Data processing method, medium, service platform, and computer program product

Abstract

A data processing method, a medium, a service platform, and a computer program product. The method is applied to the service platform, and the service platform has established a communication connection with an invocation interface of a large language model in advance. The method comprises: acquiring service data on the service platform and prompt information for input into the large language model, wherein the prompt information is used for prompting the large language model to extract at least one piece of attribute information comprised in the service data (S12); sending the service data and the prompt information to the invocation interface, so that the large language model acquires the service data and the prompt information from the invocation interface, and extracts the at least one piece of attribute information from the service data on the basis of the prompt information (S14); and acquiring the at least one piece of attribute information returned by the large language model by means of the invocation interface, and associating the at least one piece of attribute information with target service objects to which the at least one piece of attribute information respectively belongs (S16).

Description

Data processing method, medium, service platform and computer program product

[0001] This disclosure claims priority to the Chinese patent application filed with the China Patent Office on March 26, 2024, with application number 202410355146.1 and application name “Data processing method, medium, service platform and computer program product”, the entire contents of which are incorporated by reference into this disclosure. Technical Field

[0002] The present disclosure relates to the field of Internet technology, in particular to a data processing method, a medium, a service platform, and a computer program product. Background Art

[0003] The service platform can provide service objects. Operators can define attribute information for the service objects on the service platform. Users can then filter target service objects from the service objects provided by the service platform based on the attribute information. However, manually defining attribute information is labor-intensive and expensive. Summary of the Invention

[0004] In a first aspect, an embodiment of the present disclosure provides a data processing method, which is applied to a service platform, wherein the service platform has pre-established a communication connection with a calling interface of a large language model; the method comprises: obtaining service data on the service platform and prompt information for inputting the large language model; the prompt information is used to prompt the large language model to extract at least one attribute information included in the service data; sending the service data and the prompt information to the calling interface, so that the large language model obtains the service data and the prompt information from the calling interface, and extracts the at least one attribute information from the service data based on the prompt information; obtaining the at least one attribute information returned by the large language model through the calling interface, and performing association processing on the at least one attribute information and the target service object to which the at least one attribute information belongs.

[0005] In some embodiments, the service platform is used to implement service object calls between the service provider of the service platform and the service demander of the service platform, and the service data includes communication records between any service provider of the service platform and at least one service demander of the service platform, and the at least one attribute information includes at least one service supply capability.

[0006] In some embodiments, the service data includes: communication records within a preset time window between a service provider who has chat records with multiple service demanders on the service platform and the multiple service demanders.

[0007] In some embodiments, the large language model is also used to extract descriptive information corresponding to the at least one service supply capability from the communication record, and the descriptive information corresponding to the service supply capability is used to indicate whether the service provider to which the service supply capability belongs has the service supply capability; obtaining the at least one attribute information returned by the large language model through the calling interface, and associating the at least one attribute information and the target service object to which the at least one attribute information belongs, includes: obtaining the at least one service supply capability returned by the large language model through the calling interface, and the descriptive information corresponding to the at least one service supply capability; for any one of the at least one service supply capability, determining whether the service provider to which the service supply capability belongs has the service supply capability based on the descriptive information corresponding to the service supply capability; if so, associating the service supply capability with the service provider to which the service supply capability belongs.

[0008] In some embodiments, the method further includes: for any one of the at least one service supply capability, based on the descriptive information corresponding to the service supply capability extracted from multiple communication records of the same service provider by the large language model, determining the score of the service provider to which the service supply capability belongs for possessing the service supply capability; and determining whether the service provider to which the service supply capability belongs possesses the service supply capability based on the score.

[0009] In some embodiments, the plurality of communication records include at least one of the following: communication records between the same service provider and multiple service demanders; and communication records between the same service provider and at least one service demander within multiple time windows.

[0010] In some embodiments, the service platform is used to implement service object calls between the service provider of the service platform and the service demander of the service platform, the service data includes a set of attributes of the service object called on the service platform, and the at least one attribute information includes target attribute information extracted from the attribute set and ranked in order of importance from high to low, and the target attribute information includes part of the attribute information in the attribute set.

[0011] In some embodiments, the attribute information in the attribute set includes multiple attribute categories; the at least one attribute information extracted from the service data by the large language model based on the prompt information includes: target attribute information extracted from the attribute information of each attribute category in the attribute set, and ranked from high to low in importance.

[0012] In some embodiments, the method further includes: obtaining historical search data of at least one service demander of the service platform on the service platform; the historical search data includes attribute information in the attribute set; according to the frequency of the attribute information included in the historical search data, adjusting the importance ranking of each attribute information determined by the large language model to obtain an adjusted importance ranking; wherein, the at least one attribute information includes target attribute information extracted from the attribute set, and the adjusted importance ranking is from high to low.

[0013] In some embodiments, the ranking of each attribute information determined by the large language model is adjusted according to the frequency of the attribute information included in the historical search data to obtain an adjusted importance ranking, including: determining a first ranking of the attribute information in the attribute set according to the frequency of the attribute information included in the historical search data; obtaining a second ranking of the attribute information in the attribute set through the large language model; and performing weighted processing on the first ranking and the second ranking to obtain the adjusted importance ranking.

[0014] In some embodiments, before associating the at least one piece of attribute information with the target service object to which the at least one piece of attribute information belongs, the method also includes: normalizing the at least one piece of attribute information; and / or calling a reserved verification interface to verify the authenticity of the at least one piece of attribute information.

[0015] In a second aspect, an embodiment of the present disclosure provides a data processing method, which is applied to a service platform, and the method includes: receiving a data request from a client for a target service object on the service platform; obtaining at least one attribute information associated with the target service object in response to the data request; pushing the at least one attribute information to the client so that the client displays the at least one attribute information; wherein, after the at least one attribute information is extracted from the service data on the service platform by a large language model based on prompt information, it is associated with the target service object, and the prompt information is used to prompt the large language model to extract the at least one attribute information included in the service data.

[0016] In some embodiments, the target service object includes at least one service provider of the service platform, and the at least one piece of attribute information includes at least one service supply capability possessed by each service provider; pushing the at least one piece of attribute information to the client so that the client displays the at least one piece of attribute information includes: pushing the at least one service supply capability possessed by each service provider to the client so that the client displays the at least one service supply capability possessed by each service provider.

[0017] In some embodiments, the target service object includes a set of attributes of the service object called on the service platform, the at least one attribute information includes target attribute information extracted from the attribute set and ranked in order of importance from high to low, the target attribute information includes partial attribute information in the attribute set, and the data request is used to obtain target attribute information of at least two service objects; pushing the at least one attribute information to the client so that the client displays the at least one attribute information includes: pushing the target attribute information of the at least two service objects to the client so that the client performs a structured display of the target attribute information of the at least two service objects.

[0018] In some embodiments, the target service object includes a set of attributes of the service object called on the service platform, the at least one attribute information includes target attribute information extracted from the attribute set and ranked in order of importance from high to low, the target attribute information includes part of the attribute information in the attribute set, the data request is used to trigger the dialogue service assistant of the service platform to output dialogue information including the target attribute information, and the dialogue information is used to guide the user to input the target attribute value of the target attribute information; pushing the at least one attribute information to the client so that the client displays the at least one attribute information includes: pushing the target object including the target attribute value to the client so that the client displays the target object including the target attribute value.

[0019] In a third aspect, an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of the present disclosure.

[0020] In a fourth aspect, an embodiment of the present disclosure provides a service platform, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in any embodiment of the present disclosure when executing the program.

[0021] In a fifth aspect, an embodiment of the present disclosure provides a computer program product, including a computer program, which implements the method described in any embodiment of the present disclosure when executed by a processor.

[0022] In the embodiment of the present disclosure, a communication connection is pre-established between the service platform and the calling interface of the large language model. The service platform can send the service data on this platform together with the prompt information to the calling interface of the large language model through the communication connection. The large language model can obtain the service data and prompt information from the calling interface, automatically extract the attribute information from the service data according to the prompt information, and return the attribute information through the calling interface. After obtaining the attribute information extracted by the large language model, the service platform can associate the attribute information with the target service object to which it belongs, thereby mounting the attribute information on the corresponding target service object, so that the user can filter out the target service object from the service platform based on the attribute information associated with the target service object. The embodiment of the present disclosure extracts the attribute information of the service object through the large language model, without the need to manually define the attribute information, thereby reducing the labor cost in the process of obtaining the attribute information.

