family travel, according to the resource association graph, the attractions that are highly suitable for family travel (such as parent-child parks) are loaded first, and displayed on the recommendation interface in the form of pictures and brief text introductions. This process uses front-end development technologies such as HTML, CSS, and JavaScript to present resources in a visual way.

[0043] According to the interactive behavior of the target user, such as the user's scrolling behavior on the recommendation interface or the behavior of clicking on a specific type of resource (such as clicking on a short video), analyze the user's content resource browsing tendency, such as whether they tend to watch short videos, short information, or long videos and long texts, whether they tend to read news or commentary, and whether they tend to prefer highly professional or highly interesting content. Determine the resources to be loaded based on the user's browsing tendency and the multi-level expansion resource map, and load the resources to the recommendation interface in a lazy loading manner. For example, if the user tends to watch short videos, then on the recommendation interface, based on the multi-level expansion resource map, the short video resources in the multi-level expansion resource set are preferentially loaded.

[0044] Through the above steps, the recommended resource set and the multi-level expanded resource set are presented to the user in a structured manner. The relationship between resources is analyzed using the resource association graph, and intelligent lazy loading is performed in combination with the user's interactive behavior to ensure the logic, hierarchy and dynamic adaptability of the recommendation results.

[0045] Furthermore, step S1 of the embodiment of the present application includes:

[0046] Step S11: Apply NLP technology to extract keywords from the real-time user needs to obtain real-time demand features.

[0047] Step S12: input the real-time requirement characteristics into the authority management module, and match and output the real-time authority requirements.

[0048] Step S13: Perform a coverage judgment on the real-time permission requirement according to the historical authorization information of the target user to obtain a real-time permission request.

[0049] Step S14: updating the real-time permission requirement according to the authorization response feedback of the target user to the real-time permission request, and using the updated result to map and activate the data collection permission channel.

[0050] Specifically, when the target user enters the real-time user demand in the recommendation interface, the keyword extraction algorithm in NLP technology (such as TF-IDF algorithm) is applied to process the real-time user demand, and the real-time user demand in the form of natural language entered by the user is converted into more representative and operational real-time demand features, which provides clear input content for the subsequent processing of the permission management module. For example, if a user enters "I want to go to the beach for a vacation in the summer, and I hope the hotel is cheap and close to the beach" on the travel booking platform, the algorithm will calculate the importance of each word and extract keywords such as "summer", "seaside vacation", "cheap hotel", "close to the beach", etc. These keywords constitute the real-time demand features. This process relies on natural language processing libraries such as NLTK (Natural Language Processing Toolkit in Python) or Stanford NLP and other tools to achieve this.

[0051] The permission management module is responsible for managing various permission-related operations, including permission matching, verification and other functions. The module contains a permission rule base that stores the correspondence between different requirements and permissions. The real-time permission requirements are determined by querying this rule base. The real-time requirement characteristics are input into the permission management module, and the module searches the permission rule base for permission requirements corresponding to the current user requirements and outputs them as real-time permission requirements. For example, in a social application, if the real-time requirement characteristics include "find nearby friends", the permission management module will match the real-time permission requirements such as the need to obtain the user's geographic location permission.

[0052] Based on the target user's past authorization records for various permissions (i.e. historical authorization information), the obtained real-time permission requirements are overwritten to determine whether the permissions in the real-time permission requirements have been authorized. If not, a new request or supplementary request is required. For example, in a news and information application, the real-time permission requirements include obtaining the user's reading history and push notification permissions, and the user's historical authorization information shows that the permission to obtain the reading history has been authorized, so the push notification permission needs to be re-requested. Through overwrite judgment, repeated requests for permissions that the user has already authorized are avoided, and the permissions that need to be re-requested or supplemented are accurately determined, which improves the user experience and reduces unnecessary permission requests that disturb users.

[0053] Authorization response feedback is the target user's response of consent or disagreement to the real-time permission request. For example, when the mobile application pops up the permission request dialog box, the user clicks "Allow" or "Reject" is the authorization response feedback. When the target user makes an authorization response feedback to the real-time permission request, the real-time permission requirement is updated according to the feedback result. For example, if the user agrees to obtain push notification permission, this permission is added to the authorized real-time permission requirement. Obtain the mapping relationship table of pre-set permission requirements and data collection permission channels (this table is stored inside the application system), and find the corresponding data collection permission channel that needs to be activated according to the updated real-time permission requirements. For example, in the mapping relationship table, the permission to obtain user browsing history corresponds to the browser history data collection channel, and the updated real-time permission requirements include the permission to obtain user browsing history, so the browser history data collection channel is activated.

[0054] Through the above steps, user needs can be accurately identified, and data collection permissions can be dynamically adjusted to ensure that data collection not only meets user privacy protection requirements but also provides high-quality data support for accurate recommendations.

[0055] Furthermore, step S2 of the embodiment of the present application includes:

[0056] Step S21: Preset the data backtracking time scale.

