Searching and sorting method oriented to medicine e-commerce platform

Abstract

The invention discloses a medicine e-commerce platform-oriented search sorting method and device and a computer readable storage medium, and the method comprises the steps: responding to a received search request, executing retrieval in a preset index database based on the search request, and obtaining a plurality of candidate data objects; based on a preset classification standard, identifying an attribute category to which the heterogeneous attribute parameter belongs; calling a mapping mechanism pre-associated with the attribute category, and executing mapping calculation on the heterogeneous attribute parameters belonging to the attribute category to obtain a sorting metric value of a unified dimension; according to the sorting metric value corresponding to the heterogeneous attribute parameter, calculating a comprehensive sorting score of the candidate data object; and performing descending sort on the plurality of candidate data objects based on the comprehensive sorting score, generating a search result list and returning the search result list. The method has the advantages that a search sorting result is guaranteed, and meanwhile basic text correlation and multi-dimensional business logic balance are considered.

Description

Technical Field

[0001] This invention relates to the field of Internet data processing technology, and in particular to a search and sorting method, device, and computer-readable storage medium for pharmaceutical e-commerce platforms. Background Technology

[0002] In pharmaceutical e-commerce platforms (covering both B2B and B2C models), search engines are the core entry point for enterprise customers or end users to quickly locate the medicines, medical devices, and consumables they need. With the exponential growth in the number of products on the platform and the increasing complexity of transaction scenarios, users have placed higher demands on the accuracy, timeliness, and commercial value matching of search results.

[0003] In current technologies, mainstream search engine solutions typically employ inverted index-based text matching algorithms (such as TF-IDF or BM25) to calculate the relevance between query keywords and document fields, using this as the default ranking criterion. While this approach can effectively measure the literal matching degree between query terms and documents, it suffers from significant technical limitations when applied to complex vertical domains such as pharmaceutical e-commerce. On the one hand, ranking logic that relies solely on text relevance leads to a disconnect between search results and actual commercial value. The procurement decision for pharmaceutical products is a comprehensive process that considers not only name matching but also key business dimensions such as price competitiveness, sales volume, marketing incentives, and logistics efficiency. Ignoring these dimensions often results in high-quality, cost-effective, or fulfillment-capable products failing to receive the exposure they deserve, thus reducing the platform's overall transaction efficiency.

[0004] On the other hand, although some existing search engine frameworks provide mechanisms for customizable weighted or basic function scores, existing technical solutions are still relatively crude in their processing strategies, usually remaining at the level of simple linear superposition of single or a few factors. Because data in the pharmaceutical e-commerce field has significant multi-source heterogeneous characteristics (e.g., text is discrete semantic matching, prices and sales are continuous numerical distributions, while marketing campaigns are discrete state events), existing technologies lack a systematic mechanism to address the incomparability of these heterogeneous attributes across measurement scales and the differences in business response characteristics.

[0005] This lack of processing capability means that when faced with complex sorting requirements that combine "textual relevance" and "multi-dimensional business logic," the system often struggles to balance one aspect with another, making it difficult to achieve a reasonable balance and integration of various business dimensions while ensuring relevance. Summary of the Invention

[0006] This application provides a search ranking method for pharmaceutical e-commerce platforms, aiming to solve the technical problem in the prior art that the lack of a systematic classification and differentiated mapping mechanism for multi-source heterogeneous business attributes makes it difficult to balance the basic text relevance with multi-dimensional business logic in search ranking results.

[0007] To achieve the above objectives, embodiments of this application provide a search and ranking method for pharmaceutical e-commerce platforms, including: In response to a received search request, a search is performed in a preset index based on the search request to obtain multiple candidate data objects, wherein the candidate data objects have multiple heterogeneous attribute parameters; Based on a preset classification standard, the attribute category to which the heterogeneous attribute parameters belong is identified; Invoke the mapping mechanism pre-associated with the attribute category to perform mapping calculations on the heterogeneous attribute parameters belonging to the attribute category, and obtain a unified dimension ranking metric value; Calculate the comprehensive ranking score of the candidate data object based on the ranking metric value corresponding to the heterogeneous attribute parameters; Based on the comprehensive ranking score, the multiple candidate data objects are sorted in descending order to generate a search results list and return it.

[0008] In one embodiment, the attribute categories include content matching attribute categories, numeric attribute categories, and state event attribute categories; The step of invoking a mapping mechanism pre-associated with the attribute category to perform mapping calculations on heterogeneous attribute parameters belonging to the attribute category includes: For heterogeneous attribute parameters belonging to the content matching attribute category, identify the content relevance characteristics corresponding to the heterogeneous attribute parameters, and match the preset relevance scoring rules based on the content relevance characteristics to obtain the ranking metric value; For heterogeneous attribute parameters belonging to the numerical attribute category, identify the business sensitivity characteristics corresponding to the heterogeneous attribute parameters, and perform mapping based on the business sensitivity characteristics by matching a preset mapping function to obtain the ranking metric value; For heterogeneous attribute parameters belonging to the state event attribute category, identify the discrete business state represented by the heterogeneous attribute parameters, and determine the discrete metric value corresponding to the discrete business state based on the preset state mapping rule, as the sorting metric value.

[0009] In one embodiment, the content matching attribute category includes search keyword category, and the heterogeneous parameters include search keyword parameters; For search keyword parameters belonging to the content matching attribute category, the content relevance characteristics corresponding to the search keyword parameters are identified, and a preset relevance scoring rule is matched based on the content relevance characteristics to obtain the ranking metric, including: Identify the target field type that the search keyword parameters match in the metadata structure of the candidate data object; The system invokes a preset field weight value allocation rule to retrieve a preset weight value associated with the target field type, which is then used as the sorting metric; wherein, The field weighting rules are set based on the field type to determine the strength of the representation of the candidate data object.

[0010] In one embodiment, the numerical attribute category includes a price category, and the heterogeneous attribute parameter includes a price parameter; For price parameters belonging to the numerical attribute category, the business sensitivity characteristic corresponding to the price parameter is identified, and a preset mapping function is matched based on the business sensitivity characteristic to obtain the ranking metric value, including: Call the preset attribute definition configuration table to retrieve the business impact type identifier associated with the price parameter; In response to the business impact type identifier indicating that the price parameter is a cost-related indicator, the business sensitivity characteristic of the price parameter is identified as a negatively correlated competitive advantage characteristic. Based on the aforementioned negative correlation competitive advantage characteristics, a preset negative correlation mapping function is invoked; Based on a preset negative correlation mapping function, the price parameter is mapped to the ranking metric value.

[0011] In one embodiment, mapping the price parameter to the ranking metric value based on a preset negative correlation mapping function includes: Obtain the preset benchmark price value and preset unit score coefficient; The sorting metric is calculated using the following formula: ; In the formula, The ranking metric is calculated based on the price parameters. The benchmark price value, The actual value of the price parameter is given, and k is a preset unit score coefficient.

[0012] In one embodiment, after calculating the... Then, based on a preset negative correlation mapping function, the price parameter is mapped to the ranking metric value, which also includes: Get the preset maximum score threshold; Determine the Is it greater than the maximum score threshold? If so, the maximum score threshold shall be used as the ranking metric. If not, then the above As the sorting metric.

[0013] In one embodiment, the numerical attribute category further includes a sales popularity category, and the heterogeneous attribute parameters include sales data parameters; For sales data parameters belonging to the aforementioned sales popularity category, the business sensitivity characteristics corresponding to the sales data parameters are identified, and a preset mapping function is matched based on the business sensitivity characteristics to obtain the ranking metric value, including: Call the preset attribute definition configuration table to retrieve the business impact type identifier associated with the sales data parameter; In response to the business impact type identifier indicating that the sales data parameter is a cumulative growth indicator, the business sensitivity characteristic of the sales data parameter is identified as a diminishing marginal effect characteristic. Based on the diminishing marginal utility characteristic, a preset nonlinear saturation mapping function is invoked; The sales data parameters are mapped to the sorting metric using the nonlinear saturation mapping function.

[0014] In one embodiment, mapping the sales data parameters to the ranking metric using the nonlinear saturation mapping function includes: Obtain the preset smoothing constant and preset scaling factor; The sorting metric is calculated using the following formula: ; In the formula, To calculate the ranking metric based on the sales parameters, λ represents the actual value of the sales data parameter, b is the base of the logarithmic operation, c is the preset smoothing constant, and λ is the preset scaling factor.

[0015] In one embodiment, the numerical attribute category further includes a logistics distance category, and the heterogeneous attribute parameters include logistics distance parameters; For logistics distance parameters belonging to the aforementioned logistics distance category, the business sensitivity characteristics corresponding to the logistics distance parameters are identified, and a preset mapping function is matched based on the business sensitivity characteristics to obtain the ranking metric value, including: Call the preset attribute definition configuration table to retrieve the business impact type identifier associated with the logistics distance parameter; In response to the business impact type identifier indicating that the logistics distance parameter is a spatial cost indicator, the business sensitivity characteristic of the logistics distance parameter is identified as a nonlinear spatial decay characteristic. The preset reciprocal decay mapping function is invoked based on the aforementioned nonlinear spatial decay characteristics; The reciprocal decay mapping function is used to map the logistics distance parameter to the sorting metric value.

[0016] In one embodiment, mapping the logistics distance parameter to the sorting metric using the reciprocal decay mapping function includes: calculating the sorting metric according to the following formula: ; In the formula, C is the sorting metric calculated based on the logistics distance parameters, where C is a preset distance weight constant. The geographic arc distance between the storage location parameters and the user location parameters of the candidate data object.