[0023] It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings herein are incorporated into the specification and constitute a part of the present disclosure. These drawings illustrate embodiments consistent with the present disclosure and, together with the specification, are used to explain the technical solutions of the present disclosure.

[0025] FIG1 is a schematic diagram of an application scenario of an embodiment of the present disclosure.

[0026] FIG2 is a schematic diagram of a system architecture according to an embodiment of the present disclosure.

[0027] FIG3 is a flow chart of a data processing method according to an embodiment of the present disclosure.

[0028] FIG4A is a schematic diagram of a process of extracting service provision capabilities from communication records using a large language model according to an embodiment of the present disclosure.

[0029] FIG4B is a schematic diagram of a process of extracting key attributes from an attribute set by a large language model according to an embodiment of the present disclosure.

[0030] FIG5 is a flowchart of a process of extracting service provision capabilities from communication records according to an embodiment of the present disclosure.

[0031] FIG6 is a schematic diagram of a process of normalizing service provision capabilities using a large language model according to an embodiment of the present disclosure.

[0032] FIG. 7 is a schematic diagram showing a label display of a seller's service provision capabilities according to an embodiment of the present disclosure.

[0033] FIG8 is a flowchart of a process of extracting key attributes from an attribute set according to an embodiment of the present disclosure.

[0034] FIG9 is a schematic diagram of a process of classifying attribute information in an attribute set according to an embodiment of the present disclosure.

[0035] FIG10 is a schematic diagram of a process of sorting attribute information in an attribute set according to an embodiment of the present disclosure.

[0036] FIG11 is a schematic diagram of key attributes of various attribute categories of a pest control product according to an embodiment of the present disclosure.

[0037] FIG12 is a schematic diagram of the sorting results of the full attribute information of the pest control products according to the embodiment of the present disclosure.

[0038] FIG13 is a schematic diagram of a process of comparing key attributes of commodities according to an embodiment of the present disclosure.

[0039] FIG14 is a schematic diagram of a process of guiding user demand convergence through a conversational service assistant according to an embodiment of the present disclosure.

[0040] FIG15 is a flowchart of a data processing method according to another embodiment of the present disclosure.

[0041] FIG16 is a schematic diagram of a service platform according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0042] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

[0043] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. The singular forms "a", "the" and "the" used in this disclosure and the appended claims are also intended to include plural forms, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items. In addition, the term "at least one" herein means any combination of at least two of any one or more of a plurality of.

[0044] It should be understood that although the terms first, second, third, etc. may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining."

[0045] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present disclosure and to make the above-mentioned purposes, features and advantages of the embodiments of the present disclosure more obvious and easy to understand, the technical solutions in the embodiments of the present disclosure are further described in detail below with reference to the accompanying drawings.

[0046] A service platform is a platform that provides one or more service objects. As shown in Figure 1, users can access service platform 20 through client 10 to obtain service objects on service platform 20. Service objects can include various resources (including physical resources and / or digital resources). Physical resources include, but are not limited to, clothing, jewelry, electronic products, and food. Digital resources include, but are not limited to, music, news, and digital goods (such as virtual currency, software applications, e-books, online courses, etc.).

[0047] Based on the supply and demand relationship for service objects, users of the service platform can be divided into service providers and service demanders. Service providers can provide service objects, and service demanders can call the service objects provided by service providers. The service platform can be used to implement service object calls between service providers and service demanders of the service platform. For example, on an e-commerce platform, the service provider is the seller on the e-commerce platform, and the service demander is the buyer on the e-commerce platform. Service object calls are buyers purchasing goods sold by sellers. For another example, on a news platform, the service provider is the news publisher on the news platform, and the service demander is the news reader on the news platform. Service object calls are news readers reading news published by the news publisher. In other types of service platforms, the service provider and service demander can differ depending on the service object. In embodiments where the service platform 20 includes service providers and service demanders, the service demander may need to understand the service provider's service provision capabilities to determine whether the service provider can meet its needs. In this case, the service object can also be the service provider itself.

[0048] Typically, each service object has certain attribute information. Service demanders can use this attribute information to filter out target service objects from the service objects provided by the service platform 20. When the service object is a resource, the resource's attribute information includes, but is not limited to, color, size, function, price, style, and material. For example, if the service platform 20 is an e-commerce platform, and a buyer wants to purchase a piece of clothing from the service platform 20, the clothing becomes the service object. The attribute information of the service object includes the material (e.g., chiffon, cotton, wool, etc.), style (e.g., casual, sporty, ethnic, etc.), and so on. By providing this attribute information, the service platform 20 can filter out clothing of specific materials and styles and return a list of the filtered clothing to the client 10 for display. When the service object is the service provider itself, the resource's attribute information can be the service provider's service provision capabilities, including, but not limited to, OEM capabilities, customization capabilities, and innovation capabilities. The service platform 20 can send the service provider's service provision capabilities to the client 10, allowing the client to display each service provider and its service provision capabilities.

[0049] In related technologies, operators are typically required to manually define attribute information for service objects on the service platform 20. This manual definition of attribute information is labor-intensive. Therefore, the disclosed embodiments utilize a large language model (LLM) to automatically mine the attribute information of service objects and associate the mined attribute information with the target service object to which the attribute information belongs. This automatically attaches the attribute information to the corresponding target service object, reducing manual intervention in the process of obtaining the attribute information of the service object and lowering labor costs.

[0050] Figure 2 shows a schematic diagram of the system architecture of an embodiment of the present disclosure. As shown in Figure 2, the system architecture of an embodiment of the present disclosure includes a service platform 20, a call interface 30, and a large language model 40. The service platform 20 is connected to the large language model 40 via the call interface 30. The service platform 20 can send service data and prompt information on the platform to the large language model 40 via the call interface 30. The large language model 40 can extract attribute information from the service data based on the prompt information and return the extracted attribute information to the service platform 20 via the call interface 30. The service platform 20 can further associate the extracted attribute information with the target service object to which the attribute information belongs. Furthermore, the service platform 20 can also connect to the client 10, and the client 10 can send a data request to the service platform 20 to obtain attribute information associated with the target service object on the service platform 20. In response to the data request sent by the client 10, the service platform 20 can push the attribute information associated with the target service object to the client 10, so that the client 10 can display the received attribute information. The following examples illustrate specific solutions of the embodiment of the present disclosure.

[0051] As shown in FIG3 , an embodiment of the present disclosure provides a data processing method, which is applied to a service platform 20 . The service platform 20 has pre-established a communication connection with a call interface 30 of a large language model 40 . The method includes:

[0052] Step S12: Obtain service data on the service platform 20 and prompt information (prompt) for inputting into the large language model 40; the prompt information is used to prompt the large language model 40 to extract at least one attribute information included in the service data;

[0053] Step S14: sending the service data and prompt information to the calling interface 30, so that the large language model 40 obtains the service data and prompt information from the calling interface 30, and extracts at least one attribute information from the service data based on the prompt information;

[0054] Step S16: obtaining at least one piece of attribute information returned by the large language model 40 by calling the interface 30, and performing association processing on the at least one piece of attribute information and the target service object to which the at least one piece of attribute information belongs.

[0055] In this embodiment, large language model 40 refers to a natural language processing model with a large parameter scale, rich training data, and strong generation and comprehension capabilities. Large language model 40 can be deployed within service platform 20 or on other platforms or systems outside of service platform 20. In embodiments where large language model 40 is deployed within service platform 20, call interface 30 can be a communication interface of service platform 20; in embodiments where large language model 40 is deployed on other platforms or systems, call interface 30 can be a communication interface external to service platform 20.

