Community operation methods, devices, equipment, and media based on large models

Abstract

One or more embodiments of this specification provide a community operation method, apparatus, device, and medium based on a large model. The community operation method can obtain inquiry information raised by users in an offline tenant community, perform semantic recognition on the corresponding inquiry information based on a pre-trained main model, and determine from which inquiry dimensions the user raises the corresponding inquiry information. Furthermore, based on the sub-model pre-trained under each inquiry dimension, the inquiry result information of the above inquiry information under the corresponding inquiry dimension is determined. Based on this, the main model can return the inquiry result information determined by all sub-models from different inquiry dimensions as reply information to the corresponding community for the user to view.

Description

Technical Field

[0001] This specification relates to the field of data processing technology in one or more embodiments, and more particularly to a community operation method, apparatus, device and medium based on a large model. Background Art

[0002] With the rapid development of internet technology and the increasing diversification of consumer behavior, brands have gradually recognized the importance of membership and fan engagement in market competition, and have entered the deep waters of membership and fan management. In this field, community management, as an efficient and direct form of interaction, has become a key means to enhance user stickiness, promote brand loyalty, and achieve precise marketing. Through communities, brands can reach their target user groups in real time to promote products, communicate brand awareness, collect user feedback, and engage in emotional exchange, thereby increasing user activity and conversion rates.

[0003] However, with the continuous expansion of community scale, the traditional manual community operation model faces many challenges. Specifically, most brands currently rely on dedicated operators to manage the daily operations of their communities. Under this model, with the surge in the number of communities and their followers, a single or small number of operators cannot effectively cover all communities, leading to decreased operational efficiency, insufficient community activity, and consequently affecting brand influence and user retention rates. In the long run, this results in declining community conversion rates, significant follower loss, and even the disbandment of some communities due to a lack of effective maintenance, causing substantial losses for brands. Summary of the Invention

[0004] This specification provides one or more embodiments of a community operation method, apparatus, device, and medium based on a large model, which simulates manual operation to provide users with timely services quickly and efficiently, reduce labor costs, and maintain users' trust in the brand.

[0005] Firstly, one or more embodiments of this specification provide a community operation method based on a large model, comprising: acquiring first query information input by a first user within a first community; determining at least one query dimension corresponding to the first query information based on a main model; the main model being obtained by model training from first training data, wherein the first training data is historical query information under different query scenarios; determining first sub-models associated with the at least one query dimension and the association order among all the first sub-models; each first sub-model being obtained by model training from second training data under a query dimension, wherein the second training data is historical question-answer pair information under the corresponding query dimension, and different first sub-models being used to determine the query result information of the first query information under the corresponding query dimension; acquiring the query result information determined by each first sub-model sequentially according to the association order; and returning the query result information determined by all the first sub-models as first response information to the first community.

[0006] In one optional embodiment, determining at least one query dimension corresponding to the first query information based on the master model includes: inputting the first query information into the master model; and obtaining at least one query dimension determined by the master model after performing semantic recognition on the first query information.

[0007] In one optional embodiment, determining at least one query dimension corresponding to the first query information based on the master model includes: performing compliance verification on the first query information; and after the verification is passed, determining at least one query dimension corresponding to the first query information based on the master model.

[0008] In one optional embodiment, determining the first sub-models associated with each of the at least one query dimension and the association order among all the first sub-models includes: determining a first query scenario corresponding to the first query information based on the semantic recognition result of the first query information by the main model, wherein the first query scenario corresponds to the at least one query dimension; obtaining pre-configured mapping information for the first query scenario; the mapping information includes the second sub-models associated with each of the query dimensions corresponding to the first query scenario and the association order among all the second sub-models, wherein all the second sub-models include all the first sub-models; and determining the first sub-models associated with each of the at least one query dimension and the association order among all the first sub-models based on the mapping information.

[0009] In an optional embodiment, according to the association order, the query result information determined by each first sub-model is obtained sequentially, including: determining the first first sub-model in the association order; inputting the first query information into the first first sub-model and obtaining the query result information determined by the first first sub-model; according to the association order between the first first sub-model and other first sub-models, sequentially inputting the query result information determined by the previous first sub-model as input parameters into the next first sub-model, and sequentially obtaining the query result information of each first sub-model.

[0010] In an optional embodiment, inputting the first query information into the first sub-model and obtaining the query result information determined by the first sub-model includes: formatting the first query information according to the semantic recognition result of the first query information by the main model; inputting the formatted first query information as an input parameter into the first sub-model; and obtaining the query result information determined by the first sub-model.

[0011] In an optional embodiment, according to the association order between the first sub-model and other first sub-models, the query result information determined by the previous first sub-model is sequentially input as input parameters into the next first sub-model, and the query result information of each first sub-model is sequentially obtained, including: according to the association order between the first sub-model and other first sub-models, the formatted processing result of the query result information determined by the previous first sub-model is sequentially input as input parameters into the next first sub-model, and the query result information of each first sub-model is sequentially determined.

[0012] In one optional embodiment, returning all query result information determined by the first sub-model as first response information to the first community includes: performing compliance verification on all query result information determined by the first sub-model; and returning all query result information determined by the first sub-model as first response information to the first community after the verification is passed.

[0013] In an optional embodiment, the method further includes: obtaining second query information input by a second user within the first community; determining the association between the first query information and the second query information; obtaining second response information based on the association and returning the second response information to the first community.

[0014] Secondly, one or more embodiments of this specification also provide a community operation device based on a large model, comprising: a first acquisition module, configured to acquire first query information input by a first user within a first community; a first determination module, configured to determine at least one query dimension corresponding to the first query information based on a main model; the main model is obtained by training a model with first training data, the first training data being historical query information under different query scenarios; a second determination module, configured to determine the first sub-models associated with the at least one query dimension and the association order among all the first sub-models; each first sub-model is obtained by training a model with second training data under a query dimension, the second training data being historical question-answer pair information under the corresponding query dimension, different first sub-models being used to determine the query result information of the first query information under the corresponding query dimension; a second acquisition module, configured to sequentially acquire the query result information determined by each first sub-model according to the association order; and an output module, configured to return the query result information determined by all the first sub-models as first response information to the first community.

[0015] In an optional embodiment, the first determining module determines at least one query dimension corresponding to the first query information based on the master model, and is used to: input the first query information into the master model; and obtain at least one query dimension determined by the master model after performing semantic recognition on the first query information.

[0016] In an optional embodiment, the first determining module determines at least one query dimension corresponding to the first query information based on the master model, and is used to: perform compliance verification on the first query information; and after the verification is passed, determine at least one query dimension corresponding to the first query information based on the master model.

[0017] In an optional embodiment, the second determining module determines the first sub-models associated with the at least one query dimension and the association order among all the first sub-models, and is configured to: determine the first query scenario corresponding to the first query information based on the semantic recognition result of the first query information by the main model, wherein the first query scenario corresponds to the at least one query dimension; obtain pre-configured mapping information for the first query scenario; the mapping information includes the second sub-models associated with all the query dimensions corresponding to the first query scenario and the association order among all the second sub-models, wherein all the second sub-models include all the first sub-models; and determine the first sub-models associated with the at least one query dimension and the association order among all the first sub-models based on the mapping information.

[0018] In an optional embodiment, the second acquisition module sequentially acquires the query result information determined by each first sub-model according to the association order, and is used to: determine the first first sub-model in the association order according to the association order; input the first query information into the first first sub-model and acquire the query result information determined by the first first sub-model; and sequentially input the query result information determined by the previous first sub-model as input parameters into the next first sub-model according to the association order between the first first sub-model and other first sub-models, and sequentially acquire the query result information of each first sub-model.

