A cross-platform e-commerce resource dynamic matching search method and system

Abstract

The application relates to the technical field of resource matching, and discloses a cross-platform e-commerce resource dynamic matching search method, which comprises the following steps: real-time analysis of commodity description information from different e-commerce platforms, extraction of core attributes of commodities and generation of unified semantic representation; dynamic generation of a conflict resolution strategy when cross-platform data conflicts occur, the conflict resolution strategy being executed by a local preprocessing module; acquisition of data content after conflict resolution, selection of a target edge node based on the data content and user geographic location information, and dynamic configuration of a cache data set of the target edge node; generation of corresponding multi-dimensional matching features based on user intent information in a natural query language and commodity data in the configured cache, determination of a candidate commodity set and corresponding search results based on the multi-dimensional matching features, acquisition of feedback information of the user on the search results, and updating of an intent analysis model and a matching strategy based on the feedback information. The application can realize efficient cross-platform resource matching.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of resource matching, in particular to a cross-platform e-commerce resource dynamic matching search method and system. BACKGROUND

[0002] With the globalization of e-commerce, the heterogeneity (such as format difference, semantic ambiguity) and real-time demand of cross-platform commodity data are increasingly prominent, and traditional methods rely on artificial rules or static mapping, which is difficult to dynamically adapt to multi-platform data structure, resulting in low matching efficiency and high delay. In addition, the existing technology solves the strategy of data conflict (such as inconsistent price and inventory) for fixed priority or local optimization, lacks a dynamic coordination mechanism based on user behavior and real-time demand, and the collaborative ability of edge computing and center server is insufficient, which further aggravates the system load and response delay problem.

[0003] The current cross-platform e-commerce system generally adopts independent modular design, and each function (such as data standardization, conflict detection, order synchronization) is separated from each other, which is difficult to form a closed loop optimization. As known from the above, how to realize efficient cross-platform resource matching still needs to be solved. SUMMARY

[0004] In order to realize efficient cross-platform resource matching, the present application provides a cross-platform e-commerce resource dynamic matching search method and system.

[0005] In the first aspect, the present application provides a cross-platform e-commerce resource dynamic matching search method, which adopts the following technical scheme:

[0006] A cross-platform e-commerce resource dynamic matching search method, comprising:

[0007] Real-time analysis of commodity description information from different e-commerce platforms at the data access end, extraction of core attributes of commodities and generation of unified semantic representation, commodity description information including text and image, core attributes including name, specification, price; The semantic representation establishes a semantic mapping relationship between cross-platform attributes based on a synonym replacement mechanism and attribute association rules based on context perception, and transmits the standardized data to the center database through a distributed consistency synchronization mechanism;

[0008] Based on the commodity attribute information in the standardized data and the historical behavior characteristics of the user, a conflict resolution strategy is dynamically generated when cross-platform data conflict occurs, and the conflict resolution strategy is executed by a local preprocessing module deployed in the edge node, wherein the local preprocessing module responds quickly to sudden data, and the local preprocessing module processes periodic data in batches;

[0009] After conflict resolution, the data content after conflict resolution is obtained, in the computing architecture composed of the edge node and the center server, a target edge node is selected based on the data content in combination with user geographic location information, and a cache data set of the target edge node is dynamically configured;

[0010] A natural query language input by a user is obtained, corresponding user intent information is parsed based on the natural query language, corresponding multi-dimensional matching features are generated based on the user intent information in combination with commodity data in the configured cache, a candidate commodity set is determined based on the multi-dimensional matching features through approximate nearest neighbor search, and corresponding search results are output based on the candidate commodity set, feedback information of the user on the search results is obtained, and an intent parsing model and a matching strategy are updated based on the feedback information.

[0011] Optionally, when generating the unified semantic representation, the method further comprises:

[0012] The commodity core attributes are associated and mapped with preference labels in the user historical behaviors, and corresponding cross-platform behavior patterns are extracted;

[0013] Corresponding dynamic attribute weights are generated based on the cross-platform behavior patterns, the priority of different attributes in the semantic representation is calculated through a dynamic attribute weight generation module, and the priority of different attributes in the semantic representation is dynamically updated through an attribute priority adjustment module.

[0014] Optionally, in the conflict resolution strategy, the method further comprises:

[0015] The source platform of the conflict data and the geographic location information of the target user are obtained, and associated features of the source platform and the region are extracted based on the source platform and the geographic location information;

[0016] Cross-regional data consistency check rules are dynamically generated based on the associated features, wherein the cross-regional data consistency check rules preferentially use platform data with the highest matching degree with the region where the user is located as a conflict resolution benchmark.

[0017] Optionally, when dynamically configuring the cache data set of the target edge node, the method further comprises:

[0018] Cross-platform price fluctuation trends of commodities and user purchase periodicity features are extracted, and cache update priorities of the commodities are divided based on the cross-platform price fluctuation trends and the user purchase periodicity features;

[0019] Based on the cache update priorities, a short-cycle forced refresh strategy is performed on commodities with frequent price fluctuations, and a long-cycle passive update strategy is performed on commodities with stable prices.

[0020] Optionally, when generating the multi-dimensional matching features, the method further comprises:

[0021] retrieve user intention information and cross-platform attribute association rules, and generate corresponding dynamic attribute filtering conditions based on the user intention information and the cross-platform attribute association rules;

[0022] narrow the range of candidate goods step by step based on the dynamic attribute filtering conditions through a hierarchical screening mechanism, and preferentially match core attributes explicitly mentioned in the user intention.

[0023] Optionally, when updating the matching strategy based on the feedback information, the method further comprises:

[0024] obtaining user operation behaviors of the user on the search results, the user operation behaviors including clicking, purchasing, and collecting the search results, obtaining cross-platform attribute features corresponding to the goods, performing association analysis based on the user operation behaviors and the cross-platform attribute features, and generating attribute association enhancement rules driven by user behaviors;

[0025] updating the synonym replacement mechanism and the attribute alignment logic through a semantic mapping optimization mechanism based on the attribute association enhancement rules.

[0026] Optionally, when parsing the natural query language, the method further comprises:

[0027] obtaining semantic mapping relationships between cross-platform attributes, and generating a multilingual compatible standardized query template based on the semantic mapping relationships between the natural query language and the cross-platform attributes;

[0028] automatically adapting attribute description habits of different platforms through a semantic expansion mechanism based on the standardized query template, and generating cross-platform compatible query features.

[0029] In a second aspect, the present application provides a cross-platform e-commerce resource dynamic matching search system, which adopts the following technical solution:

[0030] A cross-platform e-commerce resource dynamic matching search system comprises:

[0031] A semantic mapping relationship establishment module is configured to parse real-time commodity description information from different e-commerce platforms at a data access end, extract core attributes of the commodities, and generate unified semantic representations, the commodity description information including text and images, and the core attributes including names, specifications, and prices; the semantic representations are based on a synonym replacement mechanism and attribute association rules for context-aware semantic mapping between cross-platform attributes, and the standardized data is transmitted to a central database through a distributed consistency synchronization mechanism.