[0056] In step S12, service data refers to data generated on the service platform 20 to implement the invocation of a service object, including data generated before, during, and after the invocation of the service object. The service platform 20 can obtain and store the service data generated before and after the service invocation. Service data includes, but is not limited to, at least one of the following: communication records between the service provider and the service demander before the service invocation, service invocation data generated by the service invocation, post-processing data after the service invocation, and attribute sets of the invoked service object. For example, if the service platform 20 is an e-commerce platform, the service object can be a product on the e-commerce platform, and thus the service object invocation can refer to a product transaction on the e-commerce platform. The communication records can be communication records between the seller and the buyer before the product transaction, the service invocation data can be order data generated by the product transaction, the post-processing data can be after-sales data after the product transaction, and the attribute sets of the invoked service object can be the attribute sets of each product on the e-commerce platform. A category tree of service objects on the service platform 20 can be pre-established, and attribute sets can be generated for each service object under a leaf category in the category tree. Service objects in different leaf categories can include different attribute sets. In other application scenarios, service data may also include other types of data. For example, on a news platform, the service provider may be the news publisher, and the service demander may be the news reader. Service data may include private messages between the news publisher and the news reader (i.e., chat logs), news interaction information (such as comments and forwarding data), and news attribute information (such as the news topic, type, number of characters, whether it includes images, and publication time).

[0057] Prompt information refers to the input text provided to the large language model 40 to guide the model in generating corresponding responses or derived text. Prompt information typically includes questions, topic descriptions, task instructions, etc., helping the model understand the user's intent and generate a text response that meets the expectations. In the present disclosure, prompt information is used to prompt the large language model 40 to extract at least one attribute information included in the service data. For example, the prompt information can be a sentence similar to the following: "Please output the attribute information included in the following service data."

[0058] The attribute information of a service object is used to describe the characteristics of the service object. Since service data is data generated on the service platform 20 to implement service object invocation, and service invocations are often closely related to the characteristics of the service object, service data often includes content related to the attribute information of the service object, allowing the attribute information of the service object to be extracted from the service data. For example, when the service data includes communication records between the service provider and the service requester prior to the service invocation, the attribute information includes the service provider's service provision capabilities. For another example, when the service data includes an attribute set of the invoked service object, the attribute information includes key attributes within the attribute set. These key attributes are typically the attributes that have a significant impact on the service requester's service invocation. For example, when the service object is a commodity (taking clothing as an example), the attribute set includes the clothing's color, size, style, material, thickness, origin, price, brand, pattern, etc. When a buyer purchases clothing, attributes such as color, thickness, origin, brand, and pattern typically have a smaller impact on the buyer's decision to purchase the clothing, while style, size, material, and price typically have a greater impact on the buyer's decision to purchase the clothing. Therefore, key attributes include style, size, material, and price.

[0059] In step S14, the service platform 20 may send the service data and prompt information to the calling interface 30 via the communication connection between the service platform 20 and the calling interface 30. The service platform 20 may package the service data and prompt information into prompt text and then send the prompt text to the calling interface 30. For example, the prompt text may be a text similar to "Please output the attribute information included in the following service data: data A, data B, data C", where "Please output the attribute information included in the following service data" is the prompt information and "data A, data B, data C" is the service data.

[0060] Alternatively, the service platform 20 may also send the service data and prompt information to the calling interface 30 separately. For example, the service platform 20 first sends the service data to the calling interface 30, and then sends the prompt information to the calling interface 30; or the service platform 20 first sends the prompt information to the calling interface 30, and then sends the service data to the calling interface 30.

[0061] In some embodiments, the service data may include sensitive data, such as mobile phone numbers, email addresses, addresses, names, etc. To protect user privacy, the service data may be desensitized before being sent to the calling interface 30. It should be noted that all user information and data, including the above-mentioned sensitive data, are authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data must comply with the relevant laws, regulations, and standards of the relevant countries and regions, and corresponding operation portals are provided for users to choose to authorize or refuse.

[0062] The large language model 40 can obtain service data and prompt information from the calling interface 30, and extract at least one attribute information from the service data based on the prompt information. As shown in Figure 4A, assuming that the service data is the communication record between the service provider and the service demander on the service platform 20, the service demander may consult the service provider's service supply capabilities. For example, the service demander may ask questions like "Does your store have OEM capabilities?" and "Does the store support customization?". The above communication records can be used as service data and input into the large language model 40 together with the prompt information, so that the large language model 40 extracts the attribute information included in the communication records based on the prompt information. In the embodiment shown in Figure 4A, the service data and prompt information input into the large language model 40 can be as follows:

[0063] Please output the attribute information included in the following communication record:

[0064] {Buyer: Does your store have the ability to OEM?

[0065] Seller: Yes, we have OEM capabilities.

[0066] Buyer: Does the store support customization?

[0067] Seller: Sorry, this store does not support customization.}

[0068] In the above example, the communication records input into the large language model 40 also include the source information (buyer or seller) of each conversation, indicating whether the conversation originated with the buyer or seller. In other examples, each conversation may also include time information to indicate when the conversation occurred. In addition, communication records may also include address information, contact information, and other information, which are not listed here.

[0069] The large language model 40 can determine based on the prompt information that the current task is to extract attribute information, and the object of extraction is the chat record, so that the two service supply capabilities of "private label" and "customization" can be analyzed based on the context information of the chat record.

[0070] In an embodiment where the service target is a communication record between any service provider of the service platform 20 and at least one service demander of the service platform 20, the at least one piece of attribute information includes at least one service provision capability of the service provider of the service platform 20. Due to the large volume of communication records, communication records within a preset time window (e.g., 7 days, 15 days, etc.) can be selected. Furthermore, to increase confidence in the determination of whether a service provider has the service provision capability, communication records between a service provider with existing chat records with multiple service demanders and these multiple service demanders can be selected. For example, suppose the service demanders include sellers X1, X2, and X3. Seller X1 has communication records with only one buyer, buyer X2 has communication records with two buyers (referred to as buyer Y1 and buyer Y2), and seller X3 has communication records with three buyers (referred to as buyer Y3, buyer Y4, and buyer Y5). The communication records between seller X2 and buyer Y1 and seller X2 and buyer Y2 can be selected. Alternatively, the communication records between seller X3 and buyer Y3, seller X3 and buyer Y4, and seller X3 and buyer Y5 can also be selected. Since seller X1 only has communication records with one buyer, in order to reduce the impact of factors such as unreliable data, the communication records between seller X1 and the buyer can be filtered out, thereby increasing the confidence level of the final judgment result.

[0071] In some embodiments, the communication records can be filtered by considering both the preset time window and the number of buyers communicating with the seller. Specifically, the filtered communication records are communication records within the preset time window between a service provider with whom multiple service demanders have chat records on the service platform 20 and the multiple service demanders.

[0072] Specifically, the communication records between each service provider and service demander on the service platform 20 within the preset time window can be first screened out, and then the communication records corresponding to the service provider with which multiple service demanders have communication records can be screened out from the above communication records. Furthermore, the number of service demanders with which the service provider has communication records needs to be greater than or equal to a preset number threshold (for example, 3, or 5), thereby further improving the confidence level. Assuming that the preset number threshold is 3, in the above example, the communication records corresponding to seller X3 can be screened out, that is, the communication records between seller X3 and buyer Y3, the communication records between seller X3 and buyer Y4, and the communication records between seller X3 and buyer Y5.

[0073] In some embodiments, the large language model 40 is also used to extract descriptive information corresponding to at least one service supply capability from the communication record, and the descriptive information corresponding to the service supply capability is used to indicate whether the service provider to which the service supply capability belongs has the service supply capability. The descriptive information can be a binary number, for example, the binary number "0" indicates that the corresponding service supply capability is not available, and the binary number "1" indicates that the corresponding service supply capability is available. Alternatively, the descriptive information can also be other characters, such as the character "yes" indicates that the corresponding service supply capability is available, and the character "no" indicates that the corresponding service supply capability is not available. Continuing with Figure 4A, the large language model 40 can parse from the communication record that the service provider has the OEM capability, but does not have the customization capability. Therefore, the large language model 40 can output descriptive information indicating that the service provider has the OEM capability, and output descriptive information indicating that the service provider does not have the customization capability. For ease of viewing, the large language model 40 can output structured text (such as a table) based on the extracted service supply capabilities and their descriptive information. Continuing with the embodiment shown in Figure 4A, the structured text output by the large language model 40 can be in the following form:

[0074] As shown in FIG4B , assuming that the service data is a set of attributes of a service object invoked on the service platform 20, the at least one attribute information extracted from the service data includes target attribute information extracted from the attribute set of the service object, ranked in descending order of importance. The target attribute information includes a portion of the attribute information in the attribute set. In some embodiments, the number of target attribute information can be pre-set. For example, if the number of target attribute information is set to 3, the top three attribute information in the attribute set in terms of importance can be determined as the target attribute information.