[0019] In an optional embodiment, the second acquisition module inputs the first query information into the first sub-model and acquires the query result information determined by the first sub-model, for the following purposes: formatting the first query information according to the semantic recognition result of the first query information by the main model; inputting the formatted first query information as an input parameter into the first sub-model; and acquiring the query result information determined by the first sub-model.

[0020] In an optional embodiment, the second acquisition module, according to the association order between the first sub-model and other first sub-models, sequentially inputs the query result information determined by the previous first sub-model as input parameters into the next first sub-model, and sequentially acquires the query result information of each first sub-model, for the purpose of: sequentially inputting the formatted processing result of the query result information determined by the previous first sub-model as input parameters into the next first sub-model according to the association order between the first sub-model and other first sub-models, and sequentially determining the query result information of each first sub-model.

[0021] In an optional embodiment, the output module returns all the query result information determined by the first sub-model as the first response information to the first community, for the purpose of: performing compliance verification on all the query result information determined by the first sub-model; and returning all the query result information determined by the first sub-model as the first response information to the first community after the verification is passed.

[0022] In an optional embodiment, the first acquisition module is further configured to: acquire second inquiry information input by a second user within the first community; determine the association between the first inquiry information and the second inquiry information; acquire second response information based on the association and return the second response information to the first community.

[0023] Thirdly, one or more embodiments of this specification also provide a computer device, including: a memory and a processor, the memory storing computer program instructions, which, when executed by the processor, are used to implement the large-model-based community operation method of the first aspect.

[0024] Fourthly, one or more embodiments of this specification also provide a computer-readable storage medium storing computer program instructions that, when executed, implement the community operation method based on a large model described in the first aspect.

[0025] In summary, based on the above, the community operation method based on a large model provided in one or more embodiments of this specification can acquire inquiry information raised by users within offline tenant communities, perform semantic recognition on the corresponding inquiry information based on a pre-trained main model, and determine from which inquiry dimensions the user raised the corresponding inquiry information. Furthermore, based on sub-models pre-trained under each inquiry dimension, the inquiry result information of the above inquiry information under the corresponding inquiry dimension is determined. Based on this, the main model can return the inquiry result information determined by all sub-models from different inquiry dimensions as the reply information to the corresponding community for users to view. In this method, the automatic identification and timely feedback of various inquiry information raised by users within the community based on each model not only simulates manual operation and reduces labor costs, but also provides users with timely service quickly and efficiently, helping to maintain user trust in the brand, improve user retention rate, and provide a foundation for user conversion for other derivative transactions. Attached Figure Description

[0026] To more clearly illustrate the technical solutions of one or more embodiments of this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of one or more embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1a A system architecture diagram of a community operation system based on a large model, provided for one or more embodiments of this specification;

[0028] Figure 1b A system architecture diagram of another community operation system based on a large model, provided for one or more embodiments of this specification;

[0029] Figure 1c A system architecture diagram of another community operation system based on a large model, provided for one or more embodiments of this specification;

[0030] Figure 2a A flowchart illustrating a community operation method based on a large model, provided for one or more embodiments of this specification;

[0031] Figure 2b A flowchart illustrating another community operation method based on a large model, provided for one or more embodiments of this specification;

[0032] Figure 3 A structural block diagram of a community operation device based on a large model, provided for one or more embodiments of this specification;

[0033] Figure 4 This is a structural block diagram of a computer device provided for one or more embodiments of this specification. Detailed Implementation

[0034] The present specification provides one or more embodiments in further detail below with reference to the accompanying drawings and examples. Through these descriptions, the features and advantages of one or more embodiments of the present specification will become clearer and more apparent.

[0035] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments. Although various aspects of embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless specifically indicated otherwise.

[0036] Furthermore, the technical features described below in one or more embodiments of this specification may be combined with each other as long as they do not conflict with each other.

[0037] To facilitate understanding, the application scenarios of the technical solutions provided in one or more embodiments of this specification will be described below.

[0038] In terms of user operations, brands typically maintain merchant relationships through community building and regularly track user activity within these communities to provide timely and targeted services, thereby enhancing user engagement and trust. However, as the number of users within these communities increases, the difficulty of community operations also grows. It becomes challenging for operations staff to respond promptly to the various questions raised by users, leading to a poor user experience and, in the long run, jeopardizing user retention rates.

[0039] Therefore, one or more embodiments of this specification provide a community operation method based on a large model, which is used to automatically identify and provide timely feedback on various questions raised by users in the community, thereby improving operational efficiency and user experience while reducing the labor costs of community operation.

[0040] In one or more embodiments of this specification, the executing entity of the above method is not limited. Depending on the application scenario, the executing entity of the above method may also be different. Therefore, before describing each step in the above method, the application scenarios of the above method are illustrated by several optional methods.

[0041] Figure 1a This is a system architecture diagram for an optional community operation system based on a large model. In this embodiment, the above method is applied to a centralized deployment system architecture, such as... Figure 1a As shown, the above method is executed by a cloud server. The cloud server can communicate with the brand's local server via an interface to obtain basic brand information, local community user information, community messages, and other information. Then, based on the obtained information, it performs semantic recognition, question confirmation, and feedback on user-generated questions within the community. This can be understood as the cloud server being the server corresponding to the service platform that provides operational services to various brands, such as... Figure 1a As shown in this embodiment, the program files of the above method and the multiple models that it depends on during execution are deployed in the service platform. Each model and the above method can be packaged into a service product and provided to each brand. For example, each brand can be a tenant that leases the above service product from the service platform.

[0042] Figure 1b Here is a system architecture diagram for another alternative community operation system based on a large model, such as... Figure 1b As shown, in this embodiment, the service platform provides the program files corresponding to the above methods to various brands in an open-source format, and deploys the various models involved in the execution of the above methods on a cloud server. Based on this, each brand can deploy the corresponding program files on a local server, and the local server can execute the above methods. Accordingly, as... Figure 1b As shown in the embodiment, the cloud server can communicate with the brand's local server through an interface to obtain basic information about the brand, user information of the local community, community messages, and other information from the brand's local server; then, based on the obtained information, each model performs semantic recognition, question confirmation, and feedback on the questions raised by users in the community.

[0043] Figure 1c Here is a system architecture diagram for another alternative community operation system based on a large model, such as... Figure 1c As shown, in this embodiment, the service platform provides the program files corresponding to the above methods and the various models involved in the execution of the above methods to each brand in an open-source form. Based on this, each brand can deploy the corresponding program files and various models on a local server, and the local server can execute the above methods. Correspondingly, as... Figure 1cAs shown, during the execution of the above method, basic information of the brand, user information of the local community, community messages, and other information can be obtained from the local server. Then, based on the obtained information, each model performs operations such as semantic recognition, question confirmation, and feedback on questions raised by users in the community.

[0044] exist Figure 1b and Figure 1c In the example shown, regarding the program files and various models provided to brands in open-source form, each brand can modify the corresponding program code appropriately according to its own needs before deployment to match its own brand's operational requirements. Correspondingly, in another optional approach, the service platform also provides each brand with the program files corresponding to the above methods, or provides both the program files and various models simultaneously. However, it requires that the program files and various models be deployed on each brand's local server in a private manner; that is, each brand cannot modify the program code corresponding to the above methods and various models. For details on the corresponding system architecture and specific execution methods in this embodiment, please refer to [link to relevant documentation]. Figure 1b and Figure 1c The explanation of the example shown will not be repeated here.

[0045] It should be noted that the specific form of the program files for the above methods and the deployment methods of the various models involved in the execution can be determined according to the specific needs of the service platform or the agreement with each brand, and no specific restrictions are imposed.