[0032] The conflict resolution strategy generation module is configured to generate a conflict resolution strategy based on the standardized data and historical behavior characteristics of the user when a cross-platform data conflict occurs, and the conflict resolution strategy is executed by a local preprocessing module deployed on an edge node, wherein the local preprocessing module is configured to quickly respond to burst data and the local preprocessing module is configured to batch process periodic data.

[0033] The target edge node selection module is configured to obtain data content after conflict resolution, select a target edge node based on the data content and user geographic location information in a computing architecture formed by the edge node and the central server, and dynamically configure a cache data set of the target edge node.

[0034] The update module is configured to obtain a natural query language input by the user, parse corresponding user intent information based on the natural query language, generate corresponding multi-dimensional matching features based on the user intent information and product data in the configured cache, determine a candidate product set based on the multi-dimensional matching features through approximate nearest neighbor search, output corresponding search results based on the candidate product set, obtain feedback information of the user on the search results, and update the intent parsing model and the matching strategy based on the feedback information.

[0035] In a third aspect, the present application provides a cross-platform e-commerce resource dynamic matching search system, which adopts the following technical solution:

[0036] A cross-platform e-commerce resource dynamic matching search system includes a processor, and the processor runs a program of the cross-platform e-commerce resource dynamic matching search method according to any one of the preceding aspects.

[0037] In a fourth aspect, the present application provides a storage medium, which adopts the following technical solution:

[0038] A storage medium stores a program of the cross-platform e-commerce resource dynamic matching search method according to any one of the preceding aspects.

[0039] In summary, the present application has at least one of the following beneficial technical effects:

[0040] By dynamic semantic mapping and distributed consistency synchronization, efficient standardization of cross-platform product data is achieved. At the data access end, the system combines a context-aware synonym replacement mechanism and attribute association rules to convert product descriptions (such as text and images) of heterogeneous platforms into unified semantic representations, solving the matching obstacles caused by differences in cross-platform attribute definitions. At the same time, based on the distributed consistency synchronization mechanism, only differential data is transmitted instead of full data, significantly reducing the storage pressure and data synchronization delay of the central database.

[0041] In the resource matching stage, the scheme realizes efficient cross-platform matching through multi-dimensional feature fusion and dynamic strategy optimization. First, the system generates multi-dimensional matching conditions by combining user intent analysis and product features in cache data, and preferentially matches the core attributes (such as name, specification) explicitly mentioned by the user through a hierarchical screening mechanism, gradually narrowing down the range of candidate products, and avoiding invalid matching. Second, based on the association analysis of user feedback (click, purchase, collection) and cross-platform attribute features, the system dynamically generates attribute association enhancement rules, and reversely optimizes the semantic mapping logic (such as synonym replacement and attribute alignment), forming a closed-loop improvement. Finally, through the multi-language compatible standardized query template and semantic expansion mechanism, the system automatically adapts to the attribute description habits of different platforms, eliminating the influence of language and format differences on matching accuracy. The response time of cross-platform resource matching is shortened to milliseconds, and the matching accuracy is significantly improved, finally realizing efficient, low-latency cross-platform resource dynamic matching. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 is a flow chart of a cross-platform e-commerce resource dynamic matching search method according to an exemplary embodiment.

[0043] Figure 2 is a structural block diagram of a cross-platform e-commerce resource dynamic matching search system according to an exemplary embodiment. DETAILED DESCRIPTION

[0044] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0045] In the description of the present specification, the description of the terms "certain embodiments", "one embodiment", "some embodiments", "illustrative embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in connection with the described embodiments or examples are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0046] The embodiments of the present application disclose a cross-platform e-commerce resource dynamic matching search method, referring to Figure 1 , comprising:

[0047] S100, real-time parsing of product description information from different e-commerce platforms at the data access end, extracting the core attributes of the product and generating a unified semantic representation, the product description information includes text and images, the core attributes include name, specification, price; the semantic representation establishes a semantic mapping relationship between cross-platform attributes based on a context-aware synonym replacement mechanism and attribute association rules, and transmits the standardized data to the central database through a distributed consistency synchronization mechanism.

[0048] Wherein, S100 is to standardize heterogeneous product data (such as text, image) from different e-commerce platforms, generate a unified semantic representation, and transmit it to the central database through a distributed consistency synchronization mechanism. The following is the specific execution steps and function analysis of S100:

[0049] Step 1, real-time acquisition of product description information (including text description, picture, price, specification, etc.) from the API interface or crawler system of multiple e-commerce platforms, and use natural language processing (NLP) technology to parse the text content and extract the core attributes of the product (such as name, specification, price).

[0050] At the same time, through image recognition technology, the features of the product image (such as color, shape) are extracted to supplement the lack of text description. For example, "waterproof phone case" in the description of one e-commerce platform and "IP68 level dustproof and waterproof protective cover" in another e-commerce platform will be parsed as the same semantic core attribute. The above process solves the problem of different e-commerce platforms describing the same product in different ways (such as "waterproof" and "IP68 waterproof"), and through multi-modal (text + image) parsing ensures the completeness of product information, avoiding matching failure due to missing of single data source.

[0051] Step 2, based on the semantic dictionary of e-commerce platforms and user behavior data, through the context-aware synonym replacement mechanism, dynamically replace the cross-platform synonyms (such as "phone case" → "PhoneCase"), and adjust the replacement priority according to the context (for example, "waterproof phone case" may be replaced by "WaterproofPhoneCase" instead of "CaseforMobile").

[0052] Meanwhile, through the pre-set rule base (such as the "brand + model" combination uniquely identifies the goods) and the machine learning model, the mapping relationship of cross-platform attributes is established (such as mapping the "memory capacity" of one e-commerce platform to the "RAMSize" of another e-commerce platform as the same attribute). Finally, a unified semantic vector representation is generated as the basis for standardized data. The above process eliminates the semantic ambiguity problem of cross-platform goods description (such as "waterproof" and "IP68 waterproof"), ensures the accurate matching of goods attributes on different platforms, and provides consistent attribute definitions for subsequent cross-platform resource matching (such as search, recommendation), reducing matching errors caused by semantic differences.

[0053] Step 3, the standardized data is transmitted to the central database through a distributed consistency synchronization mechanism (such as a consistent hash algorithm), only transmitting differential data (such as newly added goods, price changes), rather than full data, to reduce network bandwidth consumption. For example, if the price of a certain product on an e-commerce platform is updated from 100 yuan to 120 yuan, the synchronization mechanism only transmits the incremental data "price = 120 yuan", rather than the entire product description.

[0054] The distributed consistency synchronization mechanism significantly reduces the storage and bandwidth overhead of cross-platform data transmission through incremental synchronization and differential data compression (such as traditional full synchronization requires the transmission of GB data, while incremental synchronization only requires the transmission of several KB), while ensuring that the data state of the central database and each edge node is consistent, avoiding the distortion of matching results caused by data delay or conflict (such as users seeing old prices when searching).

[0055] Step 4, the standardized data is written into the central database (such as MySQL or Elasticsearch), and an index is established to support subsequent fast retrieval. The database design needs to be compatible with the attribute structure of multiple platforms (such as the "store rating" of one e-commerce platform and the "user evaluation" of another e-commerce platform are mapped to the unified "reputation score" field).