[0075] In the related art, operators are usually required to manually define the service supply capabilities of the service providers on the service platform 20. This method has high labor costs and the categories of service supply capabilities that can be defined are relatively few, resulting in the inability to explicitly reflect the service supply capabilities of the service providers in all aspects on the link between the service providers and the service demanders. Some service demanders may lose service from the service platform 20 before reaching the communication stage with the service demanders because they do not understand whether the service providers meet their service supply capability needs. This solution uses a large language model 40 to automatically extract the service supply capabilities of the service provider from the communication records between the service provider and the service demander. On the one hand, it reduces human participation and reduces labor costs; on the other hand, the service supply capabilities extracted from the communication records by the large language model 40 are relatively comprehensive, thereby being able to explicitly reflect the service supply capabilities of the service provider in all aspects. The disclosed embodiment realizes the automated mining and definition of the service supply capabilities of the service provider, as well as the corresponding mounting of service supply capabilities and service providers, which can effectively assist the service demanders in making service object call decisions on the service platform 20.

[0076] In some embodiments, a set of attributes for the service objects of each leaf category can be pre-established on the service platform 20. For example, the service objects of the leaf category may include but are not limited to clothes, skirts, mobile phones, cameras, rice, pesticides, tables, etc., and a corresponding set of attributes can be established for each of the service objects listed above. For example, the attribute set corresponding to clothes can be {color, size, style, material, thickness, origin, price, brand, pattern}, and the attribute set corresponding to mobile phones can be {size, thickness, operating system, battery capacity, pixels, brand, price, color}. Taking clothes as an example, the service data and prompt information input into the large language model 40 can be as follows:

[0077] Please sort the attribute information in the following attribute set according to importance:

[0078] {color, size, style, material, thickness, origin, price, brand, pattern}.

[0079] The prompt "Please sort the attributes in the following attribute set by importance" is provided, and the service data "color, size, style, material, thickness, origin, price, brand, pattern" is provided. The attributes output by the large language model 40, ranked from most important to least important, may be: material, size, thickness, style, price, brand, pattern, color, and origin. Assuming the top three attributes are identified as key attributes, these key attributes include material, size, and thickness.

[0080] Furthermore, the attribute information in the attribute set may include multiple attribute categories. Based on the prompt information, the large language model 40 can extract target attribute information from the attribute information of each attribute category in the attribute set, sorted from highest to lowest importance. For example, the attribute categories of the attribute information may include, but are not limited to, basic attributes, marketing attributes, technical attributes, logistics attributes, and functional attributes. Basic attributes may include color, size, weight, material, style, pattern, thickness, etc. Marketing attributes may include price, discount information, and discount information. Technical attributes may include technical specifications, performance parameters, and hardware configuration. Logistics attributes may include information such as transportation method, logistics company, and shipping location. Functional attributes may include information such as waterproof, dustproof, warm, and cold-proof. The prompt information can be used to instruct the large language model 40 to extract key attributes from the attribute information of a specified attribute category in the attribute set. Still using clothing as an example, the prompt information may be in the form of: "Please rank the basic attributes of the clothing in terms of importance" or "Please rank the basic attributes and marketing attributes of the clothing in terms of importance." In these examples, the prompt information includes the attribute categories of the attribute information. In other examples, the prompt information may include both the attribute categories of the attribute information and the attribute information itself. The prompt information may be in the following format: "Please rank the importance of the following basic attributes of the clothing: color, size, material, style, pattern, thickness." In different application scenarios, the attribute categories of the attribute information that the service requester is concerned about may vary. By categorizing the attribute information and ranking the attribute information of each attribute category in terms of importance, it can adapt to different application scenarios.

[0081] On the service platform 20, some service demanders do not have clear requirements when calling service objects, and these service demanders often use relatively broad search keywords to search for service objects. For example, in an e-commerce scenario, buyers may use the search keyword "women's clothing" to search for products. Because the search keyword "women's clothing" is too broad, it is difficult for the service platform 20 to clearly identify the buyer's needs. The embodiment of the present disclosure uses a large language model 40 to extract key attributes that affect the decision-making of the service demander, so that when the service demander calls the service object, it can guide the service demander to converge on its needs based on these key attributes, thereby more accurately obtaining the service demander's needs.

[0082] It can be understood that the method of sorting the importance of attribute information of different attribute categories is only one implementation method of the present disclosure. In other implementation methods, the attribute information of all attribute categories of service objects under a certain leaf category can also be uniformly sorted by importance without distinguishing the attribute categories of the attribute information.

[0083] It will be understood that the above embodiments are merely illustrative. In actual applications, the prompt information input to the large language model 40 may differ from that listed in the above embodiments, as long as the prompt information enables the large language model to extract attribute information from the service data. The service object and its service data may be other types of service objects and service data than those in the above embodiments.

[0084] In some embodiments, the input content of the large language model 40 is limited to a certain number of characters, while the service data may contain a large number of characters. Therefore, the service data can be divided into multiple data blocks, and prompt information and a portion of the data block are input to the large language model 40 each time, and the attribute information extracted from this portion of the data block is obtained by the large language model 40. By merging the data information extracted from each data block, the complete attribute information extracted from the service data can be obtained. For example, the service data may include all communication records between service providers and service demanders on the service platform 20. Because the content of these communication records is too large to be input into the large language model 40 at once, the communication records between a single service provider and service demander can be treated as a data block and input into the large language model one data block at a time.

[0085] In step S16, the service platform 20 can obtain at least one piece of attribute information returned by the large language model 40 through the calling interface 30 through the communication connection with the calling interface 30. The service platform 20 can also determine the target service object to which the at least one piece of attribute information belongs, and associate the at least one piece of attribute information with the target service object to which the at least one piece of attribute information belongs. For example, in the case where the service data is a communication record, the attribute information extracted by the large language model 40 is the service supply capability extracted from the communication record, and the target service object to which the attribute information extracted from the communication record belongs can be the service provider that generated the communication record. The communication record can include the identification information of the service provider, so that the service provider that generated the communication record can be determined based on the identification information, thereby determining the target service object to which the attribute information extracted from the communication record belongs. The above-mentioned service supply capability can be associated with the service provider. For another example, when the service data is an attribute set, the attribute information extracted by the large language model 40 is the key attribute extracted from the attribute set. A category tree of service objects can be pre-established on the service platform 20. Each category of service object corresponds to a node in the category tree. A correspondence between the attribute information in the attribute set and the nodes in the category tree can be established. Based on the above correspondence, the node corresponding to the key attribute is determined, and the service object of the category corresponding to the node is determined as the target service object to which the key attribute belongs, and the above key attribute is associated with the service object.

[0086] In some embodiments, the large language model 40 can return at least one service supply capability and the descriptive information corresponding to the at least one service supply capability by calling the interface 30. On this basis, for any service supply capability extracted by the large language model 40, the service platform 20 can determine whether the service provider to which the service supply capability belongs has the service supply capability based on the descriptive information corresponding to the service supply capability. If so, the service supply capability and the service provider to which the service supply capability belongs are associated. For example, assuming that the service supply capability extracted by the large language model 40 includes OEM capability and customization capability, and the descriptive information corresponding to the OEM capability is "yes", and the descriptive information corresponding to the customization capability is "no", indicating that the service provider can provide OEM capability but cannot provide customization capability, the service platform 20 can associate the OEM capability with the service provider, but not associate the customization capability with the service provider.