[0017] In one embodiment, the state event attribute category includes a marketing campaign category, and the heterogeneous attribute parameters include marketing campaign parameters; For marketing activity parameters belonging to the aforementioned state event attribute category, the discrete business states represented by the heterogeneous attribute parameters are identified, and based on preset state mapping rules, discrete metric values ​​corresponding to the discrete business states are determined as the ranking metric values, including: Parse the marketing campaign parameters and identify the marketing campaign tags contained in the marketing campaign parameters; In response to the identification of the marketing campaign tag, the type of campaign to which the marketing campaign tag belongs is determined; The preset activity incentive rule table is invoked to retrieve the preset fixed incentive score corresponding to the activity type, which is used as the ranking metric.

[0018] In one embodiment, the activity types include flash sale activities, discount activities, and coupon activities; Retrieving a preset fixed incentive score corresponding to the activity type includes: In response to the activity type being a flash sale, a first fixed incentive score is matched; In response to the activity type being a discount activity, a second fixed incentive score is matched; In response to the activity type being a coupon activity, a third fixed incentive score is matched; The first fixed incentive score, the second fixed incentive score, and the third fixed incentive score decrease sequentially.

[0019] In one embodiment, the preset index is built on a search engine architecture that supports function-based scoring queries.

[0020] To achieve the above objectives, this application also proposes a search and ranking device for a pharmaceutical e-commerce platform, including a memory, a processor, and a search and ranking program for a pharmaceutical e-commerce platform stored in the memory and executable on the processor. When the processor executes the search and ranking program for a pharmaceutical e-commerce platform, it implements the search and ranking method for a pharmaceutical e-commerce platform as described in any of the above claims.

[0021] To achieve the above objectives, embodiments of this application also propose a computer-readable storage medium storing a search and ranking program for a pharmaceutical e-commerce platform. When executed by a processor, the search and ranking program for the pharmaceutical e-commerce platform implements the search and ranking method for the pharmaceutical e-commerce platform as described in any of the preceding claims.

[0022] The search and sorting method of the technical solution in this application has the following beneficial effects: 1. Achieving a dynamic balance between the commercial value of search results and text relevance: This application can organically integrate key commercial dimensions such as "price competitiveness," "sales popularity," "logistics timeliness," and "marketing incentives" into the ranking logic while ensuring that the search results do not deviate from the user's query keywords. This makes the ranking results no longer a cold text matching list, but a high-quality product recommendation with both high cost performance and high service quality, effectively solving the problem of low commercial value of search results in the existing technology.

[0023] 2. Enhanced ability to process complex and heterogeneous data: By constructing a "classification-mapping" mechanism, the system can flexibly adapt to the business response characteristics of different types of data (such as negative correlation of prices, marginal decrease of sales volume, etc.), avoiding the ranking distortion caused by single linear weighting, and enabling the ranking model to more accurately simulate and respond to users' real purchasing decision psychology.

[0024] 3. Improved platform transaction efficiency: The final search results list prioritizes products that meet user needs and have high conversion potential, thereby directly shortening the user's decision-making path and improving the platform's overall transaction conversion rate (GMV) and user satisfaction. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0026] Figure 1 This is a module structure diagram of an embodiment of the search and sorting device for a pharmaceutical e-commerce platform according to the present invention; Figure 2 This is a flowchart illustrating an embodiment of the search and ranking method for a pharmaceutical e-commerce platform according to the present invention.

[0027] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0028] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0029] To better understand the above technical solutions, exemplary embodiments of this disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of this disclosure to those skilled in the art.

[0030] It should be noted that any reference signs placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The quantifier "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of "first," "second," and "third," etc., does not indicate any order and can be interpreted as names.

[0031] like Figure 1 As shown, Figure 1 This is a schematic diagram of the hardware operating environment of server 1 (also called a search and sorting device for a pharmaceutical e-commerce platform) involved in the embodiments of the present invention.

[0032] The server in this embodiment of the invention includes devices with display functions such as "Internet of Things devices", smart air conditioners, smart lights, smart power supplies with network connectivity, AR / VR devices with network connectivity, smart speakers, autonomous vehicles, PCs, smartphones, tablets, e-book readers, and portable computers.

[0033] like Figure 1 As shown, the server 1 includes: a memory 11, a processor 12, and a network interface 13.

[0034] The memory 11 includes at least one type of readable storage medium, such as flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of the server 1, such as the hard disk of the server 1. In other embodiments, the memory 11 can also be an external storage device of the server 1, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the server 1.

[0035] Furthermore, the memory 11 may include both internal storage units of the server 1 and external storage devices. The memory 11 can be used not only to store application software and various types of data installed on the server 1, such as the code of the search and ranking program 10 for the pharmaceutical e-commerce platform, but also to temporarily store data that has been output or will be output.

[0036] In some embodiments, processor 12 may be a central processing unit (CPU), controller, microcontroller, microprocessor or other data processing chip, used to run program code stored in memory 11 or process data, such as executing search and sorting program 10 for a pharmaceutical e-commerce platform.

[0037] The network interface 13 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface), which is typically used to establish communication connections between the server 1 and other electronic devices.

[0038] The network can be the Internet, a cloud network, a Wi-Fi network, a Personal Area Network (PAN), a Local Area Network (LAN), and / or a Metropolitan Area Network (MAN). Various devices in the network environment can be configured to connect to the communication network according to various wired and wireless communication protocols. Examples of such wired and wireless communication protocols may include, but are not limited to, at least one of the following: Transmission Control Protocol and Internet Protocol (TCP / IP), User Datagram Protocol (UDP), Hypertext Transfer Protocol (HTTP), File Transfer Protocol (FTP), ZigBee, EDGE, IEEE 802.11, Li-Fi, 802.16, IEEE 802.11s, IEEE 802.11g, multi-hop communication, wireless access points (APs), device-to-device communication, cellular communication protocols, and / or Bluetooth communication protocols, or combinations thereof.

[0039] Optionally, the server may also include a user interface, which may include a display, an input unit such as a keyboard, and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be referred to as a screen or display unit, used to display information processed in server 1 and to display a visual user interface.

[0040] Figure 1 Only server 1, which includes components 11-13 and a search ranking program 10 for a pharmaceutical e-commerce platform, is shown. Those skilled in the art will understand that... Figure 1 The structure shown does not constitute a limitation on server 1 and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0041] In this embodiment, the processor 12 can be used to call the search and sorting program for the pharmaceutical e-commerce platform stored in the memory 11, and perform the following operations: In response to a received search request, a search is performed in a preset index based on the search request to obtain multiple candidate data objects, wherein the candidate data objects have multiple heterogeneous attribute parameters; Based on a preset classification standard, the attribute category to which the heterogeneous attribute parameters belong is identified; Invoke the mapping mechanism pre-associated with the attribute category to perform mapping calculations on the heterogeneous attribute parameters belonging to the attribute category, and obtain a unified dimension ranking metric value; Calculate the comprehensive ranking score of the candidate data object based on the ranking metric value corresponding to the heterogeneous attribute parameters; Based on the comprehensive ranking score, the multiple candidate data objects are sorted in descending order to generate a search results list and return it.

[0042] Based on the hardware architecture of the search ranking device for pharmaceutical e-commerce platforms described above, an embodiment of the search ranking method for pharmaceutical e-commerce platforms of the present invention is proposed. The search ranking method for pharmaceutical e-commerce platforms of the present invention aims to solve the technical problem in the prior art where the lack of a systematic classification and differentiated mapping mechanism for multi-source heterogeneous business attributes makes it difficult to balance the basic text relevance with multi-dimensional business logic in search ranking results.

[0043] Reference Figure 2 , Figure 2 This is an embodiment of the search and ranking method for a pharmaceutical e-commerce platform according to the present invention. The search and ranking method for a pharmaceutical e-commerce platform includes the following steps: S10. In response to the received search request, perform a search in a preset index based on the search request to obtain multiple candidate data objects, wherein the candidate data objects have multiple heterogeneous attribute parameters.

[0044] In this step, the system first receives a search request from the user's terminal (such as a PC browser or mobile app used by a pharmacy purchasing agent). This search request typically encapsulates the user's input search keywords (such as "amoxicillin") and the user's location parameters (such as latitude and longitude coordinates, used for subsequent calculation of logistics distance).

[0045] The pre-defined index is preferably built on a high-performance search engine architecture that supports inverted indexes and function score queries, such as an Elasticsearch cluster. In this index, each product (SKU) on the platform is stored as an independent document, i.e., the data object described in this application.

[0046] To support subsequent multi-dimensional fusion and sorting, each data object is designed in the index as structured data containing multiple heterogeneous attribute parameters. Specifically, these heterogeneous attribute parameters are physically represented by different field types and, in terms of business logic, cover various dimensions of pharmaceutical B2B transactions. As a non-limiting example, the heterogeneous attribute parameters include at least: Text-based parameters include generic names ("Amoxicillin Capsules"), product names, manufacturers ("XX Pharmaceutical"), brands, and store names. These parameters are typically used for keyword matching based on inverted indexes.

[0047] Numerical parameters include: selling price (in yuan), historical sales volume (in boxes / month), inventory quantity, and store rating. These parameters are continuous numerical values ​​that can be used for mathematical calculations.

[0048] Spatial parameters: such as the geographical coordinates of a warehouse (Geo-point). These parameters are used for spatial geometric calculations.

[0049] Tag / status parameters: such as whether to participate in "flash sale", whether to participate in "discount promotion", whether to be "VIP exclusive", etc. These parameters are usually represented as Boolean values ​​or enumerated tags.

[0050] During the retrieval process, the system parses the search keywords in the search request and uses the inverted indexing mechanism of the search engine to quickly filter out all products containing the keyword from massive amounts of product data. For example, when the search keyword is "amoxicillin", the system will recall all product documents that contain the words "amoxicillin" in text fields such as "drug name", "synonyms" or "indications".