[0046] It should be further noted that the above application scenarios are merely illustrative examples and are not limited to these in actual applications. Any scenario in which the above methods are applicable can be considered to be within the protection scope of one or more embodiments of this specification, and will not be elaborated upon here.

[0047] The following describes, with reference to the accompanying drawings, one or more embodiments of the community operation method based on a large model provided in this specification.

[0048] Figure 2a A flowchart illustrating a community operation method based on a large model, provided for one or more embodiments of this specification, is shown below. Figure 2a As shown, the method includes:

[0049] S102. Obtain the first query information input by the first user in the first community;

[0050] S104. Based on the main model, determine at least one query dimension corresponding to the first query information. The main model is obtained by training the model from the first training data. The first training data is historical query information under different query scenarios.

[0051] S106. Determine the first sub-model associated with at least one query dimension and the association order among all the first sub-models; wherein, each first sub-model is obtained by training the model with second training data under a query dimension, the second training data is the historical question-answer pair information under the corresponding query dimension, and different first sub-models are used to determine the query result information of the first query information under the corresponding query dimension.

[0052] S108. According to the association order, obtain the query result information determined by each first sub-model in sequence;

[0053] S110. Return all query result information determined by the first sub-model to the first community as the first response information.

[0054] Before explaining each step of the above method, we will first explain the functions of each model involved in the execution of the above method.

[0055] In one or more embodiments of this specification, a master model is provided, which is obtained by training a model using first training data. The first training data refers to historical query information under different query scenarios. Each query scenario corresponds to at least one query dimension. The query dimensions corresponding to different query scenarios can be the same or different, and this is not limited here. The query scenario refers to the application scenario type of each brand, such as, but not limited to, catering, clothing, furniture and appliances, cultural and creative products, etc. The query dimension corresponding to any query scenario can be understood as the perspective from which users raise various questions to a specific object within the community, targeting that particular query scenario.

[0056] For example, taking the restaurant inquiry scenario, users in the corresponding community could ask questions from different perspectives: "Does restaurant A have mushroom flavor?" (based on taste); "Can different types of coupons be used together?" (based on promotions); "Is store B still open?" (based on the merchant); or "What should I do if I find a foreign object in restaurant C and no one is responsible?" (based on complaints or suggestions). Similarly, for other inquiry scenarios, users can ask different questions from different perspectives to address specific individuals within those scenarios, which will not be elaborated upon here.

[0057] In one or more embodiments of this specification, a corresponding sub-model is pre-trained for each query dimension. Each sub-model is trained using second training data for a given query dimension, where the second training data consists of historical question-answer pairs for that query dimension. Different sub-models are used to determine the query result information corresponding to that query dimension. Based on this, each sub-model can perform semantic recognition on various questions raised by the user, using the historical question-answer pairs for that query dimension, to determine the query result information for that query dimension.

[0058] Based on the above, in order to achieve mutual scheduling between the main model and the various sub-models, in one or more embodiments of this specification, corresponding mapping information is pre-configured for each query scenario. The mapping information for each query scenario may include the sub-models associated with all query dimensions corresponding to the relevant query scenario, as well as the association order between all sub-models. It may also include the association relationships between each sub-model and other sub-models used to perform general functions. Based on this association order, orderly scheduling between the main model and the various sub-models can be achieved, so that for various questions raised by the user, the corresponding query result information can be determined sequentially from different query dimensions.

[0059] Optionally, embodiments of this specification provide a processing sub-model for handling complaint or suggestion information. This processing sub-model can be considered as a sub-model for performing general functions. If the main model identifies the content of the first query information as a complaint or suggestion, it can directly hand it over to the processing sub-model to determine the corresponding solution and return it to the requesting end. For example, if the first query information is a complaint about a product having a quality problem, after identifying the specific complaint information in the first query information, the processing sub-model can determine the product's details through the product sub-model, determine the user information corresponding to the complaining user through the user model, and determine the merchant information corresponding to the product through the merchant sub-model. Then, it can notify the corresponding merchant of the above user information and corresponding complaint information for processing, or it can return the queried merchant information to the requesting end for the user to contact the corresponding merchant for processing. Of course, the sub-model for performing general functions is not limited to the above processing sub-model. Accordingly, the operations performed by the processing sub-model in the above embodiments are only illustrative examples and are not limited thereto.

[0060] Based on the above, the following is... Figure 1a Taking the centralized deployment method shown as an example, this specification describes the community operation method based on a large model proposed in one or more embodiments. Accordingly, this method can be executed by a cloud server.

[0061] It should be noted that, in one or more embodiments of this specification, the execution entity for scheduling each sub-model according to the above-described association order is not limited. Optionally, after the main model determines the inquiry scenario corresponding to the user's question, it can also determine the various sub-models associated with the corresponding inquiry scenario and the association order between the various sub-models based on pre-configured mapping information, and then schedule the various sub-models in an orderly manner according to the association order; or, after the main model determines the inquiry scenario corresponding to the user's question, it can output it as a recognition result to the cloud server, and the cloud server can determine the various sub-models associated with the corresponding inquiry scenario and the association order between the various sub-models based on pre-configured mapping information, and then schedule the various sub-models in an orderly manner according to the association order.

[0062] In one or more embodiments of this specification, the method of determining the association order between various sub-models and scheduling each sub-model is used as an example for illustration.

[0063] For clarity, in one or more embodiments of this specification, any community is referred to as the first community, and different users within the first community are distinguished as first users and second users; different questions entered by users within the first community are distinguished as first query information and second query information; the tenant corresponding to the first community is referred to as the first tenant, and the query scenario corresponding to the first tenant is referred to as the first query scenario; the sub-model associated with all query dimensions corresponding to the first query information entered by the first user is referred to as the first sub-model, and the sub-model associated with all query dimensions corresponding to the first query scenario is referred to as the second sub-model, that is, the second sub-model includes the first sub-model; the request information sent by the local server of the first tenant to the cloud server each time is distinguished as first group message and second group message, and so on, without further details.

[0064] Based on this, when the first user enters the first query information in the first community, the local server of the corresponding first tenant can initiate an operation service request to the cloud server through the communication interface between the two servers, and send the first group message to the cloud server. The first group message includes the first query information, and may also include the first tenant's basic information, necessary information for communication, etc. The specific content can be determined according to actual needs and is not limited here.

[0065] Furthermore, after receiving the first group of messages, the cloud server can perform semantic recognition on the first query information based on the main model to determine the first query scenario and at least one query dimension corresponding to the first query information. Further, based on the mapping information corresponding to the first query scenario, the first sub-model associated with each query dimension and the association order among all first sub-models can be determined. Then, according to this association order, the query result information determined by each first sub-model under the corresponding query dimension is obtained sequentially, and the query result information determined by all first sub-models is returned to the first community as the first response information.

[0066] In one or more embodiments of this specification, the specific method by which the cloud server determines at least one query dimension based on the master model is not limited.

[0067] In one optional embodiment, the cloud server can first input the first query information into the main model. Since the main model is trained based on historical query information under different query scenarios, which are questions raised by historical users from different query dimensions regarding specific objects in different query scenarios, the main model can perform semantic recognition on the first query information based on the historical query information under different query scenarios to determine the first query information corresponds to the first query scenario and at least one query dimension corresponding to the first query scenario.

[0068] In another optional embodiment, before inputting the first query information into the main model, the cloud server may first perform a compliance check on the first query information to determine whether the request meets the requirements. Further, if the check passes, the first query information is input into the main model to determine at least one query dimension corresponding to the first query information based on the main model; correspondingly, if the check fails, a prompt message is returned to the requesting end to inform it that the initiated request does not meet the requirements. The specific content of the compliance check is not limited; optionally, it may check whether the first query information involves harmful information that violates laws and regulations, and it may also check whether the first query information includes pre-specified sensitive information, such as information related to user identity, address, contact information, ID number, and other personal privacy information.