[0087] As described in the aforementioned embodiment, due to the input character limit of the large language model 40, the service data is divided into multiple data blocks and input into the large language model 40. However, the large language model 40 cannot achieve 100% accuracy in recognizing attribute information. Therefore, the attribute information extracted by the large language model 40 for the same service object from different data blocks may be inconsistent. For example, assuming that the service object is a communication record between a service provider and a service demander, and the attribute information is the service provider's service provision capability, the following situation may occur: the large language model 40 extracts that the service provider has service provision capability P from one communication record, but extracts that the service provider does not have service provision capability P from another communication record. To improve the accuracy and reliability of the attribute information extraction results, for any service provision capability, the service platform 20 can determine the score of the service provider to which the service provision capability belongs based on the descriptive information corresponding to the service provision capability extracted by the large language model 40 from multiple communication records of the same service provider, and then determine whether the service provider to which the service provision capability belongs has the service provision capability based on the score.

[0088] The plurality of communication records may include at least one of the following:

[0089] (1) Communication records between the same service provider and multiple service demanders. For example, assuming that service provider X1 communicates with multiple service demanders (assuming that they include service demanders Y1, Y2, and Y3), the communication records between service provider X1 and service demander Y1, the communication records between service provider X1 and service demander Y2, and the communication records between service provider X1 and service demander Y3 can be obtained respectively.

[0090] (2) Communication records between the same service provider and at least one service demander within multiple time windows. Assuming the time window is 7 days, the communication records between service provider X1 and one or more service demanders in the last 7 days, the communication records between service provider X1 and one or more service demanders in the last 7 to 14 days, and the communication records between service provider X1 and one or more service demanders in the last 14 to 21 days can be obtained.

[0091] By obtaining multiple communication records and determining the score of the service provider's service provision capability based on the descriptive information extracted from the multiple communication records, it is possible to determine whether the service provider has the service provision capability. This can effectively reduce the errors caused by inaccurate recognition results of large language models and improve the accuracy and reliability of the service provision capability extraction results.

[0092] For example, assuming there are five communication records, recorded as record 1, record 2, record 3, record 4 and record 5, and the large language model 40 extracts from record 1 and record 2 that the service provider X1 does not have the OEM capability, and extracts from record 3, record 4 and record 5 that the service provider X1 has the OEM capability. Then, based on the above five communication records, the score of the service provider X1's OEM capability can be determined, and based on the above score, it can be determined whether the service provider X1 has the OEM capability.

[0093] In an optional embodiment, for any service provision capability P, if a communication record determines that the service provider possesses the service provision capability P, a base score (e.g., 1 point) can be added to the original score (e.g., 0 points). If a communication record determines that the service provider does not possess the service provision capability P, a base score can be subtracted from the original score. A determination can be made as to whether the final score is greater than the original score. If so, the service provider is determined to possess the service provision capability P; otherwise, the service provider is determined to not possess the service provision capability P. For example, in the above example, assuming the original score is 0 and the base score is 1, the score for service provider X1's OEM capability can be subtracted by 1 point based on records 1 and 2, respectively. Based on records 3, 4, and 5, the score for service provider X1's OEM capability can be increased by 1 point, resulting in a final score of 1. Since the final score is greater than the original score, it can be determined that service provider X1 possesses OEM capability.

[0094] It will be understood that the above embodiments are merely illustrative. In other examples, other methods may be used to determine whether a service provider has the capability to provide services, which will not be listed here one by one.

[0095] Since the attribute information extracted by the large language model 40 is relatively objective, there may be certain deviations from the actual situation of the service providers and service demanders of the service platform 20. In order to solve this problem, the output results of the large language model 40 can be adjusted based on the historical search data of at least one service demander of the service platform 20 on the service platform 20.

[0096] Specifically, historical search data of at least one service demander on the service platform 20 can be obtained. This historical search data includes attribute information in an attribute set of service objects on the service platform 20. Then, the importance ranking of each attribute information determined by the large language model 40 can be adjusted based on the frequency of the attribute information included in the historical search data to obtain an adjusted importance ranking. Target attribute information extracted from the attribute set and ranked from high to low in the adjusted importance ranking can then be associated with the target service object.

[0097] For example, in an e-commerce platform, when a buyer searches for a product, he or she may enter the attribute information of the product in the search information. For example, when a buyer searches for clothes, he or she may enter the material (such as "chiffon") or style (such as "casual") of the clothes. The higher the frequency with which a buyer enters a certain attribute information, the greater the influence of the attribute information on the buyer's decision, and thus the more important the attribute information is. Taking these important attribute information as key attributes is in line with the buyer's operating habits and psychological expectations. Therefore, by adjusting the importance ranking of each attribute information determined by the large language model 40 based on the frequency of the attribute information included in the historical search data, a ranking result that is more in line with the operating habits and psychological expectations of the users of the service platform 20 can be obtained. At the same time, since in some cases the historical search data may be relatively sparse, combining the historical search data with the large language model 40 to obtain the importance ranking result of the attribute information can reduce the problem of insufficient accuracy caused by the sparsity of the historical search data.

[0098] In some embodiments, the first ranking of the attribute information in the attribute set can be determined based on the frequency of the attribute information included in the historical search data, the second ranking of the attribute information in the attribute set can be obtained through the large language model 40, and the first ranking and the second ranking can be weighted to obtain an adjusted importance ranking. Among them, the higher the frequency of the attribute information included in the historical search data, the higher the attribute information is in the first ranking; conversely, the lower the frequency of the attribute information included in the historical search data, the lower the attribute information is in the first ranking. For a certain piece of attribute information P, assuming that the first ranking of the attribute information P is k1 and the second ranking is k2, the first ranking k1 and the second ranking k2 can be weighted to obtain an adjusted importance ranking of K = α*k1+β*k2. Among them, K is the adjusted importance ranking, and α and β are preset weights.

[0099] As described in the aforementioned embodiment, due to the limitation on the number of input characters of the large language model 40, the service data is divided into multiple data blocks and input into the large language model 40. Different data blocks may be communication records between a service provider and different service demanders. Since different service demanders may have different expression habits, the description of the same service supply capability may be different in different communication records. For example, in one communication record, the service supply capability for personalized customization of products is referred to as "customization capability", while in another communication record, the same service supply capability is referred to as "commodity customization". Based on this, before associating at least one piece of attribute information with the target service object to which the at least one piece of attribute information belongs, the at least one piece of attribute information may also be normalized. For example, "customization capability" and "commodity customization" are normalized into "customization capability".

[0100] In some embodiments, before associating at least one piece of attribute information with the target service object to which the at least one piece of attribute information belongs, a reserved verification interface may be called to verify the authenticity of the at least one piece of attribute information. Still taking the example that the service data is a communication record and the attribute information is a service supply capability, some service supply capabilities may need to provide certificate authentication. Therefore, a reserved verification interface may be called, and the verification interface may call an authenticity verification algorithm, for example, obtaining a certificate pre-uploaded by the service provider, and performing authenticity verification on the service supply capability of the service provider based on the certificate. In addition to the situations listed above, other methods may be used to verify the authenticity of the attribute information, and the specific authenticity verification method is not limited in this disclosure.

[0101] FIG5 illustrates the overall process of the embodiment of the present disclosure using the specific application scenario of extracting the service supply capabilities of the service provider from the communication records as an example. Referring to FIG5 , the service provider is a seller on the e-commerce platform, and the service demander is a buyer on the e-commerce platform. The specific process of extracting the service supply capabilities of the service provider from the communication records includes:

[0102] Step S22: Preprocess the communication records. Specifically, sellers with communication records with at least n buyers within a recent preset time window (assuming 7 days) can be selected, and the communication records between the seller and each buyer can be obtained. Here, n is a positive integer greater than 1, and its value can be pre-set, for example, 5. Furthermore, sensitive information such as telephone numbers and mobile phone numbers can be removed from the obtained communication records.

[0103] Step S24: The pre-processed communication record and prompt information from step S22 are sent together as prompt text to the large language model 40. The prompt information can be used to express the requirement to extract the seller's capabilities (i.e., service provision capabilities). For example, the prompt information can be "Please extract the seller's capabilities from the following text." The large language model 40 extracts the capabilities from the communication record and outputs corresponding description information to indicate whether the seller possesses the capabilities.