[0051] It should be noted that the main task of step S10 is "recall," which involves identifying a set of candidate data objects that are literally related to the search term from the entire database. The list of candidate data objects obtained at this stage is typically only initially ranked based on basic text relevance (such as TF-IDF scores) and has not yet undergone the commercial dimension re-ranking process in subsequent steps of this application. Therefore, there may be cases where products with excessively high prices, excessively distant locations, or extremely low sales but high text matching are ranked highly, which is precisely the problem that this application needs to further address.

[0052] S20. Based on a preset classification standard, identify the attribute category to which the heterogeneous attribute parameters belong.

[0053] After obtaining candidate data objects containing multiple heterogeneous attribute parameters in step S10, in order to use appropriate algorithms to process data with different physical properties, the system first needs to perform a qualitative analysis of these parameters at the logical level.

[0054] Here, the pre-defined classification criteria are preferably represented in the implementation as an attribute definition configuration table or metadata mapping table stored in the system backend, configuration center, or search engine index mapping settings. This classification criterion pre-establishes a logical mapping relationship between the specific field identifiers in the data object and the abstract attribute categories defined internally by the system. This mapping relationship is pre-configured based on business requirements and data characteristics.

[0055] During the identification process, the system iterates through the heterogeneous attribute parameters contained in each candidate data object obtained in step S10. For each heterogeneous attribute parameter, the system extracts its field identifier (Key) and uses this field identifier as an index key to perform a search and match within a preset classification standard. Through this matching process, the system can determine the attribute category label to which the heterogeneous attribute parameter belongs in the current ranking model.

[0056] In step S20, the system transforms the originally messy and isolated data fields in the candidate data object into categorized data with clearly defined processing logic. This process is equivalent to building a routing and distribution layer between the original data and the subsequent scoring algorithm, ensuring that subsequent steps can automatically call the matching algorithm for processing based on the attribute category identified here, without having to hardcode processing logic for each data field separately.

[0057] S30. Invoke the mapping mechanism pre-associated with the attribute category to perform mapping calculations on the heterogeneous attribute parameters belonging to the attribute category, and obtain a unified dimension ranking metric value.

[0058] The core task in this step is to address the incomparability of heterogeneous data in terms of physical dimensions and value ranges. The system pre-maintains a library of mapping mechanisms or a set of algorithmic strategies, establishing a one-to-one or many-to-one correspondence between "attribute categories" and specific "mapping mechanisms." This correspondence ensures that the system can automatically index to the correct processing logic based on the category labels identified in step S20.

[0059] During the mapping calculation, for each heterogeneous attribute parameter, the system calls the corresponding algorithm model or calculation function from the mapping mechanism library according to its attribute category. The system passes the original data of the heterogeneous attribute parameter (such as a specific numerical value, text, or status identifier) ​​as an input parameter to the called mapping mechanism. The mapping mechanism then performs a preset mathematical transformation or logical judgment to convert the original data into a dimensionless numerical result that can be used for ranking weights.

[0060] A unified-dimensional ranking metric refers to the numerical result output after the above mapping calculation. Regardless of the physical nature of the original heterogeneous attribute parameters (e.g., amount, distance, quantity, or text matching degree), after processing by their respective mapping mechanisms, they are all transformed into metrics on the same mathematical plane (e.g., all mapped to a certain "score" or "weight coefficient"). This transformation eliminates the metric barriers between different data, enabling heterogeneous data that were originally impossible to directly compare or calculate to have the prerequisites for mathematical fusion (such as summation) in subsequent steps.

[0061] In existing search ranking technologies, different types of heterogeneous attribute parameters often have completely different physical meanings and units of measurement. For example, "price parameters" are usually measured in currency units (such as "yuan"), and their values ​​may range from tens to thousands; while "logistics distance parameters" are measured in length units (such as "kilometers"); and "sales data parameters" are measured in count units (such as "transactions / units"), and their values ​​may be as high as tens of thousands. If these raw values ​​are directly weighted or compared without the processing in step S30, parameters with larger numerical values ​​(such as sales volume) will naturally dominate the ranking, thereby masking the influence of other key dimensions (such as price or distance), leading to distorted ranking results.

[0062] It is understood that by executing step S30, this application achieves a "dimensionless" or "homogeneous" transformation process. The mapping mechanism acts as an adaptive "adaptor" that, based on the attribute categories identified in step S20, projects or maps various original data with different physical properties onto a preset, unified mathematical plane (i.e., the "unified dimension sorting metric").

[0063] Under this unified dimension, all parameters no longer represent "how much money" or "how many kilometers," but rather uniformly represent the "contribution" or "score" of that dimension to the final ranking objective. This makes previously isolated and incompatible business data mathematically additivity and comparability, thereby solving the technical problem in existing technologies that cannot take into account multi-dimensional business logic due to the lack of a systematic heterogeneous data processing mechanism.

[0064] S40. Calculate the comprehensive ranking score of the candidate data object based on the ranking metric value corresponding to the heterogeneous attribute parameter.

[0065] In this step, the system performs mathematical aggregation on the multiple unified-dimensional ranking metrics output in step S30. Since all heterogeneous attribute parameters (whether representing content matching degree, numerical competitiveness, or state incentive score) have been mapped to values ​​on the same mathematical plane in the previous steps, they have the basis for direct algebraic operations.

[0066] When calculating the overall ranking score, the system obtains each ranking metric corresponding to the current candidate data object. As a specific implementation, the system can sum all the ranking metrics to obtain a scalar value that represents the overall performance of the candidate data object across all business dimensions; this is the overall ranking score.

[0067] Of course, in practical applications, to reflect the different importance of various business dimensions in the platform's operational strategy (for example, the platform may consider the "price" dimension more important than the "distance" dimension), the system can also pre-configure corresponding fusion weight coefficients for each attribute category. In this scenario, the system first calculates the product of each ranking metric and its corresponding fusion weight coefficient, and then sums all the product results to calculate the comprehensive ranking score. The higher the comprehensive ranking score, the more the candidate data object matches the user's search intent and the platform's business recommendation standards.

[0068] Through step S40, the system converges the multi-dimensional discrete evaluation indicators into a single sortable indicator, providing direct numerical basis for the final list generation.

[0069] S50. Based on the comprehensive ranking score, sort the multiple candidate data objects in descending order, generate a search results list, and return it.

[0070] In this step, the system performs a sorting operation on the multiple candidate data objects recalled in step S10 based on the comprehensive sorting score calculated in step S40.

[0071] When performing the sorting, the system uses a descending order. Since the overall ranking score represents a comprehensive score of candidate data objects in terms of text matching accuracy and multi-dimensional commercial value, a higher score indicates that the data object better meets the user's search needs and has higher commercial recommendation value. Therefore, the system places the candidate data object with the highest score at the beginning of the sequence, and objects with lower scores are ranked accordingly.

[0072] After sorting, the system can extract the top N data objects according to preset pagination rules (e.g., displaying 20 data items per page) and generate a search results list. This search results list is usually encapsulated in a standardized data format (such as JSON or XML), which not only contains basic information about the data objects (such as name, price, and image URL), but can also optionally include the object's overall ranking score or score details for each dimension, so that the front-end can display or debug it.

[0073] Finally, the system transmits the generated search results list back to the requesting user's terminal via the network. Upon receiving the list, the user's terminal uses front-end rendering technology to display the sorted products, allowing the user to prioritize high-quality products that offer both accurate matching and high commercial value, thus completing a full search interaction process.

[0074] In summary, steps S10 to S50 work together to construct a complete search ranking processing closed loop from "basic recall" to "multi-dimensional fusion".

[0075] Specifically, step S10 first utilizes an inverted index mechanism to ensure the basic relevance of search results at the "textual semantic" level, ensuring that the recalled candidate data objects literally match the user's query intent. Subsequently, steps S20 and S30, as the core data processing hub, eliminate the measurement barriers in physical dimensions of different business data (text, numerical values, and states) through the classification, identification, and adaptive mapping of heterogeneous attribute parameters, solving the technical challenge of unifying the quantification of multi-source heterogeneous data. Finally, steps S40 and S50 perform mathematical fusion and descending sorting based on unified-dimensional measurement values, making the implicit multi-dimensional business logic explicit in the final ranking result.

[0076] Through the implementation of the above technical solution, this application has achieved the following beneficial technical effects: 1. Achieving a dynamic balance between the commercial value of search results and text relevance: This application can organically integrate key commercial dimensions such as "price competitiveness," "sales popularity," "logistics timeliness," and "marketing incentives" into the ranking logic while ensuring that the search results do not deviate from the user's query keywords. This makes the ranking results no longer a cold text matching list, but a high-quality product recommendation with both high cost performance and high service quality, effectively solving the problem of low commercial value of search results in the existing technology.

[0077] 2. Enhanced ability to process complex and heterogeneous data: By constructing a "classification-mapping" mechanism, the system can flexibly adapt to the business response characteristics of different types of data (such as negative correlation of prices, marginal decrease of sales volume, etc.), avoiding the ranking distortion caused by single linear weighting, and enabling the ranking model to more accurately simulate and respond to users' real purchasing decision psychology.

[0078] 3. Improved platform transaction efficiency: The final search results list prioritizes products that meet user needs and have high conversion potential, thereby directly shortening the user's decision-making path and improving the platform's overall transaction conversion rate (GMV) and user satisfaction.

[0079] In some embodiments of this application, in order to adapt to the complex data characteristics in the pharmaceutical B2B scenario, the system pre-classifies the attribute categories into content matching attribute categories, numerical attribute categories, and status event attribute categories.