[0057] Step S22: extracting a first authorization channel from the data collection authority channel, wherein the data collection authority channel includes K authorization channels.

[0058] Step S23: Using the data backtracking time scale as the time boundary constraint and the real-time demand characteristics as the content boundary constraint, crawler technology is used to call user historical data in the first authorized channel to obtain first channel historical data.

[0059] Step S24: Similarly, the user history data of the K authorized channels are called to obtain K-dimensional channel history data, wherein the K-dimensional channel history data constitutes the user panoramic data.

[0060] Specifically, the data backtracking time scale is a pre-set standard for limiting the length of data backtracking. The data backtracking time scale is preset based on factors such as business needs and the timeliness of the data. For example, in the movie recommendation example, if the user enters "I want to watch a sci-fi adventure movie that was recently released", the data backtracking time scale can be preset to the past month to obtain the user's recent viewing behavior and preferences. Setting the data backtracking time scale can limit the scope of data collection, avoid collecting too much and too old data, and ensure that the collected data has a high relevance to current needs in terms of time.

[0061] The first authorized channel is extracted from the K authorized channels included in the data collection permission channel in a certain order (such as according to the priority of the channel or a random order), and is recorded as the first authorized channel. K is a positive integer, indicating the number of authorized channels.

[0062] Time boundary constraints are used to limit the time range of data calls, and content boundary constraints are used to limit the content range of data calls. Using the preset data backtracking time scale as the time boundary constraint and the real-time demand characteristics as the content boundary constraint, crawler technology is used to call user historical data in the extracted first authorized channel to obtain the first channel historical data. For example, on a news information platform, if the data backtracking time scale is the past month, the real-time demand characteristics are "technology news", and the first authorized channel is the user's browsing history, then the crawler technology will search for records related to technology news in the user's browsing history in the past month. These records are the first channel historical data.

[0063] In the same way as obtaining the historical data of the first channel, the user historical data is called from K authorized channels in turn. For example, the data collection permission channels include three authorized channels (K=3), such as browser history records, social media accounts, and e-commerce platform purchase history. After completing the data call of the browser history records, continue to call the data in the social media accounts and e-commerce platform purchase history, and finally obtain the channel historical data of three dimensions, which together constitute the user panoramic data.

[0064] Through the above steps, based on user authorization permissions, user historical data can be traced back from multiple data sources to build a complete user panorama data, so as to accurately and compliantly obtain the user's past behavior data, and provide comprehensive data support for subsequent in-depth analysis and precise recommendations.

[0065] Further, such as Figure 2 As shown, step S3 of the embodiment of the present application includes:

[0066] Step S31: taking 1 / 24 of the data backtracking time scale as the data segmentation time scale.

[0067] Step S32: Segment the K-dimensional channel historical data into K-stage historical data sequences according to the data segmentation time scale.

[0068] Step S33: reorganize the K stage historical data sequences based on time alignment to obtain multiple groups of stage historical data.

[0069] Step S34: After using TF-IDF to extract text features from the multiple groups of stage historical data, PCA is used to reduce the dimension of the extracted results to perform dimensionality reduction identification of behavioral features, thereby obtaining panoramic insight results for multiple stages.

[0070] Step S35: By performing long-term and short-term demand evolution analysis on the panoramic insight results of the multiple stages, the short-term panoramic insight results and the long-term panoramic insight results are output.

[0071] Specifically, in order to analyze user behavior more finely, the data backtracking time scale is divided by 24 to obtain the data segmentation time scale, which serves as the time standard for segmenting user historical data.

[0072] According to the determined data segmentation time scale, the K-dimensional channel historical data is segmented. For example, when processing user browsing history data, if the data is stored in the database with timestamp as the index, the query statement is used to extract data from the database according to the time range of the data segmentation time scale, thereby obtaining K-stage historical data sequences. For example, if the K-dimensional channel historical data covers one year of user data, according to the half-month segmentation time scale, the data of every half month is a stage historical data sequence.

[0073] For K phase historical data sequences, the historical data of different channels in each phase are reassembled in chronological order to obtain multiple groups of phase historical data, each of which contains historical data of different channels in the same time period. For example, there are three channels (channel A, channel B, channel C), each of which has a phase historical data sequence divided by time. The data of channel A, channel B, and channel C in each phase (such as the first half month) are combined into a group of phase historical data. This process can be implemented using the merge function in the data processing library, such as using the merge function of the Pandas library in Python (such as merge or concat function). By aligning and reorganizing the time series, the data of the same time period scattered in different channels are integrated together, which can more comprehensively reflect the comprehensive situation of users in each time period and provide a more complete data foundation for subsequent text feature extraction.