[0069] In another optional embodiment, in addition to verifying the compliance of the content of the first query information, the type of the first query information can also be verified to determine whether the first query information is invalid information unrelated to the application scenario of the first community. For example, it can be verified whether the first query information is a third-party advertisement, spam, third-party access link, etc. Optionally, in order to maintain the community environment, if the first query information is determined to be invalid, the requesting party can be instructed to withdraw it directly. Further optionally, a corresponding input count threshold can be set for each user to determine the number of times each user inputs invalid information in the first community. Based on this, if it is determined that the number of times the first user inputs invalid information has reached the preset input count threshold, the requesting party can be instructed to perform operations such as muting, kicking out of the community, or blacklisting the first user, and the specific method is not limited.

[0070] It should be noted that, in the above example, the operations of performing compliance verification on the content of the first query information and invalidation verification on the type of the first query information can be performed by the main model or delegated to other sub-models. Further optionally, the operations of performing compliance verification on the content of the first query information and invalidation verification on the type of the first query information can be performed by a single sub-model or by different sub-models. For example, compliance verification on the content of the first query information can be performed by the verification sub-model, and invalidation verification on the type of the first query information can be performed by the masking sub-model. Of course, the above is merely an illustrative example and is not limited thereto.

[0071] In one or more embodiments of this specification, since each first sub-model is trained based on historical question-and-answer pair information under a certain question dimension, each first sub-model can perform semantic recognition on the first question information raised by the first user based on the historical question-and-answer pair information under the corresponding question dimension, and determine the question result information matched by the first question information under the corresponding question dimension.

[0072] Based on this, after determining the first sub-models associated with at least one query dimension corresponding to the first query information based on pre-configured mapping information, and the association order among all the first sub-models, the main model can determine the first first sub-model according to this association order. Further, the first query information is input into the first first sub-model, and the query result information determined by that first sub-model is obtained. Further, according to the association order between the first first sub-model and other first sub-models in the above association order, the query result information determined by the previous first sub-model is sequentially input as an input parameter into the next first sub-model, until all first sub-models in the above association order have been executed, and the query result information of each first sub-model is obtained sequentially.

[0073] In one or more embodiments of this specification, the specific method by which the main model inputs the first query information into each first sub-model and obtains query result information from each first sub-model is not limited. Optionally, before inputting the first query information into the first first sub-model in the aforementioned association order, the main model may first format the first query information based on the semantic recognition result of the first query information. Further, the formatted first query information is used as an input parameter into the first first sub-model in the aforementioned association order, thereby obtaining the query result information determined by the first sub-model.

[0074] Accordingly, for the query result information determined by the multiple first sub-models, the query result information determined by the previous first sub-model can be formatted in the order of association between the multiple first sub-models, and the formatted query result information can be used as input parameters to input the next first sub-model, until all the first sub-models in the above association order have been executed, and the query result information of each first sub-model is determined in turn.

[0075] In one or more embodiments of this specification, the executing entity for formatting the query result information determined by each first sub-model is not limited. Optionally, the main model may perform formatting on the query results of all first sub-models, or each first sub-model may perform formatting on its determined query result information.

[0076] Optionally, when executed by the main model, each first sub-model only needs to determine the query result information corresponding to the first query information under the corresponding query dimension and output it to the main model. Then, the main model performs formatting processing on the query result information output by each first sub-model. Correspondingly, when executed by each first sub-model, each first sub-model first determines the query result information corresponding to the first query information under the corresponding query dimension, then performs formatting processing on the corresponding query result information, and outputs the formatting processing result to the main model.

[0077] It should be noted that the specific method used can be selected according to actual needs. In one or more embodiments of this specification, the example of the main model performing formatting processing on the query result information determined by all first sub-models is used for illustration.

[0078] Below, using the restaurant inquiry scenario as an example, we will provide an exemplary explanation of the process by which the main model performs the above operations.

[0079] Suppose the first user's user ID is "2088XYZW". Their first query in the first community, entered in text format, is: "How many flavors of fries are there?". When the first tenant's local server initiates an operational service request to the cloud server, it can send the first query and user ID as the first group message to the cloud server. Based on this, upon receiving the first group message, the cloud server can first perform a compliance check to determine if it contains any illegal information. Furthermore, if it is determined that the first group message contains pre-defined illegal information, it returns a prompt informing the tenant that the request does not meet the requirements. Based on this, the first tenant's local server can resend a compliant second group message to re-initiate the operational service request. Conversely, if the first group message is deemed compliant, it is input into the main model for subsequent processing.

[0080] Optionally, after receiving the first set of messages, the main model can identify the user identifier corresponding to the first user and the first query information input by the first user from the first set of messages. Further, based on historical query information under different query scenarios, semantic recognition is performed on the first query information to determine at least one query dimension involved in the first query information. In addition, the main model can also identify other necessary information from the first query information and use the identified information together as the recognition result. For example, if the main model identifies the keyword "flavor" from the first query information, it can determine that the query dimension corresponding to the first query information is "flavor type"; further, if the main model identifies the keyword "fries" from the first query information, it can determine that the product type corresponding to the first query information is "fries".

[0081] Based on this, the main model can format the above recognition results. Assume that after formatting by the main model, the recognition results are obtained in the following format:

[0082]

[0083] In this structure, "GoodsType" identifies the product category, with "French Fries" representing the product category mentioned in the first query. "UserId" identifies the user, with "2088XYZW" representing the user ID of the first user who entered the first query within the first community. "InquiryType" identifies the query dimension, with "Flavor Category" representing the query dimension corresponding to the first query. As can be seen, after formatting, the first query is transformed from colloquial text into a key-value pair string, making it easier for machines to recognize and process.

[0084] It should be noted that the "key-value pair" string format of the recognition results described above is merely illustrative. Optionally, the string format can be a lightweight data interchange format (JavaScript Object Notation, JSON), but it is not limited to this in practical applications. Depending on the specific processing requirements, the formatted recognition results can also be in other formats. For example, the formatted recognition results of the first query information can also be in Extensible Markup Language (XML), binary files, or plain text files, etc. The specific format used is not limited here. In the following embodiments, JSON format is used as an example for illustrative purposes.

[0085] Assume that, based on pre-configured mapping information for the query scenario of "food and beverage," the first sub-model directly associated with the query dimension of "flavor type" is determined to be the product sub-model. The subsequent sequentially associated first sub-models are the mapping sub-model and the promotion sub-model. The product sub-model can query product information corresponding to the query dimension of "flavor type." The mapping sub-model can query corresponding user information by combining the user identifier identified by the main model, and determine the mapping information between the corresponding user information and the aforementioned product information based on the query dimension of "flavor type" identified by the main model and the product information queried by the product sub-model. The promotion sub-model can recommend available promotional activities, coupons, and related products to users based on this mapping information, with no specific limitations on the content.

[0086] It should be noted that the specific content of product information and user information, as well as the specific content of the mapping information determined by the mapping sub-model, is not limited in one or more embodiments of this specification. In specific application scenarios, user information and product information may differ, and correspondingly, the mapping information between them may also differ. For example, user information may have differences in region, level, personal information, etc., while product information may have differences in product type, flavor type, sales region, sales volume, recommendation index, inventory, etc. Therefore, the mapping information determined by the mapping sub-model based on different user information and different product information will also be different.

[0087] In one or more examples in this specification, the information regarding the determination of each first sub-model is merely illustrative and intended to explain the implementation principle of the corresponding examples, and does not constitute a specific limitation.