[0104] Step S26: Normalize the capability items extracted from each communication record to unify the expression of the same capability item. Normalization can also be achieved through the large language model 40. Referring to Figure 6, the extracted capability items and prompt information can be input into the large language model 40. The prompt information prompts the large language model 40 to normalize the input capability items. For example, the extracted capability items include: sampling, sample, free sample, customization, product customization, and full customization. The prompt information may include: "Which of these capability items express the same meaning?" Furthermore, the prompt information may also include the display method of the normalized capability items, for example, through structured text such as a table, and specify the meaning of each column or row in the table. Specifically, the prompt information may include: "Which of these capability items express the same meaning? Output as a table. The first column of the table contains the capability item, outputting the specific capability. The second column of the table contains synonyms, outputting other expressions with the same meaning as the capability item." The output of the large language model 40 is shown in Figure 6.

[0105] Step S28: The normalized capability items are input into the operation tool, allowing the operator to determine whether the capability items output in step S26 require authenticity verification. Alternatively, a whitelist of capability items can be established. If the capability item output in step S26 matches one on the whitelist, the capability item is determined to require authenticity verification; otherwise, the capability item is determined to not require authenticity verification. If the result is yes, proceed to step S30; otherwise, proceed to step S32.

[0106] Step S30: Call the verification interface to perform authenticity verification on the capability items that need to be verified. For example, determine whether the seller has uploaded a certificate for authenticity verification of the capability items on the service platform 20. If so, determine that the capability item is a real capability item and the verification is successful; otherwise, determine that the verification has failed.

[0107] Step S32: The capability items and their description information extracted by the large language model 40 from multiple communication records are integrated to give a comprehensive score on whether the seller has the capability item, thereby reducing errors caused by insufficient output accuracy of the large language model 40.

[0108] Step S34: The seller's capabilities are displayed in the foreground scene to assist the buyer in making decisions.

[0109] Figure 7 shows a schematic diagram of the seller capability item display interface of some embodiments. The interface includes seller 1 and seller 2, wherein the capability items are used as capability item labels of the sellers and are displayed in association with the corresponding sellers. For example, if seller 1 has OEM capability and customization capability, then OEM capability and customization capability will be used as capability item labels of seller 1 and displayed in the information of seller 1; if seller 2 has sampling capability and innovation capability, then sampling capability and innovation capability will be used as capability item labels of seller 2 and displayed in the information of seller 2. It can be understood that what is shown in the figure is only an exemplary illustration. In actual application, capability items can also be displayed in other ways. In addition to being displayed in text form, capability items can also be displayed in icon form, or in a way that text and icons are displayed together. In addition, the display interface of capability items can be various interfaces that buyers can view, such as the seller's homepage, product recommendation interface, and event promotion interface.

[0110] In related technologies, industry operations need to have sufficient understanding of the professional characteristics of the industry and possess professional knowledge. Operations manually define the key capabilities of merchants in each industry. The entire process relies on the knowledge output of operations personnel and cannot timely define some emerging merchant capabilities, such as some innovative technical capabilities, and requires a large investment of time and resource costs from operations, R&D, and sellers.

[0111] The present disclosure uses the AI-generated content (AIGC) capability of the large language model 40 from a large number of communication records between buyers and sellers to extract the seller's capability items from the communication records. After extracting all the capability items, only the operation personnel need to classify all the capability items. If the capability item is a capability item that requires authenticity verification, the verification interface is called to perform authenticity verification, and the authenticity verification can be undertaken according to the existing verification method. If it is a capability item that does not require authenticity verification, the large language model 40's understanding and abstraction ability of the communication records is used to determine whether the seller has a certain capability item and whether he can undertake the relevant needs.

[0112] This disclosure eliminates the need for operators to possess deep expertise, nor does it require industry operators to constantly monitor emerging capabilities within the industry. When emerging capabilities emerge within the industry, they often appear in communications between sellers and buyers. The large language model 40 can extract these emerging capabilities based on contextual information, enabling automated and timely mining and definition, as well as determining whether sellers possess the relevant capabilities. Operators simply need to review relevant documentation throughout the entire process to determine which capabilities require verification or certification.

[0113] FIG8 illustrates the overall process of the embodiment of the present disclosure by taking the specific application scenario of extracting key attributes from an attribute set as an example. Some service demanders are not clear enough in their needs when calling service objects, and often use relatively broad search keywords to search for service objects. In this case, the present disclosure provides guidance on relevant needs by extracting key attributes that affect the decision-making of the service demander, helping the service demander to converge to relatively clear needs. Referring to FIG8 , still taking the example of the service supplier being a seller on the e-commerce platform and the service demander being a buyer on the e-commerce platform, the specific process of extracting key attributes from the attribute set includes:

[0114] Step S42: Generate attribute categories and importance rankings (i.e., the second ranking in the aforementioned embodiment) for each attribute information under each commodity leaf category through the large language model 40. The importance ranking here includes the importance ranking of each attribute information under the same commodity leaf category, and the importance ranking of each attribute information under the same attribute category. Figure 9 shows the attribute categories in a specific embodiment and the classification results of the attribute information by the large language model 40. It can be understood that what is shown in the figure is only an exemplary illustration. On different service platforms 20 and in different application scenarios, the classification method may be different from that shown in the figure. In the example shown in Figure 9, all attribute information under casual clothing can be sorted, and the attribute information under each attribute category such as basic attributes and marketing attributes can also be sorted separately. The sorting results of all attribute information under casual clothing are shown in Figure 10.

[0115] Step S44: Obtain the buyer's historical search data, such as the buyer's search data from the past year. Based on the frequency with which attribute values ​​expressed in the query keywords in the past year's search data are mapped to attribute information, calculate the importance ranking of each attribute information under the product leaf category (i.e., the first ranking in the aforementioned embodiment). The query keywords include specific attribute values ​​of the attribute information. For example, when searching for clothes, a user might enter "chiffon" or "pure cotton." "Chiffon" and "pure cotton" are both attribute values ​​for the attribute information "material." The attribute values ​​can be mapped to corresponding attribute information, and the frequency with which the attribute values ​​are mapped to the attribute information can be determined. For example, if the user enters "chiffon" twice and "pure cotton" three times, these attribute values ​​are mapped to the attribute information "material" five times. Based on the frequency with which the attribute values ​​are mapped to each attribute information, the attribute information can be ranked. For example, if the attribute values ​​are mapped to the three attribute information "material" five times, "color" once, and "brand" three times, respectively, the three attribute information can be ranked in descending order of importance as follows: material, brand, color.

[0116] Since the importance ranking given by the large language model 40 may be universally applicable to various scenarios, the user groups of different service platforms 20 have some unique characteristics in terms of identity and national distribution. The attribute values ​​expressed by users in the query keywords are more concerned about and influence their sourcing decisions. The embodiment of the present disclosure uses the number of attribute items mapped to the attribute values ​​expressed in the user's query keywords in the past year as an indicator to calculate the importance ranking of each attribute information under the product leaf category, and combines the ranking results of the large language model 40 with the ranking results based on the user's historical search data, thereby reducing the importance ranking of the attribute information from deviating too much from the focus of the user of the service platform 20 when it is in line with the facts.

[0117] Step S46: Weight the rankings in steps S42 and S44 to obtain a comprehensive importance ranking of each attribute information under the product leaf category, as well as a comprehensive importance ranking of the attribute information under each attribute category. The attribute information ranked in the top K is used as the key attribute, including the attribute information ranked in the top K under the product leaf category as the key attribute under the product leaf category, and the attribute information ranked in the top K under each attribute category as the key attribute under the attribute category. Since the user penetration rate in the search scenario may not be high, many users do not express their needs through search. Therefore, combining the ranking results of the large language model 40 with the ranking results based on the user's historical search data can effectively improve the accuracy of the ranking results. Figure 11 shows an example of taking the top 5 attribute information ranked in importance within an attribute category as the key attribute. Under the product leaf category of pest control, there are four attribute categories: core attributes, selling point attributes, basic attributes, and category attributes. The attribute information under each attribute category can be ranked in importance, and one or more core attributes under the attribute category can be determined based on the ranking results. As shown in Figure 11, the core attributes category includes five core attributes: pest type, energy source, application area, usage time, and specifications. The selling point attributes category includes only one core attribute, the characteristic attribute. Figure 12 shows an example of the importance ranking results for the overall attribute information under the product sub-category. It can be seen that in this embodiment, all attribute information under the product sub-category of pest control is ranked. The top K attribute information can be selected as key attributes based on actual needs.