[0074] Use TF-IDF to extract text features from multiple sets of stage historical data. For example, when processing a user's search history data (as part of multiple sets of stage historical data), the TF-IDF algorithm calculates the importance of each word in each search record and outputs the importance weight corresponding to the word. Then, the TF-IDF extraction results are input into PCA for dimensionality reduction to identify the user behavior characteristics in each stage, thereby obtaining panoramic insights from multiple stages. In Python, the scikit-learn library can be used to implement TF-IDF (TfidfVectorizer class) and PCA (PCA class) operations.

[0075] Perform a long-term and short-term demand evolution analysis on the panoramic insight results of multiple stages to identify the short-term fluctuations and long-term trends of user demand, so as to understand the changing rules of user demand at different time scales, obtain short-term panoramic insight results and long-term panoramic insight results, and provide a more accurate basis for subsequent recommended resource positioning and other operations. Each insight result is a multidimensional vector, and each dimension represents the intensity of user demand or preference in a certain aspect. For example, for a content platform, the dimension can be the frequency of users reading or watching different types of articles and videos; for an e-commerce platform, the dimension can be the frequency of users buying different product categories, browsing time, etc. When conducting a long-term and short-term demand evolution analysis, time series analysis methods or regression analysis methods in machine learning can be used. By observing the changing trends of user demand characteristics in different stages (stages corresponding to short-term and long-term), short-term panoramic insight results and long-term panoramic insight results can be distinguished. For example, on an e-commerce platform, if users frequently search for a certain type of low-priced goods in the recent stage (short-term), and the user search behavior gradually shifts to high-quality goods over a longer period of time, these analysis results constitute short-term panoramic insight results and long-term panoramic insight results respectively.

[0076] Through the above steps, we can conduct time series analysis on the user's panoramic data, extract user behavior characteristics, and output short-term and long-term demand insights based on the long-term and short-term demand change trends to accurately understand the evolution trend of user demand.

[0077] Further, step S35 includes:

[0078] Step S351: taking 1 / 12 of the data backtracking time scale as the short-term demand evolution scale.

[0079] Step S352: Taking the short-term demand evolution scale as a constraint, extract H stage panoramic insight results from the multiple stage panoramic insight results in reverse time.

[0080] Step S353: Performing cross-stage demand evolution analysis on the H-stage demand feature vectors of the H-stage panoramic insight results, and outputting a short-term resource preference vector as the short-term panoramic insight result.

[0081] Step S354: 11 / 12 of the data backtracking time scale is used as the long-term demand evolution scale.

[0082] Step S355: Similarly, taking the long-term demand evolution scale as a constraint, performing cross-stage demand evolution analysis on the panoramic insight results of the multiple stages, and outputting the long-term panoramic insight results.

[0083] Specifically, the short-cycle demand evolution scale is the time standard used to measure short-term demand evolution analysis. The short-cycle demand evolution scale is obtained by dividing the data backtracking time scale by 12.

[0084] With the short-term demand evolution scale as a constraint, H stage panoramic insight results are extracted from multiple stage panoramic insight results in reverse chronological order. Where H is a positive integer, indicating the number of extracted stage panoramic insight results. For example, if multiple stage panoramic insight results are stored in a list in chronological order, start from the end of the list by indexing, and extract H stage panoramic insight results step by step forward with the short-term demand evolution scale as the time interval.

[0085] For the H-stage panoramic insight results, extract their respective stage demand feature vectors, and use vector difference calculation, correlation coefficient analysis and other methods to perform demand cross-stage evolution analysis. For example, if the stage panoramic insight result is a vector about the user's purchase frequency of different types of goods, these vectors are analyzed for demand cross-stage evolution. In Python, use the numpy library to perform vector operations, calculate the difference or correlation coefficient between adjacent stage demand feature vectors, and construct a short-term resource preference vector by analyzing these results to represent the user's preference for resources (such as goods, content, etc.) in the short term. This vector is used as a short-term panoramic insight result to provide a direct basis for short-term recommended resource positioning.

[0086] The long-term demand evolution scale is the time standard used to measure long-term demand evolution analysis. The long-term demand evolution scale is obtained by multiplying the data backtracking time scale by 11 / 12.

[0087] Following a similar method to obtaining short-term panoramic insight results, with the long-term demand evolution scale as a constraint, we conduct a cross-stage demand evolution analysis on the panoramic insight results of multiple stages, extract the demand feature vectors of each stage, and then analyze the changing relationship of these vectors over a long period of time. We output the long-term resource preference vector as the long-term panoramic insight result to reflect the user's long-term preference for resources (such as goods, content, etc.), providing a direct basis for long-term recommendation resource positioning.

[0088] Through the above steps, based on the user's historical data, their short-term and long-term demand evolution trends are analyzed, and finally short-term and long-term panoramic insight results are output to accurately capture the dynamic changes in user needs and predict possible future demand directions.

[0089] Furthermore, step S4 of the embodiment of the present application includes:

[0090] Step S41: The short-term panoramic insight results and real-time user needs are fused through vector stitching to obtain the first user portrait.

[0091] Step S42: vectorizing resource features of the local resource library to obtain multiple resource feature vectors of multiple local resources.