[0088] For ease of explanation, the content obtained by formatting the recognition result of the first query information by the main model is called parameter 1. Based on this, the main model can use parameter 1 as the input parameter of the product sub-model and input it into the product sub-model so that the product sub-model can perform subsequent processing according to parameter 1.

[0089] Furthermore, after receiving parameter 1 from the main model, the product sub-model can identify that the value of "GoodsType" is "French Fries" and the value of "InquiryType" is "Flavor Type". Based on this, the product sub-model can query the quantity and name of the "Flavor Type" corresponding to the product type "French Fries" from the product information database as query results. Optionally, in addition to the directly retrieved quantity and name, the product sub-model can also query other relevant information. For example, among the retrieved "Flavor Type" names, the product sub-model can also query the name of the "Flavor Type" with the highest sales volume. Of course, it is not limited to this; it can also query other relevant information, which can be determined according to actual needs.

[0090] Suppose that, based on the above query, the product sub-model determines that there are three flavors for the product type "French Fries," namely "Cheese," "Sea Salt," and "Black Pepper," with "Cheese" being the best-selling. Based on this, the product sub-model can output the query results to the main model. Upon receiving the query results from the product sub-model, the main model can perform the same formatting operation as the first query to convert the query results from the product sub-model into JSON format.

[0091] Suppose that the query results output by the product sub-model, after being formatted, result in the following content:

[0092]

[0093] For ease of explanation, the content obtained after the above formatting process will be referred to as parameter 2. Based on this, the main model can determine the input parameters of the mapping sub-model according to parameter 1 and parameter 2, and input these input parameters into the mapping sub-model to schedule the mapping sub-model to perform subsequent operations. Optionally, when determining the input parameters of the mapping sub-model, the main model can first determine at least one indicator parameter based on the query dimension of "flavor type" identified from the first query information, wherein each indicator parameter is used to instruct the mapping sub-model to perform the corresponding operation.

[0094] In this embodiment, the number and specific content of the indicator parameters corresponding to the query dimension "flavor type" are not limited. Assume that the main model determines that the query dimension "flavor type" corresponds to three indicator parameters, which are used by the mapping submodel to determine the first user's "activity level," "flavor preference," and "recommendation acceptance." Optionally, "activity level" refers to the number of items purchased most frequently by the first user, "flavor preference" refers to the name of the most frequently purchased "flavor type" among the food items purchased by the first user, and "recommendation acceptance" refers to the number of items purchased by the first user from products recommended by the brand or other users. The content corresponding to these items in the "purchase source" type can be identified as "recommendation."

[0095] Furthermore, this embodiment does not limit the method for determining the above three indicator parameters. Optionally, based on the first user's historical purchase record information, the main model can determine the "activity level" based on the quantity of the product type "French Fries" or all food purchased, determine the "flavor preference" based on the "flavor type" with the largest quantity among the product type "French Fries" or all food purchased, and determine the "recommendation acceptance" based on the quantity of the product type "French Fries" or all food purchased with "recommendation" as the "purchase source".

[0096] Of course, the above explanation of each indicator parameter and the method of determining the content of each indicator parameter are only illustrative examples and are not limited to in actual applications. For each indicator parameter, the specific content of each parameter can also be determined from other perspectives. Correspondingly, for different query scenarios, the content of the corresponding indicator parameters and the method of determining the corresponding content can also be different, which will not be detailed here.

[0097] Furthermore, after determining at least one indicator parameter corresponding to the query dimension of "flavor type", the main model can determine the input parameters of the mapping sub-model based on parameter 1, parameter 2 and at least one indicator parameter.

[0098] In this embodiment, the specific method by which the main model determines the input parameters of the mapping sub-model is not limited. Optionally, the main model can directly determine parameter 1, parameter 2, and at least one indicator parameter as the input parameters of the mapping sub-model; or, it can determine a parameter 3 based on parameter 1, parameter 2, and at least one indicator parameter, and use this parameter 3 as the input parameter of the mapping sub-model. The specific method used can be determined according to the input format of the mapping sub-model. In this embodiment, parameter 3 is used as an example for explanation. To simplify the processing, the main model can directly concatenate parameter 1, parameter 2, and at least one indicator parameter into a JSON format, and determine the concatenated content as parameter 3 and input it into the mapping sub-model.

[0099] In this embodiment, "ActivityLevel" represents "activity level", "FoodPrefer" represents "taste preference", and "AcceptLevel" represents "recommendation acceptance level". Assuming that parameter 3 is obtained by concatenating parameter 1, parameter 2, and the above three indicator parameters, it is as follows:

[0100]

[0101] Based on this, after the main model inputs parameter 3 into the mapping sub-model, the mapping sub-model can identify the value of "UserId" as "2088XYZW" from parameter 3. If it identifies "ActivityLevel", "FoodPrefer", and "AcceptLevel" as "?", it determines that the operation to be performed is to determine the values ​​of these three parameters. Based on this, the mapping sub-model can query the first user's historical purchase records based on the user identifier value "2088XYZW" and determine the three parameter values ​​from them using the aforementioned method. For example, taking the query of a user's historical purchase records for the product "French Fries" as an example, suppose that after the mapping sub-model retrieves the purchase records related to "French Fries" from the user's historical purchase records based on "2088XYZW", it determines the number of times the user purchased "French Fries", identifies the name of the "flavor" with the highest purchase volume, and counts the number of times the "purchase source" for "French Fries" was "recommended". Based on this, the mapping sub-model can output the queried information as query result information for the main model to perform subsequent processing based on this query result information.

[0102] It should be noted that when querying the first user's historical purchase records, the mapping sub-model can query all historical purchase records or historical purchase records within a specified historical period, such as the past month, three months, six months, etc., without any limitation on the specific duration.

[0103] Furthermore, after receiving the query result information output by the mapping sub-model, the main model can format the query result information in the same way, and concatenate the formatted content with parameter 3 to obtain parameter 4, which is then used as the input parameter of the promotion sub-model and input into the promotion sub-model to schedule the promotion sub-model to perform subsequent operations.

[0104] Assuming that after the above formatting and concatenation, the resulting parameter 4 is as follows:

[0105]

[0106] Based on this, after receiving parameter 4, the promotion sub-model can identify the types of each parameter and their corresponding values. For example, based on the identification of parameter 4, the promotion sub-model can determine that the first user's user ID is "2088XYZW", the product type the first user inquired about is "French fries", and the query dimension of the first user's input query information is "flavor type". Currently, there are three flavors of "French fries" on sale: "cheese", "sea salt", and "black pepper", with "cheese" being the best-selling flavor. In the first user's historical purchase records, the "activity level" of purchasing "French fries" is 7, the "flavor preference" is mostly "sweet", and the "recommendation acceptance" is 8. Furthermore, based on the identification results, the promotion sub-model can determine the first user's recommended information, such as available promotional activities, coupons, and related products to purchase.

[0107] In this embodiment, the specific method by which the recommendation sub-model determines the recommendation information is not limited. Optionally, based on the above identification results, it can be preferentially determined whether the first user's account contains coupons corresponding to the three "flavor types" of "fries". If it is determined that the first user's account contains corresponding coupons, then the corresponding coupons and the access address of the specific product type "fries" that can be purchased are used as recommendation information. Correspondingly, if the first user's account does not contain coupons, then it is queried whether there are corresponding promotional activities for the product type "fries" based on the three "flavor types". If there are corresponding promotional activities, then it is further determined whether the first user is eligible to participate in the promotional activity. If it is determined that the first user is eligible to participate, then the access address of the corresponding activity is used as recommendation information.