[0080] The reason for this limitation is that the business data in the pharmaceutical e-commerce field has significant heterogeneity. Among them, content matching (such as keywords) deals with discrete symbolic semantic relationships, and its core lies in the accuracy of matching; numerical data (such as price and sales volume) deals with continuous quantitative relationships, and its core lies in the sensitivity curve of numerical changes; state events (such as marketing tags) deal with Boolean or enumerated states, and their core lies in the incentive strength of specific conditions.

[0081] Accordingly, the process of performing the mapping calculation described in step S30 specifically includes the following sub-steps S21 to S23: S21. For heterogeneous attribute parameters belonging to the content matching attribute category, identify the content relevance characteristics corresponding to the heterogeneous attribute parameters, and match the preset relevance scoring rules based on the content relevance characteristics to obtain the ranking metric value.

[0082] In this embodiment, the content matching attribute category mainly involves non-numerical text data that carries semantic information, with the core representative being the search keyword parameters input by the user. For such parameters, the system does not simply determine whether the keyword exists in the candidate data object, but must further identify its content relevance characteristics.

[0083] Here, content relevance refers to the hit distribution and semantic matching depth of search keywords within the metadata structure of candidate data objects. In pharmaceutical e-commerce platforms, a candidate data object (such as a drug SKU) typically contains structured multidimensional information fields, such as "generic name," "product name," "brand," "manufacturer," "indication," and "store name." When a user initiates a search, the same keyword (such as "amoxicillin") may appear in different fields mentioned above.

[0084] For example, if a keyword appears in the "Generic Name" field, the "Content Relevance Feature" is characterized as "Exact Hit of Core Semantics"; if the keyword appears in the "Store Name" field, the "Content Relevance Feature" is characterized as "Fuzzy Hit of Related Semantics". Therefore, the essence of identifying content relevance features is to identify the semantic mapping relationship between the user's search intent and the attribute fields of the data object.

[0085] The reason for such fine-grained feature identification of content matching attribute categories is that in the vertical search scenario of pharmaceutical B2B, the strength of user intent represented by "hit" at different positions varies significantly. Traditional full-text search technologies (such as simple TF-IDF) often only focus on word frequency, which may result in a store's product containing "amoxicillin" in its name but actually selling masks, ranking ahead of "genuine amoxicillin capsules," causing "semantic drift" in search results. By identifying "content relevance characteristics," the system can distinguish between "what the user wants to find" (hit product name) and "what is related to what the user is looking for" (hit store name), thereby constructing an "intent layering mechanism" based on field semantic priority. This provides a logical basis for subsequent differentiated assignment based on preset "relevance scoring rules."

[0086] Understandably, by using scoring rules based on content relevance, the system can assign higher ranking metrics to core fields, forcibly locking products that most directly match the user's intent at the top of the results list. Furthermore, the ranking metrics calculated in S21 form the foundation of the overall ranking score. This ensures that subsequent commercial dimensions such as price and sales volume (S22, S23) are optimized based on "qualified text relevance," preventing products with extremely low prices but completely irrelevant to the search terms from occupying the rankings, thus achieving the integration of commercial value while ensuring search accuracy.

[0087] In some embodiments of this application, the content matching attribute category specifically includes the search keyword category, and correspondingly, the heterogeneous attribute parameter specifically includes the search keyword parameter.

[0088] For these parameters, the processes of "identifying content relevance characteristics" and "matching relevance scoring rules" in step S21 are specifically refined as follows: S211. Identify the target field type that the search keyword parameter matches in the metadata structure of the candidate data object.

[0089] Here, "target field type" refers to the specific data field (Field) that the search keyword hits in the inverted index of the data object. In the metadata structure of a pharmaceutical B2B platform, a product object typically contains multiple field types such as "generic name", "product name", "brand", "manufacturer", "approval number", "store name", and "business scope".

[0090] S212. Call the preset field weight value allocation rule to retrieve the preset weight value associated with the target field type, and use it as the sorting metric.

[0091] Here, the rules for assigning field weights are based on the fact that different types of fields have significant differences in their strength of representing the core content of the data object (i.e., the strength of representation), and this difference directly corresponds to the clarity of the user's search intent.

[0092] Specifically, for strong descriptive fields (such as "generic name of drug" or "product ID"), when a user's search terms match these fields, it is highly likely that the user is precisely looking for that specific product. In this case, the field has the strongest descriptive power for the content and should be given extremely high weight to ensure that the product has an absolute advantage in the ranking.

[0093] For weak representation fields (such as "store name" or "product description"), when a search term only hits these fields but not the name, it often means that the product is only somewhat related to the search intent (e.g., "a certain amoxicillin specialty store" sells "masks"), and its representation ability is weak. If this is not distinguished, non-target products will be mixed in with the top search results, interfering with the user's decision-making.

[0094] Therefore, this embodiment constructs a "field semantic priority system" through a preset field weight value allocation rule. This system elevates "search matching" from a flat text co-occurrence to a three-dimensional intent hierarchy, ensuring the semantic accuracy of the ranking results.

[0095] To illustrate the execution process of the above rules more intuitively, the following examples are provided in specific medical search scenarios.

[0096] Suppose the user enters the search keyword "amoxicillin". During the search process, the system will find different candidate data objects that have matched different fields. The system will then perform the following mapping operation based on the pre-configured field weighting rules: Scenario A (Hit a Core Field): If the system identifies that candidate data object A matches "Amoxicillin" in the "Drug Name" field, the system determines that this match belongs to the highest priority strong representation type. According to the rules, the system directly retrieves and assigns an extremely high preset weight value, such as +1,000,000 points. This ensures that a product named "Amoxicillin Capsules" receives an extremely high base ranking score. Similarly, if the "Product ID" field is matched (usually used for precise barcode scanning or input code search), it is also considered the highest priority and assigned, for example, +1,000,000 points.

[0097] Scenario B (Hiting Auxiliary Attribute Fields): If the system identifies that candidate data object B hits "Amoxicillin" in the "Brand" or "Manufacturer" fields (e.g., a brand name is "Amoxicillin" or a manufacturer contains this word), the system determines that the hit belongs to a medium-priority representation type. According to the rules, the system assigns a medium preset weight value, such as +100,000 points.

[0098] Scenario C (Hiting a weakly relevant field): If the system identifies that candidate data object C only matches "Amoxicillin" in the "Store Name" field (e.g., the store name is "KangKang Amoxicillin Specialty Store," but the product sold is "Medical Cotton Swabs"), the system determines that the match belongs to the weak representation type. According to the rules, the system assigns a low preset weight value, such as +50,000 points.

[0099] The above examples demonstrate that even though the store name for product C (cotton swabs) contains "amoxicillin," its ranking metric (50,000) is significantly lower than that of product A (1,000,000). This mechanism effectively implements the business logic of "what is sold comes before who is sold" (i.e., product itself > manufacturer > store), precisely aligning with the "first-come, first-served" business requirement in pharmaceutical procurement.

[0100] It is worth noting that the content matching attribute categories are not limited to the keyword matching categories mentioned above. In some embodiments of this application, they may also include semantic similarity categories and user profile matching categories.

[0101] For semantic similarity categories, the mapping method involves vectorized calculation. Specifically, the system converts the textual description information of the object and the user's search query into high-dimensional vector representations, and directly maps the semantic matching degree to a continuous numerical value within a preset score range as the ranking criterion by calculating the distance between the two vectors (such as cosine similarity).

[0102] For user profile matching categories, the mapping method involves tag-weighted matching. Specifically, the system extracts the user's historical interest tag set and the object's attribute tag set, calculates the tag overlap or weighting coefficient between the two, and generates a metric representing the degree of personalized matching.

[0103] S22. For heterogeneous attribute parameters belonging to the numerical attribute category, identify the business sensitivity characteristics corresponding to the heterogeneous attribute parameters, and perform mapping based on the business sensitivity characteristics by matching a preset mapping function to obtain the ranking metric value.

[0104] In this embodiment, the numerical attribute category mainly covers business indicator data with continuous numerical characteristics that can be quantitatively compared. Unlike the text semantics processed in S21, the core of such parameters is not "whether they match," but rather "the marginal impact of the numerical value on user decisions." For such parameters, the system must first identify their business sensitivity characteristics.

[0105] Here, business sensitivity characteristics refer to the response curve features between changes in the numerical values ​​of heterogeneous attribute parameters and user purchase intentions or commercial value. In the pharmaceutical B2B transaction scenario, different types of numerical parameters often follow completely different business response logics. For example, "drug prices" usually show a negative correlation trend (the cheaper the better), while "monthly sales customer count" usually shows a positive correlation but diminishing marginal returns trend (the higher the sales volume, the better, but the trust gain brought by growth will gradually saturate). Therefore, the essence of identifying business sensitivity characteristics is to identify the stimulus-response pattern of changes in the numerical values ​​of different physical dimensions on user psychological expectations.

[0106] The reason for identifying this characteristic of numerical attribute parameters is that simple linear weighting cannot adapt to the complex distribution of heterogeneous data. Directly normalizing or linearly superimposing all values ​​easily leads to a "numerical dominance" phenomenon. For example, sales volume might range from 0 to 100,000, while price differences might only range from 0 to 100. Without differentiated function mapping based on business characteristics (such as taking the logarithm of sales volume), the influence of sales volume will completely mask the influence of price, resulting in products with "extremely high sales but outrageous prices" dominating the market for extended periods, severely undermining fair competition. By identifying "business-sensitive characteristics," the system can match each type of numerical parameter with a mathematical model that best aligns with its business logic, thereby constructing an "adaptive measurement system" based on business logic.

[0107] In some embodiments of this application, the numerical attribute categories specifically include price categories, sales popularity categories, and logistics distance categories. Correspondingly, the heterogeneous attribute parameters specifically include price parameters, sales data parameters, and logistics distance parameters. For these parameters with different dimensions, step S22 is refined into differentiated processing for different characteristics.