[0092] Step S43: using the Pearson correlation coefficient to traverse and calculate the multiple resource matching degrees between the multiple resource feature vectors and the first user portrait.

[0093] Step S44: preset a resource matching threshold, and use the resource matching threshold to traverse the multiple resource matching degrees to screen and obtain the recommended resource set.

[0094] Specifically, both the short-term panoramic insight results and the real-time user needs are expressed in vector form. Then, the vector splicing operation is performed in a certain order (such as first the short-term panoramic insight result vector elements, then the real-time user needs vector elements) to obtain the first user portrait. This can be achieved using array or vector operation functions. For example, vector A=(a 1 , a 2 , a 3 ) is the short-term panoramic insight result, vector B=(b 1 , b 2 , b 3 ) is the real-time user demand, and the two are concatenated to obtain a new vector C=(a 1 , a 2 , a 3 , b 1 , b 2 , b 3 ), represents the first user portrait.

[0095] Extract features from each resource in the local resource library and convert these features into vector representations to obtain multiple resource feature vectors, so that resources can participate in subsequent resource matching calculations in a unified form, which is convenient for comparing and screening resources. For example, for laptops on e-commerce platforms, features include brand, price, weight, screen size, etc. These features can be converted into vectors through encoding and normalization. For example, the feature vector of a laptop may be (0.8, 0.6, 0.4, 0.7), representing brand, price, weight, and screen size, respectively.

[0096] For the multiple resource feature vectors of the multiple local resources obtained, the Pearson correlation coefficient is used in turn to calculate the resource matching degree between these vectors and the first user portrait to quantify the matching degree between the local resources and the user portrait. The Pearson correlation coefficient calculation can be implemented using a statistical computing library. For example, in Python, the pearsonr function in the scipy.stats library is used to calculate the Pearson correlation coefficient between each resource feature vector and the first user portrait vector to obtain the corresponding resource matching degree.

[0097] Preset a resource matching threshold, then traverse all calculated resource matching degrees, filter out resources with resource matching degrees greater than the resource matching threshold, and form a recommended resource set. The recommended resource set contains detailed information of the filtered resources, such as resource ID, name, link, similarity score, etc., which is used for subsequent topology construction. For example, if the resource matching threshold is set to 0.7, the matching degree of resource A is 0.8, the matching degree of resource B is 0.6, and the matching degree of resource C is 0.75, then resources A and C are filtered into the recommended resource set.

[0098] Through the above steps, a first user profile is constructed based on the user's short-term needs and real-time needs, and the profile is used to filter out recommended resources that best suit the user's interests from the local resource library to ensure the accuracy and personalization of the recommended content.

[0099] Further, step S12 includes:

[0100] Step S121: Using the access frequency threshold as a constraint, call the online statistical data to obtain multiple sample access channel topologies for multiple sample demand keywords.

[0101] Step S122: storing the plurality of sample demand keywords and the plurality of sample access channel topologies in the authority management module based on the knowledge graph association, and completing the analysis data localization of the authority management module.

[0102] Step S123: Load the M real-time demand keywords constituting the real-time demand features into the authority management module, and obtain M sample access channel topologies through retrieval and matching.

[0103] Step S124: Aggregate the M sample access channel topologies, and extract sample access channels based on the aggregation results to obtain the real-time permission requirements.

[0104] Specifically, the access frequency threshold is a pre-set value used to limit the scope of network statistical data calls. Only the data related to the sample demand keywords that reach this access frequency will be called. The sample demand keywords are representative demand keywords extracted from historical data. The sample access channel topology is the structure and relationship of the access channels corresponding to the sample demand keywords, showing how to obtain relevant data through different channels. First, determine the value of the access frequency threshold, which can be set according to business needs and data volume. Then, through the network interface or database query statement, call the network statistical data according to the access frequency threshold. For example, if you are obtaining data from a network server, you can use HTTP requests to filter out the relevant data of the sample demand keywords that meet the conditions according to the set threshold, and then build multiple sample access channel topologies for multiple sample demand keywords. Through the constraints of the access frequency threshold, the efficiency and security of data calls are ensured, and representative sample data is obtained at the same time, providing a basis for subsequent permission management module analysis.

[0105] Use a knowledge graph construction tool (such as Neo4j) to store multiple sample demand keywords and multiple sample access channel topologies in an associated manner in the local storage corresponding to the permission management module. For example, in Neo4j, you can define nodes as sample demand keywords and sample access channels, and edges as the relationship between sample demand keywords and sample access channels. Then, store these nodes and edges in the graph database according to the topological structure to complete the analysis data localization of the permission management module. Through the knowledge graph associative storage and analysis data localization, the permission management module can quickly obtain and analyze data locally, improving the data access speed and processing efficiency. At the same time, the associative storage method of the knowledge graph helps to better explore the relationship between data and provide support for accurate analysis of the permission management module.