[0108] Alternatively, if it's confirmed that the first user's account has a corresponding coupon, it's possible to further determine whether there are any promotional activities for the "French Fries" product type, and whether the first user is eligible to participate. If the first user is determined to be eligible, the corresponding coupon, the access address of the specific "French Fries" product type available for purchase, and the access address of the corresponding promotion can all be used as recommendation information. Alternatively, it's also possible to directly determine from the sales volume whether the first user has purchased a coupon for "Cheese Flavored" French Fries, whether there are any corresponding promotional activities for "Cheese Flavored" French Fries, and whether the first user is eligible to participate in the corresponding promotional activities, etc.; then, if one or more of the above information exists in the query, the retrieved information can be used as recommendation information. Of course, the above methods are only illustrative examples and are not limited to these in actual applications, and will not be detailed here.

[0109] Suppose that the recommendation sub-model, based on the user identifier "2088XYZW", finds that the first user's account contains "Coupon A", which can be used to purchase "Cheese Flavored" "French Fries Q". Furthermore, based on "French Fries Q", the corresponding access address is determined to be "http: / / XXXXXX". Based on this, the promotion sub-model can output the above information as a query result to the main model. After receiving the query result from the promotion sub-model, the main model can format it and concatenate it with parameter 4 to obtain parameter 5. Then, parameter 5 is returned to the first community as the first response information, allowing the first user to make a purchase based on this first response.

[0110] Assuming that this embodiment uses "Discount" to identify recommended information, "Purchasable" to identify purchasable products, and "Address" to identify access addresses, the query result information output by the promotion sub-model is formatted and concatenated with parameter 4 to obtain parameter 5 as shown below:

[0111]

[0112] It should be noted that in practical applications, when the main model inputs the corresponding input parameters to each first sub-model, it can only input the necessary information of each first sub-model as input parameters. In the embodiments of this specification, inputting all previously obtained information into the corresponding first sub-model is only an example. The specific form can be determined according to the processing requirements and is not limited here.

[0113] In one or more embodiments of this specification, before returning the first response information to the first community, the main model may also perform compliance verification on the first response information based on the verification sub-model to determine whether it includes information involving violations of laws and regulations or personal privacy information that is prohibited from being disclosed. Optionally, if the verification passes, the verification sub-model may directly output indication information to indicate that the verification has passed, so that the main model can return the original first response information to the first community; correspondingly, if the verification fails, the non-compliant content in the first response information may be filtered based on the shielding sub-model, and the filtered content may be output to the main model, so that the main model can return the output result as the new first response information to the first community.

[0114] Optionally, this embodiment may also provide an output sub-model for personalizing the first response information, thereby setting the output format of the first response information. For example, in the above example, if the first response information is verified to include personal privacy information, the verification sub-model may not perform filtering, but instead set the corresponding information to an output format that is "visible only to the first user" through the output sub-model, ensuring information security while also allowing the first user to confirm the relevant information.

[0115] It should be noted that the types and quantities of the first sub-models, the specific operations performed by each first sub-model, and the scheduling and information processing methods between the main model and the various first sub-models in the above examples are merely illustrative and are not limited to these in practical applications. For example, in the above examples, the mapping sub-model can also be split into two sub-models; optionally, it can be split into a user sub-model for determining user information and an order sub-model for determining historical purchase record information. As another example, the validation sub-model and the masking sub-model can also be merged into a single first sub-model, which can perform both compliance validation operations and invalid information filtering operations.

[0116] In practical applications, since different brands have their own brand characteristics, the output model can also set the output format of the first response information to adapt to the brand characteristics of the first tenant, so as to improve user experience and highlight the brand characteristics. For example, if the first tenant is mainly "anime-style", the output sub-model can set the output format of the first response information to "output in the form of a cartoon character's voice"; or if the first tenant is mainly "traditional Chinese style", the output sub-model can set the output format of the first response information to "output in the form of text or animation with a background of ink and wash landscape painting", and so on. Based on this, after receiving the first response information returned by the cloud server, the local server of the first tenant can perform output control, page rendering and other operations according to its corresponding output format. The specific implementation can be determined according to actual needs, and will not be detailed here.

[0117] In practical applications, since a community contains multiple users, each user may input queries within the community, and each user may input multiple queries. There may be correlations between multiple queries input by a single user or between multiple queries input by multiple users. Optionally, the correlation between multiple queries can be determined based on the query scenarios they correspond to. If the query scenarios corresponding to multiple queries are the same or similar, they are considered related; otherwise, they are considered unrelated. Of course, the specific determination method is not limited to this.

[0118] For example, if a user wants to inquire about a product from multiple perspectives, inputting multiple query messages, the first query message obtained by the main model can be a group message fragment within a first community. Based on this, the main model can perform context recognition on the group message fragment to determine the query scenario corresponding to each query message, thereby determining the relationship between the two. Furthermore, following the method in the above embodiment, the corresponding first response message is uniformly determined for these multiple query messages based on each first sub-model.

[0119] For example, consider a query input by one user and a reply input by another user, or similar queries input in the same context. In this case, the main model, in addition to obtaining the first query input from the first user within the first community, can also obtain the query input from the second user within the first community and use it as the second query. Optionally, the main model can obtain group message fragments within the first community and perform contextual recognition on these fragments to extract the first and second query information. Further, semantic recognition is performed on the first and second query information to determine their respective corresponding query scenarios and thus the relationship between them. Then, based on this relationship, the second reply information is obtained and returned to the first community.

[0120] In the above embodiments, the method by which the main model obtains the second response information is not limited. Optionally, if the main model determines that the first and second query information are unrelated, it can directly determine the response information corresponding to the first query information based on each of the first sub-models, and return this response information as the second response information to the first community, in accordance with the method described in the above embodiments. Correspondingly, if the main model determines that the first and second query information are related, it can determine whether the second query information can serve as the response information to the first query information, based on each of the first sub-models, in accordance with the method described in the above embodiments.

[0121] Optionally, if it is determined that the second query information is not accurate enough as a response to the first query information, the response information determined by each first sub-model can be returned to the first community as the second response information, as described in the above embodiments; or, if it is determined that the second query information can be used as a response to the first query information, an affirmative indication, such as "yes" or "correct," can be returned to the first community as the second response information to enhance the sense of interaction within the community.

[0122] It should be noted that, in the above embodiments, the methods for determining whether multiple query messages within the first community are related and for determining the response information based on the relationship between multiple query messages are merely illustrative examples and are not limited to these in practical applications. The above embodiments are intended to illustrate that the embodiments of this application support determining the response information based on the relationship between multiple query messages and do not constitute specific limitations.

[0123] The following describes the overall execution flow of the community operation method based on a large model provided in one or more embodiments of this specification, with reference to the accompanying drawings.

[0124] Figure 2b The flowchart illustrates another community operation method based on a large model, as provided in the embodiments of this specification. In this example, the method is still executed by a cloud server and the first user asks "How many flavors of fries are there?".

[0125] like Figure 2b As shown, in this embodiment, the cloud server can obtain the first query information input by the first user within the first community, and then input the first query information into a verification sub-model and a masking sub-model. The verification sub-model verifies the compliance of the first query information, and the masking sub-model filters out invalid information from the first query information. Figure 2b As shown, if the validation sub-model fails the validation, it returns a prompt message to the requesting end, indicating that the request does not comply with the regulations. If the blocking sub-model determines that the first query information includes invalid information, it can withdraw the first query information, return a warning message to the requesting end, or, based on the number of times the first user entered invalid information, perform actions such as muting, kicking out of the community, or blocking. Furthermore, if the first query information passes the validation sub-model and the blocking sub-model confirms that the first query information is not invalid, such as... Figure 2b As shown, the first query information that passes the verification is input into the main model.