[0108] It is worth noting that the numerical attribute categories are not limited to the price category, sales popularity category, and logistics distance category mentioned above. In some embodiments of this application, they may also include categories such as historical click rate and conversion rate.

[0109] Specifically, for price parameters (such as drug sales prices) in numerical attribute categories, since they are directly related to users' procurement costs, the system employs a linear mapping logic based on the "negative correlation competitive advantage characteristic." The specific processing flow is detailed into steps S221 to S224: S221. Call the preset attribute definition configuration table and retrieve the business impact type identifier associated with the price parameter.

[0110] In the implementation, the system backend maintains an attribute definition configuration table. This table records the metadata definitions of all heterogeneous attribute parameters in the system. When the system encounters a heterogeneous attribute parameter that is a "price parameter" (e.g., a field named "price"), it uses that field name as the key to query the corresponding configuration item in the configuration table. The query result contains a crucial metadata field—the business impact type identifier. This identifier qualitatively describes the direction of the parameter's impact on user value. For example, the identifier TYPE_COST represents a cost-related indicator (lower is better), and the identifier TYPE_BENEFIT represents a revenue-related indicator (higher is better).

[0111] S222. In response to the business impact type identifier indicating that the price parameter is a cost-related indicator, the business sensitivity characteristic of the price parameter is identified as a negatively correlated competitive advantage characteristic.

[0112] In this step, the system parses the identifier obtained in step S221. When the identifier of the price parameter is identified as TYPE_COST, which represents a cost-related indicator, the system's internal logic decision-maker is triggered, determining that this parameter has a negatively correlated competitive advantage characteristic in business logic. Here, "negatively correlated competitive advantage characteristic" means that the value of this parameter is inversely proportional to the competitive advantage (i.e., the ranking score it deserves) of the data object. That is, in the pharmaceutical B2B procurement scenario, the lower the procurement price of the drug, the greater the profit margin for the buyer, and therefore the stronger the competitive advantage of the product.

[0113] S223. Invoke the preset negative correlation mapping function based on the negative correlation competitive advantage characteristic.

[0114] Based on the above characteristic identification results, the system automatically routes and loads a preset negative correlation mapping function from the algorithm library. In this embodiment, in order to ensure that the sensitivity of the scoring system to price changes is uniform and predictable (for example, regardless of the base price, the score gain brought by each 1 yuan reduction should be constant), the system preferably uses a linear interpolation transformation algorithm as the negative correlation mapping function.

[0115] S224. Based on a preset negative correlation mapping function, the price parameter is mapped to the ranking metric value.

[0116] This step is the core calculation process that transforms specific business numerical values ​​into abstract ranking scores. In some preferred embodiments, this process further includes the following sub-steps S2241 and S2242: S2241. Obtain the preset benchmark price value and preset unit score coefficient.

[0117] The system first loads the two core hyperparameters required by the mapping function: Benchmark price ( This is a preset reference anchor point, usually set as the market guidance price, historical high price, or a reasonable "price ceiling" (e.g., 10,000 yuan) for this type of product. It represents the "zero score line," meaning that when the actual price reaches or exceeds this benchmark value, this dimension will no longer receive bonus points (or will receive 0 points).

[0118] Unit score coefficient (k): This is a scaling factor used to control the sensitivity of the weight. It defines the number of ranking points that can be converted from "each unit price advantage (e.g., each 1 yuan cheaper)" (e.g., 100 points / yuan).

[0119] S2242. Calculate the sorting metric value according to the following formula: ; In the formula, The ranking metric is calculated based on the price parameters. The benchmark price value, The actual value of the price parameter is given, and k is a preset unit score coefficient.

[0120] To illustrate the above calculation logic more clearly, let's take the search for "amoxicillin capsules" as an example: Assume the system has a preset benchmark price value. = 50 yuan (meaning that anything over 50 yuan is considered uncompetitive), preset unit score coefficient k=100 (meaning 100 points are added for every 1 yuan cheaper). At this point, the following situations exist: Scenario 1: The actual price of candidate product A = 10 yuan. The system calculates its numerical advantage range as 50 - 10 = 40 yuan.

[0121] The final sorting metric = 40 × 100 = 4000 points.

[0122] Scenario 2: The actual price of candidate product B = 40 yuan. The system calculates that its numerical advantage is only 50 - 40 = 10 yuan.

[0123] The final sorting metric = 10 × 100 = 1000 points.

[0124] The calculations above show that product A, due to its significant price advantage (30 yuan cheaper), gained a ranking advantage of 3,000 points higher than product B in terms of price. This mechanism precisely quantifies the demand for cost control in B2B procurement into ranking weights, ensuring that high-value products receive more exposure.

[0125] Furthermore, in some preferred embodiments, considering the monotonically increasing characteristic of linear functions, if the price of a candidate data object is extremely low (e.g., a promotional product with a price of "001 yuan" or a negative value caused by a system entry error), it may result in the calculation of an extremely large value. This could mask the weighting of other business dimensions (such as text relevance and security). To avoid this, the system sets an upper limit threshold for the price parameter to limit its maximum score. The specific capping constraint operation includes the following steps S2243 to S2246: S2243. Obtain the preset maximum score threshold; S2244, Determine the above Is it greater than the maximum score threshold? S2245. If so, the maximum score threshold shall be used as the ranking metric. S2246. If not, then the aforementioned As the sorting metric.

[0126] Specifically, the system first reads the maximum score threshold for the price dimension from the configuration center. This threshold is a "circuit breaker" set according to the system's overall scoring system. For example, if the system's text relevance score is out of 100 and the sales volume dimension is out of 5000, to prevent the price dimension from "overshadowing" the price dimension, the maximum score threshold can be set to 50000. This means that no matter how cheap a product is, its maximum competitive advantage score gained through the price dimension will not exceed this value.

[0127] Then, the system will use the original sorting metric value calculated in step S2242 ( The score is then compared with the maximum bonus threshold obtained in step S2243: If the comparison result shows the original If the score exceeds the maximum score threshold, the system determines that the score is suspected of "overflow risk" or "over-incentivization" (for example, it may have hit a low-priced, low-quality product with an abnormal price). In this case, the system performs a truncation operation, directly assigning the maximum score threshold to the final ranking metric and discarding the excess.

[0128] If the comparison result shows the original If the price is less than or equal to the maximum bonus threshold, it indicates that the price is within a reasonable competitive range. The system will keep the original calculation result unchanged and use it directly as the final ranking metric.

[0129] Using the aforementioned "amoxicillin" case, let's assume a benchmark price... = 50 yuan, unit score coefficient k=2000 (high sensitivity), and preset maximum score threshold = 50000 points.

[0130] Normal low-price scenario: Product A is priced at 3000 yuan.

[0131] Calculate: (50 – 30) × 2000 = 40000 points.

[0132] Judgment: 40000 < 50000.

[0133] Result: Final score 40,000 points.

[0134] Extremely low price scenario: Product B is a promotional item used to attract customers, priced at only 0.1 yuan.

[0135] Calculate: (50 – 0.1) × 2000 = 99800 points.

[0136] Judgment: 99800 > 50000.

[0137] Result: The capping logic was triggered, and the final score was forcibly adjusted to 50,000 points.

[0138] It is understandable that, through the capping process in steps S2243 to S2246, this application effectively solves the "score dominance" problem that may arise in linear mapping functions under extreme boundary conditions. This ensures that even if some products have an extreme price advantage, their ranking weight is limited to a reasonable range, thereby preserving the ability of other dimensions such as "content matching degree" and "sales popularity" to correct the final ranking results, and preventing "low-priced and low-quality" products from dominating the top of search results for a long time simply because of price factors.

[0139] In this embodiment, for sales data parameters (such as "monthly customer count" and "cumulative sales") in numerical attribute categories, due to their large numerical range and cumulative characteristics, the system adopts a nonlinear mapping logic based on the "diminishing marginal utility characteristic." The specific processing flow is detailed as steps S224 to S227: S224. Call the preset attribute definition configuration table and retrieve the business impact type identifier associated with the sales data parameter.

[0140] Similar to the processing of the aforementioned price parameters, the system first accesses the attribute definition configuration table maintained in the backend. When it encounters a heterogeneous attribute parameter that is "sales data parameter" (e.g., a field named sales_count), the system uses that field name as the index key to retrieve its metadata configuration and obtain the corresponding business impact type identifier.

[0141] S225. In response to the business impact type identifier indicating that the sales data parameter is a cumulative growth indicator, the business sensitivity characteristic of the sales data parameter is identified as a diminishing marginal effect characteristic.

[0142] In this step, when the system recognizes that the identifier represents a cumulative growth indicator (e.g., identified as TYPE_CUMULATIVE_GROWTH), the system determines that the parameter follows the diminishing marginal utility characteristic in business logic.

[0143] Here, the diminishing marginal utility property means that although the increase in the value of this parameter generally represents an increase in the popularity or reputation of the product (positive correlation), as the base value increases, the marginal value gain brought by the increase in the value gradually decreases.

[0144] For example, in pharmaceutical procurement, a product's monthly sales increasing from 0 to 100 signifies a transformation from "unpopular" to "market-recognized," a qualitative change that should garner significant ranking points. However, when monthly sales increase from 10,000 to 10,100, although the absolute increase is the same (both 100), the perceived difference is negligible for buyers as both products have high credibility levels. Therefore, this portion of growth should not receive the same level of ranking points.

[0145] S226. Based on the diminishing marginal effect characteristic, a preset nonlinear saturation mapping function is invoked.