[0106] Real-time demand keywords are keywords that constitute the real-time demand characteristics and directly reflect the user's current real-time demand. The M real-time demand keywords that constitute the real-time demand characteristics are loaded into the permission management module that has completed data localization. Among them, M is a positive integer, which represents the number of real-time demand keywords. Then, a retrieval and matching operation is performed in the storage structure (knowledge graph structure) of the permission management module. For example, a graph query language (such as Cypher) is used to query the sample access channel topology related to the M real-time demand keywords, thereby obtaining M sample access channel topologies.

[0107] Aggregate the M sample access channel topologies. For example, if the sample access channel topology is a matrix representing the relationship and weight between channels, the aggregation operation is to add these matrices or perform other mathematical operations. Then, based on the aggregation results, extract all sample access channels contained therein as real-time permission requirements.

[0108] Through the above steps, based on the user's real-time demand characteristics and combined with historical sample data, real-time permission requirements suitable for the current user are dynamically generated to ensure that the data collection process only calls necessary data sources, thereby improving data utilization and reducing unnecessary privacy authorization.

[0109] Furthermore, step S11 includes:

[0110] Step S111: extracting the real-time user demand from the search box of the recommendation interface.

[0111] Step S112: Perform compliance verification on the real-time user demand.

[0112] Step S113: If the real-time user demand passes the compliance verification, after completing the content of the real-time user demand, NLP technology is applied to extract keywords of the completion result to obtain the real-time demand feature.

[0113] Step S114: If the real-time user demand fails the compliance verification, a compliance defect is generated and sent to the recommendation interface for visual warning.

[0114] Specifically, by monitoring the input events of the search box on the recommendation interface, when the user completes the input and submits (for example, pressing the Enter key or clicking the Search button), the text content in the search box is obtained, and this text content is the real-time user demand. In the actual implementation process, scripting languages ​​such as JavaScript can be used to implement monitoring of search box input events and content acquisition.

[0115] Establish a compliance rule base that contains various rules and requirements specified by the platform. For example, for an e-commerce platform, the rules include not including banned words, not having illegal product descriptions, etc. Compare the real-time user needs with the rules in the rule base to complete the compliance verification of the real-time user needs. You can use regular expressions, string matching and other technologies for verification. For example, if the rule base stipulates that certain specific banned words cannot be included, use regular expressions to check whether the real-time user needs contain these words.

[0116] When the real-time user demand passes the compliance verification, the real-time user demand is supplemented and improved using predefined templates or based on historical user demand data to make it more complete and accurate. For example, the real-time user demand is "summer" and "thin and light notebook", and the completion is "looking for a thin and light notebook suitable for summer use". Then, the keyword extraction algorithm in NLP technology, such as the TF-DF algorithm, is used to extract the keywords of the completion results, obtain the keywords and their weights, and select keywords with higher weights as real-time demand features, such as "summer", "thin", and "laptop".

[0117] When it is determined that the real-time user demand has not passed the compliance verification, analyze and find out the specific content that does not meet the rules as a compliance defect. For example, if it fails to pass the verification because it contains banned words, the banned words are regarded as compliance defects. Use front-end technology (such as JavaScript, HTML, CSS) on the recommendation interface to achieve visual warning. For example, a prompt box pops up to display the content of the compliance defect.

[0118] The above steps are used to parse the real-time requirements entered by users in the recommendation interface to ensure the compliance of user requirements, and use natural language processing (NLP) technology to extract key demand features for subsequent data collection and recommendation calculations.

[0119] In summary, the AI-driven precise demand location method provided in the embodiments of the present application has the following beneficial effects:

[0120] By inputting real-time needs by the target user, the data collection permission channel is activated, the user's current intention and demand direction are clarified, and the pertinence and legality of data collection are ensured, providing clear boundaries and basis for subsequent data collection and analysis. Within the scope of data collection authority, with real-time needs as constraints, crawler technology is used to trace back data from multiple channels, enriching the data source, providing a comprehensive data foundation for subsequent precise analysis, and ensuring a more three-dimensional and in-depth understanding of user needs. Based on channel relevance, the user's panoramic data is deeply analyzed, and short-term and long-term panoramic insight results are output, providing a key basis for accurately positioning user needs. Short-term insights and real-time needs are matched as the first user portrait to output a recommended resource set; long-term insights and real-time needs are expanded as the second user portrait to obtain a multi-level expanded resource set. This step converts the analysis results into actual recommended resources, which not only meets the user's current needs, but also expands potential needs, and improves the diversity and personalization of recommendations. Hierarchical visual lazy loading is performed on the recommendation interface to optimize the display of recommended resources, improve the interface loading speed and user experience, ensure that users can quickly obtain resources that are highly matched with their needs, and avoid resource waste.