[0126] Furthermore, based on the main model, semantic recognition can be performed on the first query information to determine whether it is a complaint or suggestion. If it is, the processing sub-model determines a specific solution and returns it to the requesting client. Conversely, if it is not, the system further determines the query scenario corresponding to the first query information and the first sub-models associated with at least one query dimension, as well as the association order between all the first sub-models. Then, according to this association order, the query result information determined by the previous first sub-model is used as input parameters to the next first sub-model, and the query result information of each first sub-model is obtained sequentially.

[0127] like Figure 2b As shown, after identification by the main model, the first sub-models sequentially associated with the first query information are identified as the product sub-model, the mapping sub-model, and the promotion sub-model. Based on this, the main model can, according to this association order, first input the formatted first query information into the product sub-model, so that the product sub-model can determine the product information corresponding to the first query information and output it as the query result. Further, the main model can obtain the product information determined by the product sub-model, format it, and input the processing result as an input parameter into the mapping sub-model. Based on this, the mapping sub-model can determine the user information corresponding to the first user who input the first query information, as well as the mapping information between the user information and the product information, and output it as the query result. Further, the main model can obtain the mapping information determined by the mapping sub-model, format it, and input the processing result as an input parameter into the promotion sub-model. Based on this, the promotion sub-model can determine the promotional activities that the first user can participate in, the coupons that can be used, the related products that can be purchased, and other recommended information based on the mapping information, and output it as the query result.

[0128] like Figure 2b As shown, after the main model obtains the query results output by all the sub-models mentioned above, it can also determine whether to set the output format of the response information through the output sub-model. If so, the output sub-model sets the corresponding output format based on the brand characteristics of the first tenant corresponding to the first community. Furthermore, as... Figure 2b As shown, the compliance of the above query results can also be verified through a validation model to ensure that the response information returned to the first community is compliant and does not involve sensitive information. Based on this, after the above validation is passed, the main model can return the query result information determined by all sub-models as the first response information corresponding to the first query information to the first community.

[0129] It should be noted that, Figure 2b The examples shown are intended to illustrate the overall process of the community operation method provided in the embodiments of this specification. The specific execution details of each step can be found in the description of the corresponding part in the above embodiments, and will not be repeated here.

[0130] Based on the above, the community operation method based on a large model provided in one or more embodiments of this specification can acquire inquiry information raised by users within offline tenant communities, perform semantic recognition on the corresponding inquiry information based on a pre-trained main model, and determine from which inquiry dimensions the user raised the corresponding inquiry information. Then, based on sub-models pre-trained under each inquiry dimension, the inquiry result information of the above inquiry information under the corresponding inquiry dimension is determined. Based on this, the main model can return the inquiry result information determined by all sub-models from different inquiry dimensions as the reply information of the above inquiry information to the corresponding community for users to view. In this method, the automatic identification and timely feedback of various inquiry information raised by users in the community based on each model not only simulates the manual operation method, reducing labor costs, but also provides users with timely service quickly and efficiently, helping to maintain users' trust in the brand, improve user retention rate, and provide a foundation for user conversion for other derivative transactions.

[0131] It is understood that the above embodiments are merely examples, and modifications can be made to the above embodiments in actual implementation. Those skilled in the art will understand that any modifications to the above embodiments that do not require creative effort fall within the protection scope of one or more embodiments of this specification, and will not be described again in the embodiments.

[0132] Based on the same inventive concept, one or more embodiments of this specification also provide a community operation device based on a large model. Figure 3 A structural diagram of the community operation device, such as Figure 3 As shown, the community operation device 300 includes: a first acquisition module 301, a first determination module 302, a second determination module 303, a second acquisition module 304, and an output module 305, wherein:

[0133] The first acquisition module 301 is used to acquire the first query information input by the first user in the first community; the first determination module 302 is used to determine at least one query dimension corresponding to the first query information based on the main model; wherein, the main model is obtained by training the model with the first training data, and the first training data is historical query information under different query scenarios; the second determination module 303 is used to determine the first sub-models associated with at least one query dimension and the association order between all the first sub-models; wherein, each first sub-model is obtained by training the model with the second training data under one query dimension, and the second training data is historical question-answer pair information under the corresponding query dimension, and different first sub-models are used to determine the query result information of the first query information under the corresponding query dimension; the second acquisition module 304 is used to acquire the query result information determined by each first sub-model in sequence according to the association order; the output module 305 is used to return the query result information determined by all the first sub-models as the first reply information to the first community.

[0134] In an optional embodiment, the first determining module 302 determines at least one query dimension corresponding to the first query information based on the main model, and is used to: input the first query information into the main model; and obtain at least one query dimension determined by the main model after performing semantic recognition on the first query information.

[0135] In an optional embodiment, the first determining module 302 determines at least one query dimension corresponding to the first query information based on the master model, for the purpose of: performing compliance verification on the first query information; and determining at least one query dimension corresponding to the first query information based on the master model after the verification is passed.

[0136] In an optional embodiment, the second determining module 303 determines the first sub-models associated with at least one query dimension and the association order among all the first sub-models, for: determining the first query scenario corresponding to the first query information based on the semantic recognition result of the first query information by the main model, wherein the first query scenario corresponds to at least one query dimension; obtaining pre-configured mapping information for the first query scenario; the mapping information includes the second sub-models associated with all the query dimensions corresponding to the first query scenario and the association order among all the second sub-models, wherein all the second sub-models include all the first sub-models; and determining the first sub-models associated with at least one query dimension and the association order among all the first sub-models based on the mapping information.

[0137] In an optional embodiment, the second acquisition module 304 sequentially acquires the query result information determined by each first sub-model according to the association order, for the following purposes: determining the first first sub-model in the association order; inputting the first query information into the first first sub-model and acquiring the query result information determined by the first first sub-model; and sequentially inputting the query result information determined by the previous first sub-model as input parameters into the next first sub-model according to the association order between the first first sub-model and other first sub-models, thereby sequentially acquiring the query result information of each first sub-model.

[0138] In an optional embodiment, the second acquisition module 304 inputs the first query information into the first sub-model and acquires the query result information determined by the first sub-model, for the following purposes: formatting the first query information according to the semantic recognition result of the first query information by the main model; inputting the formatted first query information as an input parameter into the first sub-model; and acquiring the query result information determined by the first sub-model.

[0139] In an optional embodiment, the second acquisition module 304 sequentially inputs the query result information determined by the previous first sub-model as input parameters into the next first sub-model according to the association order between the first first sub-model and other first sub-models, and sequentially acquires the query result information of each first sub-model. This is used to: sequentially input the formatted processing result of the query result information determined by the previous first sub-model as input parameters into the next first sub-model according to the association order between the first first sub-model and other first sub-models, and sequentially determine the query result information of each first sub-model.

[0140] In an optional embodiment, the output module 305 returns all the query result information determined by the first sub-model as the first reply information to the first community, for the purpose of: performing compliance verification on all the query result information determined by the first sub-model; and after the verification is passed, returning all the query result information determined by the first sub-model as the first reply information to the first community.

[0141] In an optional embodiment, the first acquisition module 301 is further configured to: acquire second query information input by a second user in the first community; determine the association between the first query information and the second query information; acquire second reply information according to the association and return the second reply information to the first community.

[0142] It should be noted that the specific functions and processing principles of each module in the above-mentioned community operation device based on the large model can be found in the description of the corresponding part in the above method embodiment, and will not be repeated here.