[0146] To adapt to the aforementioned business characteristics, the system automatically routes and loads a preset nonlinear saturation mapping function. Unlike the linear function used when processing price parameters, this function must possess the curve characteristic of "steep curves in the low-value range and gentle curves in the high-value range." In some preferred embodiments, the system employs a logarithmic smoothing transformation algorithm as the nonlinear saturation mapping function.

[0147] S227. Using the nonlinear saturation mapping function, the sales data parameters are mapped to the sorting metric value.

[0148] This step aims to compress the distribution space of long-tailed values ​​through mathematical transformations, preventing high-selling products at the top from monopolizing the ranking results. This process typically includes the following specific computational logic: S2271. Obtain the preset smoothing constant and preset scaling factor.

[0149] Specifically, the system obtains preset smoothing constants and scaling factors. The smoothing constant is typically set to 1 to prevent a mathematical error of log(0) when sales are 0. The scaling factor is used to amplify smaller values ​​after logarithmic operations (e.g., log(100) = 2) to an order of magnitude comparable to other dimensions (e.g., price).

[0150] S2272. Calculate the sorting metric value according to the following formula: ; In the formula, To calculate the ranking metric based on the sales parameters, λ is the actual value of the sales data parameter, b is the base of the logarithmic operation (usually 10 or e), c is the preset smoothing constant, and λ is the preset scaling factor.

[0151] To visually demonstrate how this mechanism suppresses the "over-dominance of high-frequency values ​​on sorting results," the following comparison is provided: Assume a scaling factor λ = 1000, a base b = 10, and a smoothing constant c = 1.

[0152] Scenario C (New Product Launch): The sales volume of a new drug product increases from 0 to 99 (an increase of 99).

[0153] The score when sales are 0 is: log_10(1) × 1000 = 0 points.

[0154] When the sales volume is 99, the sales score is: log_10(100) × 1000 = 2000 points.

[0155] The score jumps to 2000 points, which provides a huge boost to the new product's exposure.

[0156] Scenario D (Top Player): The sales volume of a certain long-established drug increased from 10,000 to 10,099 (an increase of 99).

[0157] The score when sales are 10,000 is: log_10(10001) × 1000 ≈ 4000 points.

[0158] The score when the sales volume is 10099 is: log_10(10100) × 1000 ≈ 4004 points.

[0159] The score increase is only 4 points at this point.

[0160] It's understandable that if linear logic were used, scenario D would also receive a 2000-point bonus, leading to an infinite expansion of established products' scores and their perpetual dominance in search results. However, through the non-linear saturation mapping in step S227, the system successfully compressed the marginal contribution of high-frequency values ​​(10,000 sales) to an extremely low level. This acknowledges their market position while preventing their excessive dominance over the ranking results, thus preserving a fair competitive channel for high-quality small and medium-sized businesses and new products.

[0161] In this embodiment, for logistics distance parameters (such as "warehouse distance" and "delivery mileage") in the numerical attribute category, since they are directly related to the delivery time and logistics cost of medicines, and users' perception of distance has significant spatial nonlinear characteristics, the system adopts a mapping logic based on "nonlinear spatial attenuation characteristics". The specific processing flow is refined into steps S228 to S2211: S228. Call the preset attribute definition configuration table and retrieve the business impact type identifier associated with the logistics distance parameter.

[0162] Similar to the steps described above, the system first accesses the attribute definition configuration table maintained in the background. When it encounters a heterogeneous attribute parameter that is "logistics distance parameter" (e.g., the field name is geo_distance), the system uses that field name as the index key to retrieve its metadata configuration and obtain the corresponding business impact type identifier.

[0163] S229. In response to the business impact type identifier indicating that the logistics distance parameter is a spatial cost index, the business sensitivity characteristic of the logistics distance parameter is identified as a nonlinear spatial decay characteristic.

[0164] In this step, when the system recognizes that the identifier represents a spatial cost-type indicator (e.g., identified as TYPE_SPATIAL_COST), the system determines that the parameter follows non-linear spatial decay characteristics in business logic.

[0165] Here, the nonlinear spatial decay characteristic refers to the fact that users' sensitivity to changes in logistics distance is not uniformly distributed, but rather shows a sharp decrease followed by a gradual leveling off as distance increases. Specifically, logistics distance can be divided into two dimensions: short-distance range and long-distance range. Short-distance range (highly sensitive): Within the same city or surrounding cities (e.g., 0-200 kilometers), a small increase in distance (e.g., an increase of 50 kilometers) may directly cause the delivery time to change from "same-day delivery" to "next-day delivery". This precipitous drop in service experience should lead to a significant decrease in the ranking score.

[0166] Long-distance range (low sensitivity): Within the scope of inter-provincial or long-distance transportation (e.g., over 1000 kilometers), the same increase in distance (e.g., an increase of 50 kilometers, from 1500 kilometers to 1550 kilometers) has almost no impact on the overall delivery time of "3-5 days," and the user's perception of difference is minimal. Therefore, the penalty caused by this increase in distance should be extremely negligible.

[0167] S2210. Based on the nonlinear spatial decay characteristics, call the preset reciprocal decay mapping function.

[0168] To accurately fit the aforementioned business characteristics, the system automatically routes and loads a preset inverse decay mapping function. Unlike the linear difference in price or the logarithmic growth of sales, this function utilizes the "reciprocal of the logarithmic function" to construct a scoring curve that "first drops steeply and then flattens out."

[0169] S2211. Using the reciprocal decay mapping function, the logistics distance parameter is mapped to the sorting metric value.

[0170] This step aims to map physical distance into a ranking score characterizing logistics competitiveness through mathematical transformation. In some preferred embodiments, this process calculates the ranking metric using the following formula: ; In the formula, C is the sorting metric calculated based on the logistics distance parameters, where C is a preset distance weight constant. The geographic arc distance between the storage location parameters and user location parameters of the candidate data object is defined. Here, the distance weight constant (C) is used to adjust the weighting of the logistics dimension in the overall ranking. For example, if it is desired that the logistics factor accounts for approximately 30,000 points out of a perfect score, C can be set to around 30,000 (depending on the specific order of magnitude of the logarithm). The logarithmic operation (log) is preferably performed using a base-10 or base-e logarithm to smooth the distance values.

[0171] To visually demonstrate how this mechanism reflects the "sensitivity difference in the near distance range and the scoring smoothness in the far distance range", the following comparison is made using the emergency medical delivery scenario as an example: Assume the distance weight constant C = 50000, the distance unit is meters, and the commonly used logarithm log_10 is used.

[0172] Scenario E (Same-city express delivery vs. intercity next-day delivery): Warehouse A (Very Close): 1000 meters (1 kilometer) from the user.

[0173] calculate: point.

[0174] Warehouse B (rear distance): 100,000 meters (100 kilometers) away from the user.

[0175] calculate: point.

[0176] Calculating the difference (16666 – 10000) shows that an increase of 99 kilometers in distance results in a sharp drop of 6666 points. This demonstrates the system's extremely high sensitivity to the advantages of short-distance timeliness, prioritizing the recommendation of "instantly available" goods.

[0177] Scenario F (Inter-provincial long-distance freight transfer): Warehouse C (far): 1,500,000 meters (1,500 kilometers) away from the user.

[0178] calculate: point.

[0179] Warehouse D (Further): 2,000,000 meters (2,000 kilometers) away from the user.

[0180] calculate: point.

[0181] Calculating the difference (8095 – 7935) shows that the distance increased by 500 kilometers (5 times the increment of scenario E), but the score only decreased by 160 points.

[0182] Understandably, through the inverse decay mapping in step S2211, the system successfully translates even a slight distance advantage into a significant ranking advantage within a short distance range, ensuring that urgently needed medicines are matched locally. Conversely, within a long distance range, the significant physical distance differences are mathematically smoothed out, causing distance to no longer dominate the ranking, thus giving way to other key dimensions such as price and sales volume. This mechanism perfectly aligns with the complex business logic of the pharmaceutical supply chain: "urgent needs are determined by distance, while inventory is determined by price."

[0183] S23. For heterogeneous attribute parameters belonging to the state event attribute category, identify the discrete business state represented by the heterogeneous attribute parameters, and determine the discrete metric value corresponding to the discrete business state based on the preset state mapping rule, as the sorting metric value.

[0184] In this application embodiment, the status event attribute category covers business tag data that does not have continuous numerical features, nor does it rely solely on semantic matching, but rather exhibits "yes / no", "present / absent", or "type enumeration" features. Its most typical application scenario is marketing activity parameters in pharmaceutical e-commerce platforms (such as whether it participates in flash sales, whether there is a discount for purchases over a certain amount, whether it is VIP exclusive, etc.).

[0185] Unlike the "semantic ambiguity" handled by S21 and the "numerical continuity" handled by S22, the core characteristic of this type of parameter lies in the triggering mechanism of "discrete business status". That is, as long as a product meets a certain specific business condition (such as "registering for the Double 11 promotion"), it should receive a definite, non-continuous, discrete metric value as traffic support or benefit.

[0186] Through the processing in step S23, this application reserves an interface for the platform's "macro-control" at the algorithm level. It allows operators to instantly change the ranking weight of a certain type of activity product without modifying the underlying code by configuring "state mapping rules," thereby achieving dynamic synchronization between the platform's operation strategy and search traffic distribution.

[0187] In some embodiments of this application, the state event attribute category specifically includes a marketing activity category, and the corresponding heterogeneous attribute parameters include marketing activity parameters. For parameters belonging to this category, the system identifies the discrete business states they represent and quantifies them into ranking metrics using preset mapping rules. The specific processing flow is detailed as steps S231 to S233: S231. Parse the marketing campaign parameters and identify the marketing campaign tags contained in the marketing campaign parameters.