[0121] Overall, the embodiment of the present application starts with the user inputting real-time demands, and then goes through precise data collection, multi-dimensional demand analysis, recommended resource positioning matching and expansion at different time scales, and finally presents recommended resources in an optimized interface display manner. This effectively improves the accuracy of user demand positioning, can fully grasp the user's short-term and long-term needs, provides rich and personalized recommended resources, improves the personalized matching and coverage of recommended resources, and enhances user experience and satisfaction by optimizing the display method of recommended resources, providing technical support for the company's business decision-making and resource planning.

[0122] Embodiment 2, as Figure 3 As shown, based on the same inventive concept as the aforementioned embodiment 1, the embodiment of the present application provides an AI-driven precise demand positioning system, the system comprising:

[0123] The response activation unit 10 is used to activate the data collection permission channel through permission response after the target user inputs the real-time user demand in the recommendation interface.

[0124] The panoramic data collection unit 20 is used to perform multi-channel data backtracking using crawler technology under the sampling boundary constraints of the data collection authority channel and taking the real-time user demand as the retrieval boundary constraints to obtain user panoramic data.

[0125] The demand trend analysis unit 30 is used to perform demand trend analysis on the user panoramic data based on channel relevance, and output demand panoramic insight results, wherein the demand panoramic insight results include short-term panoramic insight results and long-term panoramic insight results.

[0126] The recommended resource matching unit 40 is used to use the short-term panoramic insight result and the real-time user demand as the first user portrait to perform recommended resource positioning matching and output a recommended resource set.

[0127] The recommended resource expansion unit 50 is used to use the long-term panoramic insight results and real-time user needs as the second user portrait to locate and expand the recommended resources to obtain a multi-level expanded resource set.

[0128] The visualization unit 60 is used to perform hierarchical visualization lazy loading of the recommended resource set and the multi-level extended resource set on the recommendation interface.

[0129] Furthermore, the visualization unit 60 in the embodiment of the present application is also used to perform the following steps:

[0130] Analyze the adaptability of the recommended resource set to the real-time user needs, and construct a resource association map between the recommended resource set and the real-time user needs based on the analysis results; take the resource association map as the starting point, perform hierarchical resource association analysis on the multi-level extended resource set to obtain a multi-level extended resource map; perform visual loading of the recommended resource set on the recommendation interface based on the resource association map; and perform lazy loading of the multi-level extended resource set on the recommendation interface based on the multi-level extended resource map according to the interactive behavior of the target user.

[0131] Furthermore, the response activation unit 10 in the embodiment of the present application further includes:

[0132] The keyword extraction unit is used to apply NLP technology to extract keywords from the real-time user needs to obtain real-time demand features.

[0133] The real-time permission matching unit is used to input the real-time requirement characteristics into the permission management module, match and output the real-time permission requirements.

[0134] The permission coverage judgment unit is used to perform coverage judgment on the real-time permission demand according to the historical authorization information of the target user to obtain a real-time permission request.

[0135] The permission channel activation unit is used to update the real-time permission requirement according to the authorization response feedback of the target user to the real-time permission request, and use the update result to map and activate the data collection permission channel.

[0136] Furthermore, the panoramic data acquisition unit 20 in the embodiment of the present application is also used to perform the following steps:

[0137] Preset a data backtracking time scale; extract a first authorized channel from the data collection authority channel, wherein the data collection authority channel includes K authorized channels; use the data backtracking time scale as a time boundary constraint, and use the real-time demand characteristics as a content boundary constraint, use crawler technology to call user historical data in the first authorized channel, and obtain first channel historical data; and so on, call user historical data of the K authorized channels to obtain K-dimensional channel historical data, wherein the K-dimensional channel historical data constitutes the user panoramic data.

[0138] Furthermore, the demand trend analysis unit 30 of the embodiment of the present application is also used to perform the following steps:

[0139] 1 / 24 of the data backtracking time scale is used as the data segmentation time scale; the K-dimensional channel historical data is segmented into K stage historical data sequences according to the data segmentation time scale; the K stage historical data sequences are reorganized based on time series alignment to obtain multiple groups of stage historical data; after using TF-IDF to extract text features from the multiple groups of stage historical data, PCA is used to reduce the dimension of the extracted results to perform dimensionality reduction identification of behavioral features to obtain multiple stage panoramic insight results; by performing long-term and short-term demand evolution analysis on the multiple stage panoramic insight results, the short-term panoramic insight results and long-term panoramic insight results are output.

[0140] Furthermore, the demand trend analysis unit 30 of the embodiment of the present application is also used to perform the following steps:

[0141] 1 / 12 of the data backtracking time scale is used as the short-term demand evolution scale; with the short-term demand evolution scale as a constraint, H-stage panoramic insight results are extracted in reverse time from the multiple-stage panoramic insight results; by performing cross-stage demand evolution analysis on the H-stage demand feature vectors of the H-stage panoramic insight results, a short-term resource preference vector is output as the short-term panoramic insight result; 11 / 12 of the data backtracking time scale is used as the long-term demand evolution scale; and so on, with the long-term demand evolution scale as a constraint, a cross-stage demand evolution analysis is performed on the multiple-stage panoramic insight results, and the long-term panoramic insight result is output.