[0056] The above process provides unified data support for subsequent conflict resolution, cache configuration, and user intent analysis by centrally managing data, and achieves millisecond-level retrieval of cross-platform goods through pre-built indexes and optimized query statements (such as Elasticsearch's inverted index) (such as when a user inputs "waterproof phone case", the matching results can be returned within 100ms).

[0057] The synonym replacement and attribute association rules based on context perception are used to convert the product descriptions of different e-commerce platforms into unified semantic representations, eliminate format, terminology and semantic differences, and provide a standardized basis for cross-platform matching. At the same time, the distributed consistency synchronization mechanism is used to reduce the full data transmission, reduce the load of the central database and ensure the real-time data. In addition, the standardized data provides a reliable premise for subsequent conflict resolution, cache configuration and user intention analysis, and supports the real-time processing of millions of goods with high concurrency through multi-modal analysis and incremental synchronization mechanism, and meets the dynamic needs of e-commerce platforms.

[0058] S200, based on the standardized data of the product attribute information and the historical behavior characteristics of the user, a conflict resolution strategy is dynamically generated when a cross-platform data conflict occurs, and the conflict resolution strategy is executed by a local preprocessing module deployed in an edge node, wherein the local preprocessing module responds quickly to the burst data, and the local preprocessing module processes the periodic data in batches.

[0059] S200 is based on the standardized product attribute information and the historical behavior characteristics of the user, and a conflict resolution strategy is dynamically generated when a cross-platform data conflict occurs, and the conflict resolution strategy is executed by a local preprocessing module deployed in an edge node. The following are the specific execution steps and function analysis:

[0060] Step 1, first, the system detects potential conflicts in real time by monitoring the differences in standardized product attribute data (such as price, inventory, specifications) between different platforms. For example, if the price of a product on one e-commerce platform is 100 yuan and the price on another e-commerce platform is 120 yuan, the system will trigger a conflict signal.

[0061] At the same time, the context background of the conflict is analyzed in combination with the historical behavior characteristics of the user (such as search preferences, purchase records, geographic location). For example, when the user is located in Beijing and tends to trust the logistics service of a certain e-commerce platform, the system may prefer to use the price data of this e-commerce platform as the benchmark. This process avoids user complaints or transaction failures caused by inconsistent data by accurately positioning the conflict attributes (such as price, inventory) and the scope of influence, and improves the personalization of conflict resolution strategies and user experience through context-aware decision-making (such as the priority of the platform frequently visited by the user).

[0062] Step 2, secondly, the system dynamically generates resolution strategies based on the type of conflict data (such as price, inventory), user characteristics (such as geographic location, platform preference), and product attributes (such as price fluctuation trends). For example, for high-frequency fluctuation products (such as promotional products), the data of the platform with the latest update is preferred; for stable products (such as daily necessities), the weighted average of multiple platform data is used. The processing logic for inventory conflicts includes prioritizing data from platforms with inventory and marking the conflict source. The generated strategies are encapsulated as executable instructions (such as "use e-commerce platform price" or "merge inventory data") and passed to the local preprocessing module of the edge node. This process avoids the rigidity of a single rule by flexibly dealing with complex scenarios (such as price, inventory, and description conflicts), reduces the load on the central server, and improves response efficiency.

[0063] Step 3, thirdly, the local preprocessing module of the edge node executes the conflict resolution strategy through two mechanisms: sudden data quick response and periodic data batch processing. For sudden conflicts (such as sudden price changes caused by promotional activities), the local module directly executes the preset strategy (such as "force the latest price") without waiting for the central server to confirm; for example, when a product on an e-commerce platform suddenly drops to 50 yuan, the local module immediately overrides the original data and synchronizes it to the cache node.

[0064] For periodic conflicts (such as daily inventory updates), the local module checks data consistency in batches at fixed time intervals (such as every hour) and performs merging or overriding operations; for example, comparing the inventory data of one e-commerce platform with that of another, if the difference is less than a threshold (such as 5%), the average value is merged, and if the difference is too large, an exception is marked and manual review is triggered. This mechanism balances real-time and stability (key conflicts are resolved immediately, and periodic operations avoid system jitter), and takes advantage of edge computing (reduces cross-network communication overhead and reduces delay to milliseconds) to reduce the pressure on the central server.

[0065] Step 4, finally, the local preprocessing module modifies the conflict data (such as adjusting prices, merging inventory) according to the generated strategies and updates the product information in the edge node cache. Subsequently, through a distributed consistency synchronization mechanism (such as a consistent hash algorithm), the conflict-resolved data is transmitted to the central database to ensure global data consistency. For example, if the local module uses the price of one e-commerce platform to override the data of another, the central database will be updated synchronously to avoid cross-platform information differences. This process ensures that all platform users see unified product information (such as price, inventory), avoiding user confusion or transaction failures caused by data delays. At the same time, if the edge node executes a strategy and finds an exception (such as a misjudged conflict), it can quickly roll back to the pre-conflict state through the central database's log records, achieving fault tolerance and rollback capabilities.

[0066] By dynamically generating conflict resolution strategies (such as prioritizing trusted data sources based on user behavior characteristics) and relying on the local preprocessing module of the edge node, millisecond-level response to cross-platform data conflicts (such as price and inventory differences) is achieved, significantly reducing the load on the central server (such as processing 100,000 conflict data per second during Double Eleven); combined with personalized strategies (such as prioritizing user frequently visited platform data) to reduce user complaints caused by data inconsistency, while supporting high-concurrency real-time conflict resolution for millions of goods through bursty fast response and periodic batch processing mechanisms, ensuring the consistency of cross-platform commodity information and user experience.

[0067] S300, after conflict resolution, the data content after conflict resolution is obtained, in the computing architecture composed of edge nodes and central servers, based on the data content, the target edge node is selected combined with user geographic location information, and the cache data set of the target edge node is dynamically configured.

[0068] Among them, after the cross-platform data conflict is resolved, the target edge node is dynamically selected based on the user geographic location information and the data content after conflict resolution, and the cache data set of the target edge node is configured. The following is the specific execution steps and function analysis of S300:

[0069] Step 1, the system first obtains the standardized commodity attribute data (such as price, inventory, specification) and the execution result of the conflict resolution strategy (such as "adopting the price of one of the e-commerce platforms" or "merging inventory data") from the conflict resolution module (S200). The data after conflict resolution usually includes commodity core attributes (such as brand, model, price, inventory status), conflict resolution marks (such as "one of the e-commerce platform data is preferred" or "merged data") and time stamp (used to judge the timeliness of cache). For example, if the price of a commodity on one e-commerce platform is 100 yuan and on another e-commerce platform is 120 yuan, the conflict resolution may mark "the e-commerce platform with commodity price of 100 yuan is preferred", and record the update time as the current timestamp.

[0070] The above process ensures the freshness of the cache data through the timestamp, avoids users from getting old data due to cache expiration (such as price recovery after the end of a promotion), and provides a consistent data basis for subsequent cache configuration, ensuring the data consistency of edge nodes and central database.