[0118] Step S48: After determining the key attributes based on the sorting results, the key attributes and the attribute categories to which they belong can be automatically imported into data set platforms such as search, recommendation, scenario, and shopping guide.

[0119] Step S50: Key attributes can be applied to scenarios such as the screening area, product cards, and multi-product comparison in the foreground. For example, in the screening area, options corresponding to key attributes can be provided for users to choose from. In the product card area, products can be displayed according to their key attributes. In the multi-product comparison scenario, the comparison results of the key attributes of at least two products can be displayed. In addition, key attributes can also be used to train models for conversational service assistants. The sample data used for model training can be conversational data that includes key attributes. Key attributes under the corresponding attribute category can also be selected for application based on the actual usage scenarios of the business, thereby guiding the convergence of buyer demand.

[0120] Figure 13 shows a schematic diagram of a multi-product comparison scenario. Assuming the products to be compared are wireless headphones, and the number of products to be compared is 2, a list of their key attributes (such as communication method, function, battery life, etc.) can be displayed below the products to facilitate user comparison.

[0121] Figure 14 illustrates a schematic diagram of the process by which a conversational service assistant guides demand convergence. The conversational service assistant can communicate with the user through a conversational process. The conversational information input by the conversational service assistant may include product attribute information, which may be key attributes extracted by the large language model 40. The user can enter or select specific attribute values ​​for the attribute information through voice, text, or other means. For example, the conversational service assistant may enter, "What product would you like to buy?" The user's initial requirement may be relatively broad, such as "I want to buy headphones." The conversational service assistant may enter a conversational message containing key attributes, such as "Would you like to buy wired or wireless headphones?" The user can then further enter a more specific requirement, such as "I want to buy wireless headphones." The conversational service assistant can then enter further conversational information containing other key attributes and / or attribute values ​​of key attributes, such as "Would you like to buy wireless headphones with noise cancellation?" The user can then further enter their requirements and select attribute values. For example, if the user enters "Yes," the selected attribute value is "noise cancellation." In this way, the user can be gradually guided to converge on their needs.

[0122] 15 , the present disclosure further provides a data processing method, which is applied to the service platform 20. The method includes:

[0123] Step S62: receiving a data request from the client 10 for a target service object on the service platform 20;

[0124] Step S64: Responding to the data request, obtaining at least one piece of attribute information associated with the target service object;

[0125] Step S66: Pushing the at least one piece of attribute information to the client 10 so that the client 10 displays the at least one piece of attribute information;

[0126] Among them, after being extracted from the service data on the service platform 20 by the large language model 40 based on the prompt information, at least one attribute information is associated with the target service object, and the prompt information is used to prompt the large language model 40 to extract at least one attribute information included in the service data.

[0127] In steps S62 and S64, the service platform 20 may pre-establish a communication connection with the client 10 and, through the communication connection with the client 10, receive a data request from the client 10 for a target service object. The data request is used to obtain at least one piece of attribute information associated with the target service object. The target service object may be a service provider of the service platform 20. Accordingly, the attribute information associated with the target service object may be the service provider's service provision capabilities. For example, if the service platform is an e-commerce platform, the service provider may be a seller on the e-commerce platform. The service provision capabilities may include, but are not limited to, private label capabilities, innovation capabilities, and sampling capabilities. Alternatively, the target service object may be a resource on the service platform 20. Accordingly, the attribute information associated with the target service object may be key attributes of the resource. Each target service object may correspond to an attribute set, which includes one or more pieces of attribute information. Key attributes may be attribute information selected from the attribute set, and the key attributes may include a portion of the attribute information in the attribute set. Key attributes have a greater impact on the service requester's decision to invoke a service than other attributes in the attribute set other than the key attributes. For example, the target service object is the goods sold on the e-commerce platform, such as clothes. The attribute set of clothes includes attribute information such as color, material, origin, brand, and price. The material and price usually have a greater impact on the buyer's decision to buy clothes. Therefore, material and price are key attributes, while color, origin and brand usually have a smaller impact on the buyer's decision to buy clothes. Therefore, color, origin and brand are not key attributes.

[0128] The at least one piece of attribute information can be extracted using the large language model 40. The service platform 20 can associate and store the target service object and the at least one piece of attribute information. After receiving the data request sent by the client 10, the service platform 20 can obtain the attribute information associated with the target service object requested by the data request. The specific processing methods for extracting the at least one piece of attribute information and associating and storing the attribute information with the target service object are described in detail in the aforementioned embodiment and will not be repeated here.

[0129] In step S66 , the service platform 20 may send the acquired at least one piece of attribute information to the client 10 , so that the client 10 displays the at least one piece of attribute information.

[0130] In some embodiments, the target service object includes at least one service provider of the service platform 20, and the at least one piece of attribute information includes at least one service provision capability possessed by each service provider. Pushing the at least one piece of attribute information to the client 10 so that the client 10 displays the at least one piece of attribute information includes pushing the at least one service provision capability possessed by each service provider to the client 10 so that the client 10 displays the at least one service provision capability possessed by each service provider.

[0131] In some embodiments, the target service object includes a set of attributes of the service object called on the service platform 20, at least one attribute information includes target attribute information extracted from the attribute set and ranked in order of importance from high to low, the target attribute information includes partial attribute information in the attribute set, and the data request is used to obtain target attribute information of at least two service objects; pushing at least one attribute information to the client 10 so that the client 10 displays the at least one attribute information, including: pushing the target attribute information of at least two service objects to the client 10 so that the client 10 performs a structured display of the target attribute information of at least two service objects.

[0132] In some embodiments, the target service object includes a set of attributes of the service object called on the service platform 20, at least one attribute information includes target attribute information extracted from the attribute set and ranked in order of importance from high to low, the target attribute information includes partial attribute information in the attribute set, and the data request is used to trigger the dialogue service assistant of the service platform 20 to output dialogue information including the target attribute information, and the dialogue information is used to guide the user to input the target attribute value of the target attribute information; pushing at least one attribute information to the client 10 so that the client 10 displays the at least one attribute information, including: pushing the target object including the target attribute value to the client 10 so that the client 10 displays the target object including the target attribute value.

[0133] The specific details of the embodiment of this method are detailed in the aforementioned method embodiment and will not be repeated here.

[0134] The embodiment of the present disclosure further provides a service platform 20, which comprises at least a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method described in any of the aforementioned embodiments is implemented.

[0135] FIG16 shows a schematic diagram of the hardware structure of a more specific service platform 20 provided in an embodiment of the present disclosure. The device may include: a processor 52, a memory 54, an input / output interface 56, a communication interface 58, and a bus 60. The processor 52, the memory 54, the input / output interface 56, and the communication interface 58 are connected to each other within the device via the bus 60.

[0136] The processor 52 can be implemented using a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided by the embodiments of the present disclosure. The processor 52 may also include a graphics card, such as an Nvidia Titan X graphics card or an 1080Ti graphics card.

[0137] The memory 54 can be implemented in the form of a read-only memory (ROM), a random access memory (RAM), a static storage device, a dynamic storage device, etc. The memory 54 can store an operating system and other application programs. When the technical solutions provided by the embodiments of the present disclosure are implemented through software or firmware, the relevant program codes are stored in the memory 54 and called and executed by the processor 52.

[0138] The input / output interface 56 is used to connect to input / output modules to enable information input and output. The input / output modules can be configured as components within the device (not shown) or can be externally connected to the device to provide corresponding functions. Input devices may include a keyboard, mouse, touch screen, microphone, various sensors, etc., and output devices may include a display, speaker, vibrator, indicator light, etc.

[0139] The communication interface 58 is used to connect to a communication module (not shown) to enable communication between the device and other devices. The communication module can communicate via a wired method (such as USB, network cable, etc.) or a wireless method (such as mobile network, WiFi, Bluetooth, etc.).