[0142] Furthermore, the resource matching unit 40 recommended in this embodiment of the present application is also used to perform the following steps:

[0143] The first user portrait is obtained by fusing the short-term panoramic insight results and real-time user needs through vector splicing; resource feature vectorization is performed on the local resource library to obtain multiple resource feature vectors of multiple local resources; the Pearson correlation coefficient is used to traverse and calculate multiple resource matching degrees between the multiple resource feature vectors and the first user portrait; a resource matching threshold is preset, and the resource matching threshold is used to traverse the multiple resource matching degrees to filter and obtain the recommended resource set.

[0144] Furthermore, the real-time permission matching unit is further configured to perform the following steps:

[0145] Taking the access frequency threshold as a constraint, the online statistical data is called to obtain multiple sample access channel topologies of multiple sample demand keywords; based on the knowledge graph, the multiple sample demand keywords and the multiple sample access channel topologies are associated and stored in the permission management module to complete the analysis data localization of the permission management module; M real-time demand keywords constituting the real-time demand characteristics are loaded into the permission management module, and M sample access channel topologies are obtained through retrieval and matching; the M sample access channel topologies are aggregated, and sample access channel extraction is performed based on the aggregation result to obtain the real-time permission demand.

[0146] Furthermore, the keyword extraction unit is further configured to perform the following steps:

[0147] The real-time user demand is extracted from the search box of the recommendation interface; the real-time user demand is verified for compliance; if the real-time user demand passes the compliance verification, after completing the content of the real-time user demand, NLP technology is used to extract keywords of the completion result to obtain real-time demand characteristics; if the real-time user demand fails the compliance verification, a compliance defect is generated and sent to the recommendation interface for visual warning.

[0148] Through the above detailed description of an AI-driven precise demand positioning method in this specification, those skilled in the art can clearly understand that this embodiment is an AI-driven precise demand positioning system. For the system disclosed in Example 2, since it corresponds to the method disclosed in Example 1 and has corresponding functional units and beneficial effects, the relevant parts can be referred to the method part description.

[0149] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An AI-driven accurate demand positioning method, characterized in that: The method comprises: After the target user enters real-time user needs in the recommendation interface, the data collection permission channel is activated through permission response; Under the sampling boundary constraints of the data collection permission channel, with the real-time user demand as the retrieval boundary constraint, crawler technology is used to perform multi-channel data backtracking to obtain user panoramic data; Based on channel relevance, the user panoramic data is subjected to demand trend analysis, and a demand panoramic insight result is output, wherein the demand panoramic insight result includes a short-term panoramic insight result and a long-term panoramic insight result; The short-term panoramic insight results and the real-time user needs are used as the first user profile to perform recommended resource positioning and matching, and a recommended resource set is output; The long-term panoramic insight results and real-time user needs are used as the second user profile to perform recommended resource positioning expansion to obtain a multi-level expanded resource set; Perform hierarchical visual lazy loading of the recommended resource set and the multi-level extended resource set in the recommendation interface; Wherein, after the target user inputs the real-time user demand in the recommendation interface, the data collection permission channel is activated through the permission response, including: Apply NLP technology to extract keywords from the real-time user needs to obtain real-time demand features; Input the real-time requirement characteristics into the authority management module, and match and output the real-time authority requirements; Performing a coverage judgment on the real-time permission requirement according to the historical authorization information of the target user to obtain a real-time permission request; According to the authorization response feedback of the target user to the real-time permission request, the real-time permission requirement is updated, and the mapping activation of the data collection permission channel is performed using the updated result; The step of inputting the real-time requirement characteristics into the authority management module and matching and outputting the real-time authority requirements includes: Taking the access frequency threshold as a constraint, the online statistical data is called to obtain multiple sample access channel topologies of multiple sample demand keywords; Based on the knowledge graph, the plurality of sample demand keywords and the plurality of sample access channel topologies are stored in the authority management module to complete the analysis data localization of the authority management module; Loading M real-time demand keywords constituting the real-time demand features into the authority management module, and obtaining M sample access channel topologies through retrieval and matching; Aggregate the M sample access channel topologies, and extract sample access channels based on the aggregation results to obtain the real-time permission requirements.

2. The AI-driven precise demand positioning method according to claim 1, characterized in that: The recommended resource set and the multi-level extended resource set are lazily loaded in a hierarchical visual manner on the recommendation interface, and the method includes: Analyze the compatibility of the recommended resource set with the real-time user demand, and construct a resource association map between the recommended resource set and the real-time user demand according to the analysis result; Taking the resource association map as a starting point, performing hierarchical resource association analysis on the multi-level extended resource set to obtain a multi-level extended resource map; According to the resource association graph, visually loading the recommended resource set on the recommendation interface; According to the interactive behavior of the target user, lazy loading of the multi-level extended resource set is performed on the recommendation interface according to the multi-level extended resource map.