[0071] Step 2, obtain the user's geographic location (e.g., "Beijing user" or "Shanghai user") through the user's IP address, GPS location, or operator base station information, and refine the geographic location label (e.g., "Beijing user prefers a certain e-commerce platform") by combining the user's historical behavior (e.g., frequently visited e-commerce platforms). Based on this, the system follows the principle of proximity to preferentially select the edge node closest to the user's geographic location (e.g., "Beijing user → Beijing edge node"), while considering platform preference adaptation, and selecting the optimal node among multiple candidate nodes according to the load balancing strategy. For example, if there are three edge nodes in Beijing, the system will select the node with the lowest load that supports direct connection of data for a certain e-commerce platform.

[0072] The above significantly reduces network latency (e.g., from seconds-level response for cross-regional transmission to milliseconds-level for local transmission) by reducing cross-regional data transmission paths, and ensures that users preferentially access data from frequently visited platforms (e.g., "Beijing user sees real-time prices of a certain e-commerce platform") through platform preference adaptation, thereby improving user experience.

[0073] Step 3, dynamically generate cache strategies based on product attributes (e.g., hot-selling products, promotional products) and user needs (e.g., high-frequency search keywords), and preferentially cache high-value data (e.g., "hot-selling mobile phones") to target edge nodes, and allocate cache capacity based on product popularity (e.g., hot-selling products occupy a larger proportion, and cold products are loaded on demand). For high-frequency changing attributes such as price and inventory, use incremental update mechanism (e.g., only synchronize price changes); for static attributes (e.g., product description), refresh the cache at fixed intervals (e.g., every hour). Then, the system writes the data after conflict resolution (e.g., "the price of a certain e-commerce platform = 120 yuan") to the cache of the target edge node, and establishes multi-dimensional indexes (e.g., "product ID + platform + geographic location") to speed up query efficiency.

[0074] The above process significantly improves cache hit rate (e.g., hot-selling product cache hit rate can exceed 90%) through dynamic priority allocation and cache update mechanism, while reducing the load of the central server (only processing cold data requests), and flexibly responding to traffic fluctuations (e.g., increasing hot-selling product cache during promotions), supporting high-concurrency access requirements for millions of products.

[0075] S400, obtain the natural query language input by the user, parse the corresponding user intent information based on the natural query language, generate corresponding multi-dimensional matching features based on the user intent information and the product data in the configured cache, determine a candidate product set based on the multi-dimensional matching features through approximate nearest neighbor search, output corresponding search results based on the candidate product set, obtain feedback information of the user on the search results, and update the intent parsing model and the matching strategy based on the feedback information.

[0076] Where, by analyzing the intent information in the user's natural language query, combining the cache data to generate multi-dimensional matching features, and based on the approximate nearest neighbor search to output the search results, finally through the user feedback to dynamically optimize the intent analysis model and matching strategy. The following is the specific execution steps of S400 and the role analysis:

[0077] Step 1, first through natural language processing (NLP) technology to the user input natural language query (such as "find a waterproof performance good mobile phone") word segmentation, entity recognition and semantic analysis, extraction of key intent (such as "waterproof performance", "mobile phone").

[0078] Then, the extracted intent is classified as functional requirements (such as "waterproof"), price range (such as "2000 yuan or less"), brand preference (such as "A") and so on, and dynamically allocate weights according to user historical behavior (such as search records, purchase preferences). For example, if the user searches for "waterproof mobile phone" many times, the "waterproof" intent weight will be higher than other attributes. This process accurately understands user needs through structured intent information, avoids search result deviation due to natural language ambiguity, and improves the matching degree of search results (such as preferentially displaying user's frequently purchased brands) by combining personalized weight allocation.

[0079] Step 2, get the configured edge node cache data (such as hot-selling goods, promotional goods standardized attribute data), and combine it with the user intent information to construct a multi-dimensional matching feature vector. For example, "waterproof" is mapped to "IP68 waterproof" and other standardized attributes, if the user's intent includes "2000 yuan or less", filter out goods with a price higher than 2000 yuan; if the user prefers "A", increase the matching weight of A goods. Then, the system converts the multi-dimensional features into a numerical vector for subsequent retrieval.

[0080] The above process covers multi-dimensional requirements such as function, price, brand, etc., avoiding the limitations of single-dimensional matching, while reusing the cache data of S300 to reduce the real-time query pressure on the central database, significantly improving the response speed.

[0081] Step 3, based on the generated multi-dimensional feature vector, the system uses efficient approximate nearest neighbor (ANN) algorithm (such as Faiss, HNSW) to quickly retrieve the most similar candidate goods set in the cache data. For example, for the feature vector of "IP68 waterproof + 2000 yuan or less + A brand", retrieve the candidate goods (such as a mobile phone B corresponding to A) that meet the requirements in the cache. Then, according to the similarity score, sort the candidate goods, and filter out invalid goods (such as "out of stock" goods) combined with real-time inventory, promotional information, etc.

[0082] The above process meets the high concurrency scenario requirements through millisecond-level large-scale data retrieval (such as million-level commodity library), and ensures that the search results are highly relevant to the user's intention (such as preferentially displaying A mobile phones with strong waterproof performance) through multi-dimensional feature vectors and similarity sorting.

[0083] Step 4, the system trains the intention analysis model (such as BERT, Transformer) based on user feedback data (such as clicked keywords), optimizes the accuracy and generalization ability of intention recognition. For example, if the user frequently associates "waterproof" with "IP68 level", the model will strengthen this mapping relationship. At the same time, according to the feedback data, dynamically adjust the matching strategy: increase the brand preference weight, optimize the ANN algorithm parameters (such as increase the similarity threshold of the price dimension), and update the cache data priority (such as improve the cache hit rate of the user's high-frequency clicked goods). This process realizes the continuous evolution of the intention analysis model and the matching strategy through the feedback loop, adapts to the long-term changes in user demand (such as from "waterproof" preference to "battery life" preference), and reduces the operation and maintenance cost, while improving the accuracy of search results and user satisfaction through automatic strategy adjustment.

[0084] Step 4, the system trains the intention analysis model (such as BERT, Transformer) based on user feedback data (such as clicked keywords), optimizes the accuracy and generalization ability of intention recognition. For example, if the user frequently associates "waterproof" with "IP68 level", the model will strengthen this mapping relationship. At the same time, according to the feedback data, dynamically adjust the matching strategy: increase the brand preference weight, optimize the ANN algorithm parameters (such as increase the similarity threshold of the price dimension), and update the cache data priority (such as improve the cache hit rate of the user's high-frequency clicked goods). This process realizes the continuous evolution of the intention analysis model and the matching strategy through the feedback loop, adapts to the long-term changes in user demand (such as from "waterproof" preference to "battery life" preference), and reduces the operation and maintenance cost, while improving the accuracy of search results and user satisfaction through automatic strategy adjustment.