[0188] In this step, the system first receives and parses the metadata stream of the product or service object. Since marketing campaigns are typically represented as unstructured text tags or specific status codes, the system needs to parse the marketing campaign parameters. For example, consider the product "running shoes" in an e-commerce scenario. The product's metadata includes a field called `activity_tags`: ["Double11_Seckill", "Free_Shipping"]. By parsing this field, the system identifies the valid marketing campaign tag as "Double11_Seckill".

[0189] S232. In response to recognizing the marketing campaign tag, determine the type of campaign to which the marketing campaign tag belongs.

[0190] Specifically, the system maintains an activity type classifier or mapping table to normalize the diverse tags displayed on the front end into standard backend activity types. Continuing with the above example, the system inputs the identified tag "Double11_Seckill" into the classifier, determining its corresponding standard activity type as "Flash Sale Activity Type." Similarly, if the tag is "Store_Coupon_50," it is determined to be a "Coupon Activity Type."

[0191] S233. Call the preset activity incentive rule table, retrieve the preset fixed incentive score corresponding to the activity type, and use it as the sorting metric.

[0192] In this embodiment, to quantify the differentiated impact of different marketing efforts on users' purchasing decisions, the system pre-defines an activity incentive rule table. This rule table defines the mapping relationship f(Type) → Score between discrete "activity types" and continuous "fixed incentive scores".

[0193] Specifically, depending on the type of activity, the retrieval and matching process includes the following implementation scenarios:

[0194] S2331 (Flash Sale Scenario): In response to the activity type being a flash sale (e.g., limited-time flash sale, hourly flash sale), the system matches a first fixed incentive score (denoted as...). For example, 50,000 points.

[0195] S2332 (Discount Scenario): In response to the activity type being a discount activity (e.g., ¥30 off for every ¥200 spent, cross-store discount), the system matches a second fixed incentive score (denoted as...). For example, 40,000 points.

[0196] S2333 (Coupon Scenario): In response to the activity type being a coupon activity (e.g., store coupon, discount coupon), the system matches a third fixed incentive score (denoted as...). For example, 20,000 points.

[0197] In this application, the first fixed incentive score, the second fixed incentive score, and the third fixed incentive score decrease sequentially, i.e., satisfying... > > This numerical gradient setting is designed to mathematically simulate the intensity of psychological stimulation that different types of activities exert on users.

[0198] For example, suppose there are three product objects in the candidate pool of the sorting list, and their activity states are as follows: Drug A: Participating in the "Flash Sale at the Top of the Hour", the system executes step S2331, matching the first fixed incentive score. =50,000 points.

[0199] Drug B: Participating in "Cross-Store Discount", the system executes step S2332, matching the second fixed incentive score. =40,000 points.

[0200] Drug C: Only "Store Coupon" is available. The system executes step S2333, matching the third fixed incentive score. =20000 points.

[0201] From this moment on, considering only the marketing dimension, the system outputs a ranking metric of 50000 > 40000 > 20000, thus ensuring that flash sale items with high conversion potential receive higher ranking weight.

[0202] It is understood that by employing the aforementioned quantification method based on hierarchical fixed incentive scores, this application can transform originally discrete and incomparable marketing campaign text information into continuous numerical signals that can be processed by computers. Furthermore, by setting... > > The gradient constraint overcomes the shortcomings of traditional ranking models in accurately representing the stronger timeliness and scarcity incentive of "flash sales" compared to "ordinary coupons," and achieves accurate prediction and ranking optimization of user purchase conversion probability in different marketing scenarios.

[0203] It is worth noting that the discrete business status category is not limited to the marketing activity category mentioned above. In some embodiments of this application, it may also include inventory status category and fulfillment service category.

[0204] For inventory status categories, the mapping method adopts a discrete status lookup table method. Specifically, the system identifies the current inventory label of the object (such as "in stock", "pre-sale", "short stock"), and assigns a tiered fixed score to different discrete statuses based on a preset inventory priority mapping table (e.g., 100,000 for in stock, 60,000 for pre-sale, and 80,000 for short stock) to reflect the impact of inventory certainty on user decisions.

[0205] For service fulfillment categories, the mapping method uses Boolean state scoring. Specifically, the system determines whether an object supports specific high-value fulfillment services (such as "next-day delivery," "door-to-door delivery," and "freight insurance"). In response to identifying that an object possesses a discrete service right label, the corresponding preset incentive score is directly matched as the ranking metric.

[0206] Based on the embodiments of steps S21-S23 above, it can be seen that the technical solution of this application further achieves the following beneficial effects: 1. Achieved "logical decoupling" and "specific adaptation" in the heterogeneous data processing architecture: This application, through a pre-defined classification standard, diverts heterogeneous attribute parameters to three independent mapping channels: S21 (content), S22 (numerical), and S23 (state). For each type of data, the most suitable mathematical model (i.e., relevance scoring rules, mapping functions, or state mapping rules) is matched based on the physical properties (continuous / discrete) and business response characteristics (linear / nonlinear / step). This establishes a "divide and conquer" processing architecture, enabling multimodal data to be processed independently within its own mathematical dimension, thus achieving logical decoupling of the feature extraction process. This overcomes the technical shortcomings of traditional ranking methods that often use a "single normalization logic" (i.e., "one ruler for everything") to force the processing of all types of data, leading to loss of textual semantics or distortion of discrete state quantification. Furthermore, it ensures that feature information from different dimensions, when transformed into a unified-dimensional "ranking metric," still retains its original feature expressive power and fidelity to the maximum extent.

[0207] 2. Breakthrough in Feature Representation Bottlenecks in Specific Business Scenarios (Semantic Depth and Incentive Gradient): Under the above architecture, this application further implements differentiated deep processing strategies: on the one hand, "semantic vector / weighted calculation" is introduced for content matching attributes; on the other hand, a "fixed incentive score" with a decreasing relationship is introduced for marketing activity status; thus, firstly, it can break through the superficial limitations of "keyword hard matching" and calculate the relevance between objects and users from the deep semantic space; secondly, it can transform unstructured activity tags into continuous numerical signals that mathematically represent the "intensity of user purchase desire stimulation" and construct a clear incentive gradient; thereby overcoming the dual defects of "long-tail content omission" caused by simply relying on literal matching and "inability to distinguish marketing intensity differences" caused by existing technologies only using "0 / 1 Boolean values" to mark activity status; thus achieving "personalized" ranking results and effective exposure of high-conversion-potential products (such as flash sale items).

[0208] 3. Enhanced the "orthogonality" and "high-quality fusion foundation" of multi-dimensional ranking inputs: Through parallel processing and precise quantization of three independent channels, scores representing content relevance, numerical competitiveness, and business incentive intensity are output respectively. Thus, this application can pre-complete precise cleaning and quantification of each dimension before the final fusion ranking (weighted calculation), ensuring that text relevance, price competitiveness, and activity incentives receive the most accurate numerical mapping within their respective dimensions. This avoids noise interference introduced by rough front-end data preprocessing and solves the "garbage in, garbage out" problem caused by inconsistent input feature quality in traditional models. Furthermore, it provides a high-quality, high-discrimination input foundation for subsequent comprehensive ranking models, significantly improving the final ranking results' comprehensive responsiveness to users' true intentions and business goals.

[0209] Based on the above search keyword parameters, price parameters, sales data parameters, logistics distance parameters, and marketing activity parameters, a ranking metric is obtained. In some embodiments of this application, the ranking metrics corresponding to the keyword parameters, price parameters, sales data parameters, logistics distance parameters, and marketing activity parameters can be summed to obtain the comprehensive ranking score of the candidate data object.

[0210] In other embodiments of this application, to address the mismatch in magnitude between the non-normalized text scores output by the search engine and the business metrics calculated in this application, the system executes steps S100 to S300, employing a "basic relevance decay fusion" strategy to calculate the comprehensive ranking score of the candidate data objects. The specific process and algorithm principle are as follows: Step S100: Obtain the preset original correlation score of the candidate object and obtain the preset decay factor.

[0211] Specifically, the system first calls the underlying text retrieval engine (such as a retrieval engine based on the Lucene kernel), and obtains a preset original relevance score (denoted as ) based on the degree of matching between the user's input search keywords and candidate data objects (such as drug names and indications). This score is usually calculated based on the TF-IDF or BM25 algorithm, and its numerical range is usually large and has no fixed upper bound (e.g., 10.0 to 50.0).

[0212] Simultaneously, the system reads the preset relevance decay factor (denoted as δ) from the configuration center. This factor is a pre-set coefficient used to reduce the weight of the original relevance score, and its value is usually within the range of (0, 1) (e.g., δ = 0.05). In the context of pharmaceutical e-commerce, the setting of this factor needs to be A / B tested to ensure that the weight of business attributes is not excessively diluted while preserving the basic constraints of text relevance.

[0213] Step S200: Calculate the product of the preset original correlation score and the preset correlation decay factor to obtain the basic correlation component.

[0214] Specifically, the system performs multiplication operations, "compressing" the large original correlation components into a numerical space comparable to the business metrics. The calculation formula is: ,in This refers to the calculated basic correlation component.

[0215] Through this step, the system achieves "soft alignment" of heterogeneous scores. For example, it maps text scores that are originally very different (such as 30 points and 10 points) into smaller components (such as 1.5 points and 0.5 points), so that they no longer absolutely dominate the subsequent summation calculation in terms of numerical value.

[0216] Step S300: Sum the basic correlation components with each of the ranking metrics to obtain the comprehensive ranking score.

[0217] Specifically, the system will obtain the result from step S200. The ranking metrics (denoted as) of each dimension (such as marketing, conversion rate, and inventory) calculated in the aforementioned embodiments are compared with those of the above embodiments. Linear superposition is performed. The calculation formula is as follows: .