3. The AI-driven precise demand positioning method according to claim 1, characterized in that: Under the sampling boundary constraint of the data collection permission channel, taking the real-time user demand as the retrieval boundary constraint, using crawler technology to perform multi-channel data backtracking to obtain user panoramic data, the method includes: Preset data backtracking time scale; Extracting a first authorization channel from the data collection authority channel, wherein the data collection authority channel includes K authorization channels; Taking the data backtracking time scale as the time boundary constraint and the real-time demand feature as the content boundary constraint, crawler technology is used to call user historical data in the first authorized channel to obtain first channel historical data; By analogy, the user historical data of the K authorized channels are called to obtain K-dimensional channel historical data, wherein the K-dimensional channel historical data constitutes the user panoramic data.

4. The AI-driven precise demand positioning method according to claim 3, characterized in that: Based on channel relevance, the user panoramic data is subjected to demand trend analysis, and a demand panoramic insight result is outputted. The method includes: 1 / 24 of the data backtracking time scale is used as the data segmentation time scale; Segmenting the K-dimensional channel historical data into K-stage historical data sequences according to the data segmentation time scale; Reorganize the K stage historical data sequences based on time sequence alignment to obtain multiple groups of stage historical data; After using TF-IDF to extract text features from the multiple sets of stage historical data, PCA is used to reduce the dimension of the extracted results to perform dimensionality reduction identification of behavioral features, thereby obtaining panoramic insight results for multiple stages; By performing long-term and short-term demand evolution analysis on the panoramic insight results of the multiple stages, the short-term panoramic insight results and the long-term panoramic insight results are output.

5. The AI-driven precise demand positioning method according to claim 4, characterized in that: By performing a long-term and short-term demand evolution analysis on the panoramic insight results of the multiple stages, the short-term panoramic insight results and the long-term panoramic insight results are outputted, and the method includes: 1 / 12 of the data backtracking time scale is used as the short-term demand evolution scale; Taking the short-term demand evolution scale as a constraint, extracting H stage panoramic insight results from the multiple stage panoramic insight results in reverse time; By performing cross-stage demand evolution analysis on the H stage demand feature vectors of the H stage panoramic insight results, a short-term resource preference vector is output as the short-term panoramic insight result; 11 / 12 of the data backtracking time scale is used as the long-term demand evolution scale; By analogy, with the long-term demand evolution scale as a constraint, the panoramic insight results of the multiple stages are subjected to a cross-stage demand evolution analysis to output the long-term panoramic insight results.

6. The AI-driven precise demand positioning method according to claim 1, characterized in that: The short-term panoramic insight result and the real-time user demand are used as the first user profile to perform recommended resource positioning matching, and a recommended resource set is output. The method includes: The first user portrait is obtained by fusing the short-term panoramic insight result and the real-time user demand through vector splicing; Perform resource feature vectorization on the local resource library to obtain multiple resource feature vectors of multiple local resources; The Pearson correlation coefficient is used to traverse and calculate the resource matching degrees between the plurality of resource feature vectors and the first user portrait; A resource matching threshold is preset, and the resource matching threshold is used to traverse the multiple resource matching degrees to screen and obtain the recommended resource set.

7. The AI-driven precise demand positioning method according to claim 1, characterized in that: Extracting keywords from the real-time user demand to obtain real-time demand features, the method includes: Extracting the real-time user demand from the search box of the recommendation interface; Performing compliance verification on the real-time user requirements; If the real-time user demand passes the compliance verification, after completing the content of the real-time user demand, NLP technology is applied to extract keywords from the completion result to obtain the real-time demand features; If the real-time user demand fails the compliance verification, a compliance defect is generated and sent to the recommendation interface for visual warning.

8. An AI-driven precise demand positioning system, characterized in that: The system is used to execute the AI-driven precise demand positioning method according to any one of claims 1 to 7, comprising: The response activation unit is used to activate the data collection permission channel through permission response after the target user enters the real-time user demand in the recommendation interface; A panoramic data collection unit is used to perform multi-channel data backtracking using crawler technology under the sampling boundary constraints of the data collection authority channel and taking the real-time user demand as the retrieval boundary constraints to obtain user panoramic data; A demand trend analysis unit, configured to perform demand trend analysis on the user panoramic data based on channel relevance, and output a demand panoramic insight result, wherein the demand panoramic insight result includes a short-term panoramic insight result and a long-term panoramic insight result; A recommended resource matching unit, configured to use the short-term panoramic insight result and the real-time user demand as the first user profile to perform recommended resource positioning matching and output a recommended resource set; A recommended resource expansion unit, configured to use the long-term panoramic insight results and real-time user needs as a second user profile to locate and expand recommended resources, thereby obtaining a multi-level expanded resource set; The visualization unit is used to perform hierarchical visualization lazy loading of the recommended resource set and the multi-level expanded resource set on the recommendation interface.

Images