[0085] The multi-modal data analysis and semantic standardization are used to solve the heterogeneous problem of cross-platform commodity information and build a unified data foundation. S100 uses natural language processing (NLP) and image recognition technology to extract the core attributes of commodities (such as name, price, and specifications), and uses a context-aware synonym replacement mechanism (such as "waterproof" mapped to "IP68 waterproof") and attribute association rules to eliminate semantic differences, ensuring that commodity descriptions on different platforms can be accurately matched. At the same time, combined with a distributed consistency synchronization mechanism (such as incremental data transmission), the load of the central database is significantly reduced and the real-time performance of the data is guaranteed. S200 further generates conflict resolution strategies (such as preferentially using data from the platform frequently visited by the user) through the local preprocessing module of the edge node, realizes millisecond-level response (such as price and inventory differences), and avoids performance bottlenecks caused by centralized processing. This "edge fast response + center coordination" architecture supports high-concurrency real-time processing of millions of commodities and provides efficient and consistent data support for cross-platform resource matching.

[0086] Efficient resource matching is realized through multi-dimensional feature vectors and approximate nearest neighbor (ANN) search algorithms, and continuous optimization is realized with user feedback. S300 dynamically configures edge node caching based on user geographic location and behavior preferences (such as preferentially loading hot-selling commodities), combines incremental updates with periodic batch processing mechanisms, significantly improves cache hit rates (such as more than 90%), and reduces the pressure on central servers. S400 analyzes user intent through NLP (such as "waterproof mobile phone"), generates multi-dimensional matching features (function + price + brand) based on cache data, uses ANN algorithms to quickly search for candidate commodities in a million-level commodity library, and dynamically adjusts the intent analysis model (such as BERT) and matching strategy (such as weight allocation) based on user feedback (clicks, purchases). This "data standardization → conflict resolution → cache optimization → intent analysis → feedback loop" link not only realizes real-time and accurate matching of cross-platform resources, but also continuously improves user experience through personalized strategies and adaptive optimization, ultimately forming an efficient and flexible cross-platform e-commerce matching system.

[0087] In the embodiments of the present application, when generating a unified semantic representation, the method further comprises:

[0088] Step 1, first, extract user preference tags from user's historical behavior data (such as search keywords, click records, purchase records, and product ratings), such as "waterproof", "high cost performance", "A brand" and so on. Then, make semantic association between these fuzzy preference tags and the core attributes of the product, such as mapping "waterproof" to the "waterproof level" attribute (such as IP68, IP67), and mapping "high cost performance" to the "price / performance ratio" attribute. At the same time, by analyzing the user behavior data of different platforms, identify the common needs and different preferences across platforms, such as finding that the "waterproof" attribute has significantly increased in attention on multiple platforms, while "authenticity guarantee" is a core demand of users on one e-commerce platform.

[0089] Step 2, in the dynamic weight calculation stage, the system will integrate the cross-platform behavior pattern data (such as "waterproof" attribute proportion 30%, "price" attribute proportion 20%) and the specific values of product attributes (such as "IP68 level waterproof = 1.0", "within 2000 yuan = 0.8"), to calculate the dynamic weight of each attribute. For example, if the "waterproof" attribute has a high proportion in the behavior pattern and the corresponding product attribute value is high, its weight will be increased to 0.5, while the "price" attribute may be assigned a weight of 0.3.

[0090] Then, the priority adjustment module will dynamically correct the weight according to real-time user feedback (such as click rate, conversion rate), such as further increasing the weight of the "waterproof" attribute when the click rate of the product with this attribute significantly increases. At the same time, the system will set threshold rules (such as "when the weight of a certain attribute exceeds 0.7, trigger priority lock"), to avoid excessive tilt in weight allocation, so as to balance between global demand (such as cross-platform high-frequency attributes) and personalized demand (such as specific user preference brand).

[0091] Step 3, finally, the system will inject the dynamically calculated attribute weight into the semantic representation vector, replacing the original static weight configuration. For example, the original static weight "brand = 0.2, waterproof = 0.4, price = 0.4" will be updated to "brand = 0.3, waterproof = 0.5, price = 0.3".

[0092] The above updating process will be synchronized to the cache of the edge node, ensuring that high-priority attributes (such as "waterproof") are loaded first during retrieval. At the same time, the system will optimize retrieval efficiency through Approximate Nearest Neighbor (ANN) algorithm, and preferentially match products with high-weight attributes. For example, for the user input "IP68 level waterproof + 2000 yuan within + A brand" query, the system will preferentially retrieve products with high "waterproof" attribute matching degree, thereby improving the relevance of search results. In addition, cache priority optimization can also reduce invalid retrieval of low-weight attribute mismatched products, reducing system load and improving response speed.

[0093] By dynamically associating user behavior with product attributes, adjusting attribute weights in real time, and optimizing semantic representation priorities, adaptive optimization of cross-platform resource matching is achieved. Its role is to accurately capture user's vague needs (such as converting "waterproof" to IP68 level waterproof attribute weight), dynamically respond to market changes (adjust weight distribution by combining cross-platform behavior patterns and real-time click rate), and significantly reduce system load and improve matching efficiency in high-concurrency scenarios through cache priority and search algorithm optimization.

[0094] In the conflict resolution strategy in the embodiments of the present application, the method further comprises:

[0095] Step 1: When a data conflict occurs, the system first determines the source platform of the conflicting data and the geographic location of the target user. Then, the system analyzes historical data or preset rules to mine the relevance of different platforms and user regions. For example, statistics show that users of a e-commerce platform are mostly concentrated in China, users of a e-commerce platform prefer genuine product guarantees, and users of an overseas platform pay more attention to logistics timeliness; at the same time, combined with regional characteristics, the matching degree of platforms and regions is quantified by statistical analysis or machine learning model.

[0096] Step 2: After extracting the association characteristics of platforms and regions, the system dynamically adjusts the priority of the verification rules according to the matching degree. For example, if the user is located in China and the matching degree of a and Chinese users is the highest, the data of a e-commerce platform is preferred as the verification reference; if the user is located in another place and the matching degree of b e-commerce platform and the region is low, the priority of b e-commerce platform data is reduced. For conflict fields (such as price, inventory, and logistics information), the system generates differentiated verification rules. For example, the price field preferentially verifies the data of the platform with the highest matching degree, and if there is a conflict, it refers to the data of other platforms; the logistics information field preferentially adopts the data of the platform with high matching degree to the user's region.

[0097] Step 3: According to the generated cross-regional verification rules, the data of the platform with high matching degree is preferentially adopted as the reference to correct the conflicting data of other platforms. Then, the system collects user feedback (such as click rate and complaint rate) after the conflict is resolved to evaluate the actual effect of the dynamic rules. For example, if it is found that "the matching degree of b e-commerce platform and the corresponding local users is improved", the priority weight is adjusted to reflect this change. Through continuous feedback loop, the system can dynamically correct the association characteristics of platforms and regions, avoid rule rigidity (such as changes in user preferences in emerging markets), and automatically adapt to new scenarios (such as the addition of new user groups in local markets) when expanding cross-regional business.

[0098] By dynamically associating user geographic location with platform characteristics to generate verification rules, regional user preferences are accurately adapted (such as Chinese users prefer to verify local platform data), invalid verification steps are reduced, high matching degree data is preferentially corrected, response time is shortened, and cross-regional business scenarios are self-adaptively expanded through real-time feedback optimization rules iteration, significantly reducing manual intervention and maintenance costs.