[0218] in, This is the final overall sorting score used for sorting the list.

[0219] It is worth noting that the calculation method for the comprehensive ranking score is not limited to simple summation. In some other embodiments of this application, in order to reflect the differentiated contributions of different business dimensions to the final ranking result, the system may adopt a "weighted fusion calculation" method.

[0220] Specifically, the system pre-configures a corresponding preset fusion weight coefficient for each of the ranking metrics; during the calculation process, the product of each ranking metric and the corresponding preset fusion weight coefficient is first calculated, and then the resulting product is summed to obtain the comprehensive ranking score of the candidate data object.

[0221] It should be understood that the above weighted fusion logic is universal. It can be applied to implementation scenarios where only the weighted sum of each of the ranking metrics is performed, or to implementation scenarios where the weighted sum of each of the ranking metrics is superimposed on the basic relevance component (i.e., the text score after attenuation processing).

[0222] Furthermore, this invention also proposes a computer-readable storage medium, which can be any one or any combination of several of the following: hard disk, multimedia card, SD card, flash memory card, SMC, read-only memory (ROM), erasable programmable read-only memory (EPROM), portable compact disc read-only memory (CD-ROM), USB memory, etc. The computer-readable storage medium includes a search and ranking program 10 for a pharmaceutical e-commerce platform. The specific implementation of the computer-readable storage medium of this invention is largely the same as the search and ranking method for a pharmaceutical e-commerce platform and the specific implementation of server 1 described above, and will not be repeated here.

[0223] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0224] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0225] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0226] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0227] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0228] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A search and ranking method for pharmaceutical e-commerce platforms, characterized in that, include: In response to a received search request, a search is performed in a preset index based on the search request to obtain multiple candidate data objects, wherein the candidate data objects have multiple heterogeneous attribute parameters; Based on a preset classification standard, the attribute category to which the heterogeneous attribute parameters belong is identified; Invoke the mapping mechanism pre-associated with the attribute category to perform mapping calculations on the heterogeneous attribute parameters belonging to the attribute category, and obtain a unified dimension ranking metric value; Calculate the comprehensive ranking score of the candidate data object based on the ranking metric value corresponding to the heterogeneous attribute parameters; Based on the comprehensive ranking score, the multiple candidate data objects are sorted in descending order to generate a search results list and return it.

2. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 1, characterized in that, The attribute categories include content matching attribute categories, numerical attribute categories, and status event attribute categories; The step of invoking a mapping mechanism pre-associated with the attribute category to perform mapping calculations on heterogeneous attribute parameters belonging to the attribute category includes: For heterogeneous attribute parameters belonging to the content matching attribute category, identify the content relevance characteristics corresponding to the heterogeneous attribute parameters, and match the preset relevance scoring rules based on the content relevance characteristics to obtain the ranking metric value; For heterogeneous attribute parameters belonging to the numerical attribute category, identify the business sensitivity characteristics corresponding to the heterogeneous attribute parameters, and perform mapping based on the business sensitivity characteristics by matching a preset mapping function to obtain the ranking metric value; For heterogeneous attribute parameters belonging to the state event attribute category, identify the discrete business state represented by the heterogeneous attribute parameters, and determine the discrete metric value corresponding to the discrete business state based on the preset state mapping rule, as the sorting metric value.

3. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 2, characterized in that, The content matching attribute categories include search keyword categories, and the heterogeneous parameters include search keyword parameters; For search keyword parameters belonging to the content matching attribute category, the content relevance characteristics corresponding to the search keyword parameters are identified, and a preset relevance scoring rule is matched based on the content relevance characteristics to obtain the ranking metric, including: Identify the target field type that the search keyword parameters match in the metadata structure of the candidate data object; The system invokes a preset field weight value allocation rule to retrieve a preset weight value associated with the target field type, which is then used as the sorting metric; wherein, The field weighting rules are set based on the field type to determine the strength of the representation of the candidate data object.

4. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 2, characterized in that, The numerical attribute category includes the price category, and the heterogeneous attribute parameter includes the price parameter; For price parameters belonging to the numerical attribute category, the business sensitivity characteristic corresponding to the price parameter is identified, and a preset mapping function is matched based on the business sensitivity characteristic to obtain the ranking metric value, including: Call the preset attribute definition configuration table to retrieve the business impact type identifier associated with the price parameter; In response to the business impact type identifier indicating that the price parameter is a cost-related indicator, the business sensitivity characteristic of the price parameter is identified as a negatively correlated competitive advantage characteristic. Based on the aforementioned negative correlation competitive advantage characteristics, a preset negative correlation mapping function is invoked; Based on a preset negative correlation mapping function, the price parameter is mapped to the ranking metric value.

5. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 4, characterized in that, Based on a preset negative correlation mapping function, the price parameter is mapped to the ranking metric value, including: Obtain the preset benchmark price value and preset unit score coefficient; The sorting metric is calculated using the following formula: ； In the formula, The ranking metric is calculated based on the price parameters. The benchmark price value, The actual value of the price parameter is given, and k is a preset unit score coefficient.

6. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 5, characterized in that, The calculation yielded the above. Then, based on a preset negative correlation mapping function, the price parameter is mapped to the ranking metric value, which also includes: Get the preset maximum score threshold; Determine the Is it greater than the maximum score threshold? If so, the maximum score threshold shall be used as the ranking metric. If not, then the above As the sorting metric.

7. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 4, characterized in that, The numerical attribute category also includes the sales popularity category, and the heterogeneous attribute parameters include sales data parameters; For sales data parameters belonging to the aforementioned sales popularity category, the business sensitivity characteristics corresponding to the sales data parameters are identified, and a preset mapping function is matched based on the business sensitivity characteristics to obtain the ranking metric value, including: Call the preset attribute definition configuration table to retrieve the business impact type identifier associated with the sales data parameter; In response to the business impact type identifier indicating that the sales data parameter is a cumulative growth indicator, the business sensitivity characteristic of the sales data parameter is identified as a diminishing marginal effect characteristic. Based on the diminishing marginal utility characteristic, a preset nonlinear saturation mapping function is invoked; The sales data parameters are mapped to the sorting metric using the nonlinear saturation mapping function.

8. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 7, characterized in that, Using the aforementioned nonlinear saturation mapping function, the sales data parameters are mapped to the ranking metric, including: Obtain the preset smoothing constant and preset scaling factor; The sorting metric is calculated using the following formula: ； In the formula, To calculate the ranking metric based on the sales parameters, λ represents the actual value of the sales data parameter, b is the base of the logarithmic operation, c is the preset smoothing constant, and λ is the preset scaling factor.

9. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 7, characterized in that, The numerical attribute category also includes a logistics distance category, and the heterogeneous attribute parameters include logistics distance parameters; For logistics distance parameters belonging to the aforementioned logistics distance category, the business sensitivity characteristics corresponding to the logistics distance parameters are identified, and a preset mapping function is matched based on the business sensitivity characteristics to obtain the ranking metric value, including: Call the preset attribute definition configuration table to retrieve the business impact type identifier associated with the logistics distance parameter; In response to the business impact type identifier indicating that the logistics distance parameter is a spatial cost indicator, the business sensitivity characteristic of the logistics distance parameter is identified as a nonlinear spatial decay characteristic. The preset reciprocal decay mapping function is invoked based on the aforementioned nonlinear spatial decay characteristics; The reciprocal decay mapping function is used to map the logistics distance parameter to the sorting metric value.

10. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 9, characterized in that, Mapping the logistics distance parameter to the sorting metric using the reciprocal decay mapping function includes: calculating the sorting metric according to the following formula: ； In the formula, C is the sorting metric calculated based on the logistics distance parameters, where C is a preset distance weight constant. The geographic arc distance between the storage location parameters and the user location parameters of the candidate data object.

11. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 2, characterized in that, The status event attribute categories include marketing campaign categories, and the heterogeneous attribute parameters include marketing campaign parameters; For marketing activity parameters belonging to the aforementioned state event attribute category, the discrete business states represented by the heterogeneous attribute parameters are identified, and based on preset state mapping rules, discrete metric values ​​corresponding to the discrete business states are determined as the ranking metric values, including: Parse the marketing campaign parameters and identify the marketing campaign tags contained in the marketing campaign parameters; In response to the identification of the marketing campaign tag, the type of campaign to which the marketing campaign tag belongs is determined; The preset activity incentive rule table is invoked to retrieve the preset fixed incentive score corresponding to the activity type, which is used as the ranking metric.

12. The search and ranking method for pharmaceutical e-commerce platforms as described in claim 11, characterized in that, The activity types include flash sale activities, discount activities, and coupon activities; Retrieving a preset fixed incentive score corresponding to the activity type includes: In response to the activity type being a flash sale, a first fixed incentive score is matched; In response to the activity type being a discount activity, a second fixed incentive score is matched; In response to the activity type being a coupon activity, a third fixed incentive score is matched; The first fixed incentive score, the second fixed incentive score, and the third fixed incentive score decrease sequentially.

13. The search and ranking method for a pharmaceutical e-commerce platform as described in any one of claims 1 to 12, characterized in that, The preset index is built on a search engine architecture that supports function-based scoring queries.

14. A search and ranking device for a pharmaceutical e-commerce platform, characterized in that, The system includes a memory, a processor, and a search and ranking program for a pharmaceutical e-commerce platform stored in the memory and executable on the processor. When the processor executes the search and ranking program for the pharmaceutical e-commerce platform, it implements the search and ranking method for a pharmaceutical e-commerce platform as described in any one of claims 1-13.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a search and ranking program for a pharmaceutical e-commerce platform, which, when executed by a processor, implements the search and ranking method for a pharmaceutical e-commerce platform as described in any one of claims 1-13.

Images