[0140] Bus 60 comprises a pathway for transmitting information between various components of the device, such as processor 52 , memory 54 , input / output interface 56 , and communication interface 58 .

[0141] It should be noted that although the above device only shows the processor 52, memory 54, input / output interface 56, communication interface 58, and bus 60, in a specific implementation, the device may also include other components necessary for normal operation. In addition, those skilled in the art will understand that the above device may only include the components necessary to implement the embodiments of the present disclosure, and does not necessarily include all the components shown in the figure.

[0142] An embodiment of the present disclosure further provides a computer-readable storage medium having a computer program stored thereon, which implements the method described in any of the aforementioned embodiments when the program is executed by a processor.

[0143] Computer-readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission media that can be used to store information that can be accessed by a computer device. As defined herein, computer-readable media does not include transitory computer-readable media (transitory media), such as modulated data signals and carrier waves.

[0144] An embodiment of the present disclosure further provides a computer program product, including a computer program, which implements the method described in any embodiment of the present disclosure when executed by a processor.

[0145] Through the description of the above implementation methods, it can be seen that those skilled in the art can clearly understand that the embodiments of the present disclosure can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the embodiments of the present disclosure, or the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments of the present disclosure.

[0146] The systems, devices, modules, or units described in the above embodiments may be implemented by a computer device or entity, or by a product having certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email transceiver, game console, tablet computer, wearable device, or any combination of these devices.

[0147] Each embodiment in the present disclosure is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment. The device embodiment described above is merely illustrative, wherein the modules described as separate components may or may not be physically separated, and when implementing the embodiment of the present disclosure, the functions of each module can be implemented in the same one or more software and / or hardware. It is also possible to select some or all of the modules according to actual needs to achieve the purpose of the embodiment. A person of ordinary skill in the art can understand and implement it without paying any creative work.

[0148] The above is only a specific implementation of the embodiment of the present disclosure. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the embodiment of the present disclosure. These improvements and modifications should also be regarded as the scope of protection of the embodiment of the present disclosure.

Claims

1. A data processing method, applied to a service platform, wherein the service platform has a pre-established communication connection with a call interface of a large language model; the method comprising: Obtaining service data on the service platform and prompt information for inputting the large language model; The prompt information is used to prompt the large language model to extract at least one piece of attribute information included in the service data; Sending the service data and the prompt information to the calling interface, so that the large language model obtains the service data and the prompt information from the calling interface, and extracts the at least one attribute information from the service data based on the prompt information; The at least one piece of attribute information returned by the large language model through the calling interface is acquired, and association processing is performed on the at least one piece of attribute information and the target service object to which the at least one piece of attribute information belongs.

2. According to the method according to claim 1, the service platform is used to implement service object calls between the service provider of the service platform and the service demander of the service platform, the service data includes communication records between any service provider of the service platform and at least one service demander of the service platform, and the at least one attribute information includes at least one service supply capability.

3. The method according to claim 2, wherein the service data comprises: The communication records between the service provider who has chat records with multiple service demanders on the service platform and the multiple service demanders within a preset time window.

4. The method according to claim 2 or 3, wherein the large language model is further used to extract descriptive information corresponding to the at least one service provision capability from the communication record, the descriptive information corresponding to the service provision capability being used to indicate whether the service provider to which the service provision capability belongs has the service provision capability; The obtaining of the at least one piece of attribute information returned by the large language model through the calling interface, and associating the at least one piece of attribute information with the target service object to which the at least one piece of attribute information belongs, includes: Obtaining the at least one service provision capability returned by the large language model through the calling interface, and description information corresponding to the at least one service provision capability; For any one of the at least one service provision capability, determining, based on description information corresponding to the service provision capability, whether the service provider to which the service provision capability belongs has the service provision capability; If so, association processing is performed on the service provision capability and the service provider to which the service provision capability belongs.

5. The method according to claim 4, further comprising: For any one of the at least one service provision capability, determining a score for the service provider to which the service provision capability belongs having the service provision capability based on descriptive information corresponding to the service provision capability extracted from multiple communication records of the same service provider by the large language model; Based on the score, it is determined whether the service provider to which the service provision capability belongs has the service provision capability.

6. The method according to claim 5, wherein the plurality of communication records comprises at least one of the following: Communication records between the same service provider and multiple service demanders; Communication records between the same service provider and at least one service demander within multiple time windows.

7. According to the method according to claim 1, the service platform is used to implement service object calls between the service provider of the service platform and the service demander of the service platform, the service data includes a set of attributes of the service object called on the service platform, and the at least one attribute information includes target attribute information extracted from the attribute set and ranked in order of importance from high to low, and the target attribute information includes part of the attribute information in the attribute set.

8. The method according to claim 7, wherein the attribute information in the attribute set includes multiple attribute categories; and the at least one attribute information extracted from the service data by the large language model based on the prompt information includes: Target attribute information is extracted from the attribute information of each attribute category in the attribute set and sorted from high to low importance.

9. The method according to claim 7 or 8, further comprising: Obtaining historical search data of at least one service demander of the service platform on the service platform; The historical search data includes attribute information in the attribute set; adjusting the importance ranking of each piece of attribute information determined by the large language model according to the frequency of the attribute information included in the historical search data to obtain an adjusted importance ranking; The at least one piece of attribute information includes target attribute information extracted from the attribute set and sorted in descending order of importance after adjustment.

10. The method according to claim 9, wherein adjusting the importance ranking of each attribute information determined by the large language model based on the frequency of the attribute information included in the historical search data to obtain an adjusted importance ranking comprises: determining a first ranking of the attribute information in the attribute set according to the frequency of the attribute information included in the historical search data; obtaining a second ranking of the attribute information in the attribute set by using the large language model; The first ranking and the second ranking are weighted to obtain the adjusted importance ranking.

11. The method according to any one of claims 1 to 10, further comprising: before associating the at least one piece of attribute information with the target service object to which the at least one piece of attribute information belongs: performing normalization processing on the at least one piece of attribute information; and / or A reserved verification interface is called to perform authenticity verification on the at least one piece of attribute information.

12. A data processing method, applied to a service platform, comprising: Receiving a data request from a client for a target service object on the service platform; In response to the data request, obtaining at least one piece of attribute information associated with the target service object; Pushing the at least one piece of attribute information to the client, so that the client displays the at least one piece of attribute information; Among them, after being extracted from the service data on the service platform by the large language model based on prompt information, the at least one attribute information is associated with the target service object, and the prompt information is used to prompt the large language model to extract the at least one attribute information included in the service data.

13. The method according to claim 12, wherein the target service object comprises at least one service provider of the service platform, and the at least one piece of attribute information comprises at least one service provision capability possessed by each service provider; and pushing the at least one piece of attribute information to the client so that the client displays the at least one piece of attribute information comprises: The at least one service provision capability respectively possessed by each service provider is pushed to the client, so that the client displays the at least one service provision capability respectively possessed by each service provider.

14. The method according to claim 12, wherein the target service object comprises an attribute set of a service object called on the service platform, the at least one piece of attribute information comprises target attribute information extracted from the attribute set and sorted in descending order of importance, the target attribute information comprises partial attribute information in the attribute set, and the data request is used to obtain target attribute information of at least two service objects; and pushing the at least one piece of attribute information to the client so that the client displays the at least one piece of attribute information comprises: The target attribute information of the at least two service objects is pushed to the client, so that the client performs a structured display of the target attribute information of the at least two service objects.

15. The method according to claim 12, wherein the target service object comprises an attribute set of a service object called on the service platform, the at least one piece of attribute information comprises target attribute information extracted from the attribute set and sorted in descending order of importance, the target attribute information comprises partial attribute information in the attribute set, the data request is used to trigger a dialog service assistant of the service platform to output dialog information comprising the target attribute information, the dialog information is used to guide a user to input a target attribute value of the target attribute information; and the pushing of the at least one piece of attribute information to the client so that the client displays the at least one piece of attribute information comprises: The target object including the target attribute value is pushed to the client, so that the client displays the target object including the target attribute value.

16. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 15 is implemented.

17. A service platform comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 15 when executing the program.

18. A computer program product comprising a computer program, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 15 is implemented.

Images