[0099] In the embodiments of the present application, when the cache data set of the target edge node is dynamically configured, the method further comprises:

[0100] Step 1: First, collect the price data of goods on different platforms through a data interface or an automated tool, analyze the fluctuation frequency and amplitude, for example, identify goods that are reduced by 5% daily during a promotion period, or goods that are stable in price during a non-promotion period. At the same time, combined with historical data, determine whether the goods are in a high fluctuation period (such as Double 11, Black Friday) or a stable period (such as daily sales).

[0101] In addition, based on user historical behavior (such as purchase records, browsing period), analyze the purchase periodicity characteristics of goods, for example, distinguish between high-frequency purchase goods (such as milk, toilet paper) and low-frequency purchase goods (such as home appliances), and further refine the classification combined with the type of goods.

[0102] Step 2: According to the price fluctuation trend and purchase periodicity characteristics of the goods, the system dynamically divides the cache update priority. Goods with frequent price fluctuations (such as promotional goods, seasonal goods) are classified as high priority and require a short-cycle forced refresh strategy (such as updating every hour); goods with stable prices (such as daily necessities, basic goods) are classified as low priority and can execute a long-cycle passive update strategy (such as daily or weekly updates). For goods with purchase periodicity (such as goods that are purchased monthly), further combine the price fluctuation to subdivide the priority: for example, high-fluctuation and high-frequency purchase goods (such as promotional milk) remain high priority, while low-fluctuation and low-frequency purchase goods (such as air conditioner filters) are adjusted to low priority.

[0103] Subsequently, the system deploys high-priority goods to edge nodes close to the user area (such as front-end nodes in a multi-level cache architecture) with a short life cycle (such as cache invalidation for 1 hour), and low-priority goods to regional center nodes with a long life cycle (such as cache invalidation for 24 hours). The purpose of the above process is to optimize cache resource allocation, avoid user loss due to outdated data for high-priority goods, reduce invalid refresh costs for low-priority goods, and shorten user access delay by preferentially loading high-priority goods on edge nodes.

[0104] Step 3, for high-priority goods (such as promotional goods), the system regularly pulls the latest price data from the source station (such as hourly price synchronization through API) and updates the edge node cache; if an abnormal price fluctuation (such as a violation of low price) is detected, a forced refresh is triggered immediately (such as updating the cache through an active API request). For low-priority goods (such as daily necessities), only the latest data is pulled from the source after the cache expires (such as updating at midnight every day), and combined with the user's purchase cycle (such as the user's average consumption cycle of 68 days), the cache is updated in advance before the active period (such as 60 days) to avoid outdated data affecting conversion rates.

[0105] The role of the above process is to dynamically adapt to market changes (such as timely response to price callback after promotion ends), avoid user complaints caused by outdated edge node cache, and reduce unnecessary source requests (such as low-priority goods only updating when expired) through differentiated update strategies, saving bandwidth and server resources.

[0106] By dynamically adapting to commodity price fluctuations and user purchase cycles, differentiated edge node cache strategies are developed to achieve high-frequency refresh of high-priority goods (such as promotional goods updated every hour) and long-cycle updates of low-priority goods (such as daily necessities updated daily), accurately matching the dynamic changes of market-sensitive goods and user high-frequency demand, and reducing system load through resource hierarchical management; at the same time, relying on the edge node to load high-priority data first, shortens the response time and avoids user loss caused by price invalidation (such as timely synchronization of promotional goods cache), ultimately ensuring data timeliness while optimizing cache efficiency and user experience.

[0107] In the embodiments of the present application, when generating multi-dimensional matching features, the method further comprises:

[0108] Step 1, first extract core needs by analyzing user input content (such as search keywords, product reviews, customer service conversations), for example, identify explicit needs such as "waterproof" "A" "price range" from "waterproof mobile phone A2000 yuan below". At the same time, combined with user historical behavior and current scene, supplement implicit needs (such as when the user recently pays attention to "sports scene", it implies "durability" demand). Then, the system extracts standardized attribute rules (such as "waterproof" → "IP68 waterproof" "high cost performance" → "price / performance ratio") from the pre-set cross-platform attribute mapping table, and dynamically adjusts the rule weight according to the platform characteristics (such as b e-commerce platform prioritizes "authenticity guarantee", a e-commerce platform focuses on "low price").

[0109] Step 2, based on the extracted user intent and cross-platform rules, generate multi-dimensional dynamic filtering conditions. For example, the explicit requirement "waterproof" is converted to "IP68 level waterproof = 1.0", "A brand = 1.0", "price ≤ 2000 yuan"; the implicit requirement "sports scene" is converted to "screen drop resistance ≥ 80%", "battery endurance ≥ 4000 mAh". At the same time, combined with platform rule adjustment condition priority (such as b e-commerce platform priority verification "genuine product guarantee = 1.0"). Then, the system converts these conditions into executable query statements (such as SQL statements or search engine query expressions), and supports combination operators (such as "AND" "OR") and priority nesting (such as "(waterproof AND brand A) OR (high cost performance AND price < 1500)").

[0110] Step 3, gradually narrow down the range of candidate goods through a hierarchical screening mechanism. The first layer is the core attribute hard filter, only retaining goods that meet the user's explicit requirements (such as "IP68 waterproof" "A brand" "price ≤ 2000 yuan"); the second layer is the implicit attribute soft sorting, which sorts the first layer results according to the implicit requirements (such as "screen drop resistance" "battery endurance"), and preferentially displays high matching degree goods; the third layer is the platform rule verification, which supplements the verification according to the user's platform (such as b e-commerce platform) (such as "genuine product guarantee = 1.0"), and filters out goods that do not meet the platform characteristics. If the user's intent explicitly mentions a certain attribute (such as "waterproof"), that attribute will have the highest priority in all layers; if a user intent conflict is detected (such as "low price" and "genuine product guarantee"), a dynamic balance will be achieved through weight distribution (such as "low price weight 0.6, genuine product guarantee weight 0.4").

[0111] By combining user explicit and implicit requirements with cross-platform attribute rules, generating dynamic filtering conditions and using a hierarchical screening mechanism, both the user's unexpressed deep requirements (such as "sports scene" implies "durability") and the semantic differences between platforms (such as "waterproof" → "IP68") are eliminated, ensuring consistency in cross-platform search results; at the same time, the hierarchical screening mechanism filters core attributes first and then sorts implicit requirements, reducing computational resource consumption and preferentially displaying key attributes explicitly mentioned by the user (such as "A"), ultimately achieving efficient matching and accurate sorting, significantly improving user experience and search efficiency.

[0112] In the embodiments of the present application, when updating the matching strategy based on the feedback information, the method further comprises:

[0113] Step 1, first collect user operation behavior data on search results from multiple platforms, including click, purchase, collection, etc. For example, if a user clicks on a certain product but does not purchase it, it may indicate that they are not satisfied with the price or after-sales service; purchase behavior reflects that the product attributes (such as price, function) meet the user's requirements.

[0114] Meanwhile, the system extracts standardized attributes (e.g., "waterproof level = IP68", "price = 1999 yuan", "brand = A") from product data and unifies the expression by combining cross-platform attribute mapping tables (e.g., "waterproof" → "IP68", "high cost performance" → "price / performance ratio"). Subsequently, the association between user behavior and product attributes is analyzed by statistical methods or machine learning models. For example, click behavior may reveal user concerns about "price" or "after-sales service", purchase behavior may indicate that "screen drop resistance" and "battery life" are core driving factors, and collection behavior may reflect user preferences for "brand" or "design". Finally, the system generates dynamic rules based on the analysis results.

[0115] Step 2, according to the generated attribute association enhancement rules, dynamically adjust the semantic mapping and attribute alignment logic. In terms of semantic mapping optimization, update the synonym replacement mechanism to improve the accuracy of attribute expression. For example, if users prefer "IP68" over "waterproof", increase the priority of mapping "waterproof" to "IP68"; if users are more sensitive to "high cost performance", strengthen the mapping weight of "price / performance ratio".

[0116] At the same time, expand semantic associations combined with user behavior, such as users frequently searching for "sports phones" and then purchasing "high battery life + waterproof" products, establishing strong associations between "sports scenarios" and "battery life" and "waterproof". In terms of attribute alignment optimization, verify the consistency of attributes on different platforms. For example, after a user purchases an "IP68 waterproof phone" on platform b, if the same product on platform a only has the "waterproof" attribute, the "IP68" attribute must be added. In addition, adjust attribute weights based on user behavior. For example, if user click rate significantly increases after clicking "price <2000 yuan", increase the weight of the "price" attribute; if users have a high collection rate for "brand", strengthen the priority of brand matching.

[0117] Step 3, the system continuously monitors the impact of the optimized semantic mapping and attribute alignment logic on search results (such as click-through rate, conversion rate, complaint rate). For example, if the "IP68" mapping optimization increases user click-through rate by 10%, the rule is retained; if the "brand" weight adjustment leads to user loss, the change is rolled back. At the same time, update the attribute association enhancement rules regularly based on monitoring results. For example, if it is found that users pay more attention to "environmentally friendly materials", add a new association rule between "environmentally friendly" attributes and "user collection rate", and adjust the semantic mapping table (e.g., "environmentally friendly" → "degradable materials").

[0118] In addition, synchronize the optimized rules to all platforms to ensure consistency of cross-platform attribute expression and user behavior. For example, if the user's "brand" weight is adjusted on platform b, platform a synchronously updates the matching logic of the "brand" attribute.

[0119] By associating user click, purchase, collection, and other operation behaviors with cross-platform attribute characteristics, dynamic attribute association rules are generated, and based on this, the semantic mapping (such as "IP68" instead of "waterproof") and attribute alignment logic are optimized, eliminating the differences in attribute expression between different platforms, ensuring consistent search experience for users on a e-commerce platform, b e-commerce platform, and other platforms. At the same time, through a closed-loop feedback mechanism to continuously monitor and optimize the effect (such as improving click-through rate, adapting to emerging needs such as "environmentally friendly materials"), the matching strategy is dynamically adjusted to avoid the lag of static rules, thereby achieving precise optimization based on user behavior, cross-platform consistency guarantee, and rapid response to market changes, and comprehensively improving the accuracy of search matching and user experience.

[0120] In the embodiments of the present application, when parsing natural query language, the method further comprises:

[0121] Step 1, first extract key attributes (such as "waterproof", "high cost performance", "A") and their multilingual expressions (such as English "waterproof" corresponding to Chinese "waterproof") from user natural language input (such as search words, customer service dialogue). Then, call the cross-platform attribute semantic mapping table to map the fuzzy natural language attribute (such as "waterproof") to the standardized attribute (such as "IP68 level waterproof"), and unify the expression between different languages (such as "waterproof" is mapped to "IP68").

[0122] At the same time, combined with the characteristics of each platform (such as a e-commerce platform prefers "low price" and b e-commerce platform emphasizes "authenticity guarantee"), the attribute weight is dynamically adjusted (such as the priority of "waterproof" on b e-commerce platform is higher than that on a e-commerce platform). Finally, based on the semantic mapping result, a multilingual compatible standardized query template is constructed to ensure that whether the user inputs Chinese "waterproof mobile phone A 2000 yuan below" or English, the system generates the same structure of query template, thereby eliminating the language and platform differences and improving the consistency of cross-platform search.

[0123] Step 2, after generating the standardized query template, further analyze the attribute description habits of each platform (such as a e-commerce platform focusing on "low price", b e-commerce platform emphasizing "authenticity guarantee"), and build platform attribute rule library (such as "waterproof" is marked as "strong waterproof performance" in a e-commerce platform, and "IP68 waterproof" in b e-commerce platform). Through semantic expansion mechanism, the system dynamically converts the standardized template into query features that conform to the habits of the platform. For example, in a e-commerce platform, "IP68 waterproof" is expanded to "strong waterproof performance" and attached with "low price" label, while in b e-commerce platform, it is expanded to "IP68 certification" and emphasizes "authenticity guarantee", while keeping the core attributes (such as "A" "price ≤ 2000 yuan") to ensure the consistency of cross-platform results. In addition, the system adjusts the display priority of query features according to the characteristics of the platform (such as b e-commerce platform preferentially displays "authenticity guarantee") or supplements the implicit attributes (such as a e-commerce platform automatically adds "promotion" label), so as to adapt to the platform preference and improve the relevance of search results and user click rate.

[0124] By mapping the natural query language of user's multi-language input (such as "waterproof" or "waterproof") to standardized attributes (such as "IP68 waterproof"), and dynamically generating adapted query features according to the attribute description habits of each platform, the accurate matching of cross-platform search is realized. Its core value lies in: eliminating language and platform differences (such as "waterproof" → "IP68"), ensuring the consistency of search results under different languages or platforms; dynamically adapting to platform preferences (such as b e-commerce platform preferentially displays "authenticity guarantee"), improving the relevance of search results and user needs; enhancing user search experience, through standardized template and semantic expansion mechanism, users can obtain search results that are both unified and meet the characteristics of different platforms (such as a e-commerce platform highlights "low price", b e-commerce platform emphasizes "certification"), thereby significantly improving click rate and satisfaction.

[0125] The embodiment of the application discloses a cross-platform e-commerce resource dynamic matching search system, referring to Figure 2 , comprising:

[0126] The semantic mapping relationship establishing module 001 analyzes the product description information from different e-commerce platforms in real time at the data access end, extracts the core attributes of the products and generates a unified semantic representation, the product description information includes text and image, and the core attributes include name, specification and price; the semantic representation is based on a synonym replacement mechanism and attribute association rule based on context perception, which is used to establish the semantic mapping relationship between cross-platform attributes, and the standardized data is transmitted to the central database through a distributed consistency synchronization mechanism;

[0127] The conflict resolution strategy generation module 002 generates a conflict resolution strategy based on the standardized data and the historical behavior characteristics of the user when a cross-platform data conflict occurs. The conflict resolution strategy is executed by a local preprocessing module deployed on an edge node. The local preprocessing module responds quickly to burst data, and the local preprocessing module processes periodic data in batches.

[0128] The target edge node selection module 003 obtains the data content after the conflict resolution. In the computing architecture formed by the edge node and the center server, the target edge node is selected based on the data content and the user geographic location information, and the cache data set of the target edge node is dynamically configured.

[0129] The update module 004 obtains a natural query language input by the user, analyzes the corresponding user intent information in the natural query language, generates a multi-dimensional matching feature based on the user intent information and the configured cache, determines a candidate product set based on the multi-dimensional matching feature through approximate nearest neighbor search, and outputs a corresponding search result based on the candidate product set. The feedback information of the user to the search result is obtained, and the intent analysis model and the matching strategy are updated based on the feedback information.

[0130] The embodiments of the present application also disclose a cross-platform e-commerce resource dynamic matching search system, which comprises a processor, and the processor runs a program of the cross-platform e-commerce resource dynamic matching search method of any one of the above.

[0131] The embodiments of the present application also disclose a storage medium, which stores a program of the cross-platform e-commerce resource dynamic matching search method of any one of the above.

[0132] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application. Those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A cross-platform e-commerce resource dynamic matching search method, characterized in that, The method comprises the following steps: Real-time analysis of commodity description information from different e-commerce platforms at the data access end, extraction of core attributes of the commodities and generation of a unified semantic representation, the commodity description information including text and images, the core attributes including name, specification and price; the semantic representation establishes a semantic mapping relationship between cross-platform attributes based on a context-aware synonym replacement mechanism and attribute association rules, and transmits the standardized data to the central database through a distributed consistency synchronization mechanism; Based on the commodity attribute information in the standardized data and the historical behavior characteristics of the user, a conflict resolution strategy is dynamically generated when a cross-platform data conflict occurs, which is executed by a local preprocessing module deployed on the edge node. The local preprocessing module responds quickly to sudden data, and the local preprocessing module processes periodic data in batches. In the conflict resolution strategy, the method further comprises: obtaining the geographical location information of the source platform and the target user of the conflict data, and extracting the association characteristics of the source platform and the region based on the source platform and the geographical location information; based on the association characteristics, a cross-regional data consistency checking rule is dynamically generated, wherein the cross-regional data consistency checking rule preferentially adopts the platform data with the highest matching degree with the user's region as the conflict resolution benchmark; After the conflict is resolved, the data content after the conflict is obtained, and in the computing architecture composed of the edge node and the center server, the target edge node is selected based on the data content combined with the user's geographical location information, and the cache data set of the target edge node is dynamically configured; The natural query language input by the user is obtained, the corresponding user intent information is parsed based on the natural query language, the corresponding multi-dimensional matching features are generated based on the user intent information combined with the commodity data in the configured cache, the candidate commodity set is determined based on the multi-dimensional matching features through approximate nearest neighbor search, and the corresponding search result is output based on the candidate commodity set. The feedback information of the user to the search result is obtained, and the intent parsing model and the matching strategy are updated based on the feedback information; wherein, when generating multi-dimensional matching features, the method further comprises: calling user intent information and cross-platform attribute association rules, generating corresponding dynamic attribute filtering conditions based on the user intent information and the cross-platform attribute association rules; based on the dynamic attribute filtering conditions, the candidate commodity range is gradually narrowed through a hierarchical screening mechanism, and the core attributes explicitly mentioned in the user intent are preferentially matched. 2.The cross-platform e-commerce resource dynamic matching search method according to claim 1, characterized in that, When generating a unified semantic representation, the method further comprises: Associating and mapping the core attributes of the commodities with the preference labels in the user's historical behavior, and extracting the corresponding cross-platform behavior patterns; Based on the cross-platform behavior patterns, the corresponding dynamic attribute weights are generated, the priority of different attributes in the semantic representation is calculated through a dynamic attribute weight generation module, and the priority of different attributes in the semantic representation is dynamically updated through an attribute priority adjustment module. 3.The cross-platform e-commerce resource dynamic matching search method according to claim 1, characterized in that, When dynamically configuring the cache data set of the target edge node, the method further comprises: Extract the cross-platform price fluctuation trend of the commodity and the user purchase periodicity feature, divide the cache update priority of the commodity based on the cross-platform price fluctuation trend and the user purchase periodicity feature; Based on the cache update priority, the short cycle forced refresh strategy is executed for the commodity with frequent price fluctuations, and the long cycle passive update strategy is executed for the commodity with stable price.

4. The cross-platform e-commerce resource dynamic matching search method according to claim 1, characterized in that, When updating the matching strategy based on the feedback information, the method further comprises: Obtaining user operation behavior of the user on the search result, the user operation behavior including clicking, purchasing, and collecting the search result, obtaining cross-platform attribute features corresponding to the commodity, performing association analysis based on the user operation behavior and the cross-platform attribute features, and generating attribute association enhancement rules driven by user behavior; Based on the attribute association enhancement rules, the synonym replacement mechanism and the attribute alignment logic are updated through a semantic mapping optimization mechanism.

5. The cross-platform e-commerce resource dynamic matching search method according to claim 1, characterized in that, When parsing the natural query language, the method further comprises: Obtaining the semantic mapping relationship between cross-platform attributes, generating a multilingual compatible standardized query template based on the semantic mapping relationship between the natural query language and the cross-platform attributes; Based on the standardized query template, the attribute description habits of different platforms are automatically adapted through a semantic expansion mechanism to generate cross-platform compatible query features.

6. A cross-platform e-commerce resource dynamic matching search system for performing the cross-platform e-commerce resource dynamic matching search method according to any one of claims 1-5, characterized in that, It includes: A semantic mapping relationship establishment module, which parses real-time commodity description information from different e-commerce platforms at a data access end, extracts core attributes of the commodity and generates a unified semantic representation, the commodity description information includes text and image, and the core attributes include name, specification, and price; The semantic representation is based on a synonym replacement mechanism and an attribute association rule for context awareness to establish a semantic mapping relationship between cross-platform attributes, and the standardized data is transmitted to a central database through a distributed consistency synchronization mechanism; A conflict resolution strategy generation module, which is based on the commodity attribute information in the standardized data and the historical behavior features of the user to dynamically generate a conflict resolution strategy when a cross-platform data conflict occurs, the conflict resolution strategy is executed by a local preprocessing module deployed at an edge node, wherein the local preprocessing module quickly responds to bursty data, and the local preprocessing module processes periodic data in batches; A target edge node selection module, which obtains data content after conflict resolution, selects a target edge node based on the data content in combination with user geographic location information in a computing architecture composed of edge nodes and a central server, and dynamically configures a cache data set of the target edge node; An update module, which obtains a natural query language input by the user, analyzes corresponding user intent information based on the natural query language, generates corresponding multi-dimensional matching features based on the user intent information in combination with commodity data in the configured cache, determines a candidate commodity set based on the multi-dimensional matching features through approximate nearest neighbor search, and outputs corresponding search results based on the candidate commodity set, obtains feedback information of the user on the search results, and updates an intent analysis model and a matching strategy based on the feedback information.

7. A cross-platform e-commerce resource dynamic matching search system, characterized in that, The application relates to a processor, wherein the processor runs a program of a cross-platform e-commerce resource dynamic matching search method as claimed in any one of claims 1-5.

8. A storage medium, characterized by The application relates to a storage device, wherein the storage device stores a program of a cross-platform e-commerce resource dynamic matching search method as claimed in any one of claims 1-5.

Images