[0143] Based on the same inventive concept, one or more embodiments of this specification also provide a computer device. Figure 4 This is a block diagram of the computer device. For example... Figure 4 As shown, the computer device 400 includes a processor 401 and a memory 402, wherein the processor 401 is used to perform the following operations:

[0144] Obtain the first query information entered by the first user within the first community;

[0145] Based on the master model, at least one query dimension corresponding to the first query information is determined; wherein, the master model is obtained by training the model from the first training data, and the first training data is historical query information under different query scenarios;

[0146] Determine the first sub-model associated with at least one query dimension and the association order among all the first sub-models; wherein each first sub-model is obtained by training a model with second training data under a query dimension, the second training data being historical question-answer pair information under the corresponding query dimension, and different first sub-models are used to determine the query result information of the first query information under the corresponding query dimension;

[0147] Based on the association order, the query result information determined by each first sub-model is obtained sequentially;

[0148] All query results determined by the first sub-model are returned to the first community as the first response information.

[0149] It should be noted that the functional principles of each component in the above-mentioned computer equipment are the same as the implementation principles of the corresponding steps in the community operation method based on the large model. For details, please refer to the description of the corresponding part in the above method embodiment, which will not be repeated here.

[0150] In some alternative implementations, processor 401, sometimes also referred to as controller or operation control, may include a microprocessor or other processor device and / or logic device, which receives input and controls the operation of various components of computer device 400.

[0151] The memory 302 can be coupled to the processor 401. The memory 302 can be one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It can store programs related to the execution of the community operation method based on the large model. The processor 401 can execute the program stored in the memory 302 to perform information storage or processing, etc.

[0152] Memory 302 can be a solid-state memory, such as read-only memory (ROM), random access memory (RAM), SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROM, etc. Memory 302 can also be some other type of device. Memory 302 includes buffer memory (sometimes referred to as a buffer). Memory 302 may include an application / function storage unit for storing application programs and function programs or processes for executing operations of computer device 400 via processor 401.

[0153] The memory 302 may also include a data storage unit for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the device. The driver storage unit of the memory 302 may include various drivers for the computer device for communication functions and / or for performing other functions of the computer device (such as messaging applications, address book applications, etc.).

[0154] It is worth noting that, Figure 4 The component shown is exemplary, and other types of structures can be used to supplement or replace it to achieve telecommunications or other functions. Of course, computer equipment 400 is not necessarily required to include it. Figure 4 All components shown; in addition, computer device 400 may also include Figure 4 For components not shown, please refer to existing technologies.

[0155] In some alternative implementations, the computer device 400 may also include components such as a memory 302, a communication module, an input unit, an audio processor, a display, and a power supply.

[0156] An input unit can provide input to the processor 401. This input unit may be, for example, a keypad or touch input device. A power supply can be used to provide power to the computer device 400. A display can be used to display images and text, etc. This display may be, for example, an LCD display, but is not limited to this.

[0157] The communication module is a transmitter / receiver that sends and receives signals via an antenna. The communication module (transmitter / receiver) is coupled to the processor 401 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0158] Based on different communication technologies, multiple communication modules can be configured in the same computer device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) is also coupled to a speaker and microphone via an audio processor to provide audio output through the speaker and receive audio input from the microphone, thereby enabling typical telecommunications functions. The audio processor may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor is coupled to processor 401, enabling on-device recording via the microphone and on-device playback of stored sound via the speakers.

[0159] One or more embodiments of this specification also provide a computer-readable storage medium capable of implementing all steps of the large-model-based community operation method in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the large-model-based community operation method in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0160] Obtain the first query information entered by the first user within the first community;

[0161] Based on the master model, at least one query dimension corresponding to the first query information is determined; wherein, the master model is obtained by training the model from the first training data, and the first training data is historical query information under different query scenarios;

[0162] Determine the first sub-model associated with at least one query dimension and the association order among all the first sub-models; wherein each first sub-model is obtained by training a model with second training data under a query dimension, the second training data being historical question-answer pair information under the corresponding query dimension, and different first sub-models are used to determine the query result information of the first query information under the corresponding query dimension;

[0163] Based on the association order, the query result information determined by each first sub-model is obtained sequentially;

[0164] All query results determined by the first sub-model are returned to the first community as the first response information.

[0165] It should be noted that the functional principle of the aforementioned computer-readable storage medium is the same as the implementation principle of the corresponding steps in the above community operation method based on a large model. For details, please refer to the description of the corresponding part in the above method embodiment, which will not be repeated here.

[0166] While one or more embodiments of this specification provide method operation steps as shown in the embodiments or flowcharts, more or fewer operation steps may be included based on conventional or non-inventive labor. The order of steps listed in the embodiments is merely one possible execution order among many and does not represent the only execution order. In actual device or client product execution, the method can be executed sequentially as shown in the embodiments or drawings, or in parallel (e.g., in a parallel processor or multi-threaded processing environment).

[0167] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, apparatus (systems), or computer program products. Therefore, the embodiments of this specification can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, one or more embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0168] This specification describes one or more embodiments of a method, apparatus (system), and computer program product according to one or more embodiments of this specification with reference to flowchart illustrations and / or block diagrams. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart illustrations and / or block diagrams.

[0169] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0170] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps for implementing the functions specified in one or more flowcharts and / or one or more block diagrams.

[0171] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and system embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0172] In this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, without necessarily requiring or implying any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Those skilled in the art will understand the specific meaning of the above terms in one or more embodiments of this specification, depending on the specific circumstances.

[0173] It should be noted that, unless otherwise specified, one or more embodiments and features thereof in this specification can be combined with each other. This specification is not limited to any single aspect, nor to any single embodiment, nor to any combination and / or substitution of such aspects and / or embodiments. Furthermore, each aspect and / or embodiment of one or more embodiments of this specification can be used alone or in combination with one or more other aspects and / or embodiments thereof.

[0174] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of one or more embodiments of this specification, and are not intended to limit them. Although one or more embodiments of this specification have been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of one or more embodiments of this specification, and they should all be covered within the scope of one or more embodiments of this specification.

[0175] The foregoing description of one or more embodiments of this specification has been provided in conjunction with optional implementation methods. However, these embodiments are merely exemplary and serve only an illustrative purpose. Based on this, various substitutions and modifications can be made to one or more embodiments of this specification, all of which fall within the protection scope of one or more embodiments of this specification.

Claims

1. A community operation method based on a large model, characterized in that, include: Obtain the first query information entered by the first user within the first community; Based on the main model, at least one query dimension corresponding to the first query information is determined; The main model is obtained by training the model using the first training data, which is historical query information under different query scenarios. Determine the first sub-models associated with each of the at least one query dimension and the association order among all the first sub-models; each first sub-model is obtained by training a model with second training data under a query dimension, the second training data being historical question-answer pair information under the corresponding query dimension, and different first sub-models are used to determine the query result information of the first query information under the corresponding query dimension. According to the association order, the query result information determined by each first sub-model is obtained sequentially; All query results determined by the first sub-model are returned to the first community as the first response information.

2. The method according to claim 1, characterized in that, Based on the main model, at least one query dimension corresponding to the first query information is determined, including: Input the first query information into the main model; Obtain at least one query dimension determined by the main model after performing semantic recognition on the first query information.

3. The method according to claim 1, characterized in that, Based on the main model, at least one query dimension corresponding to the first query information is determined, including: Perform compliance verification on the first query information; After the verification is passed, at least one query dimension corresponding to the first query information is determined based on the master model.

4. The method according to claim 2, characterized in that, Determining the first sub-model associated with each of the at least one query dimension and the association order among all the first sub-models includes: Based on the semantic recognition result of the first query information by the main model, the first query scenario corresponding to the first query information is determined, and the first query scenario corresponds to the at least one query dimension; Obtain pre-configured mapping information for the first query scenario; the mapping information includes the second sub-models associated with all query dimensions corresponding to the first query scenario, and the association order between all the second sub-models, all of which include all the first sub-models; Based on the mapping information, determine the first sub-model associated with each of the at least one query dimension and the association order among all the first sub-models.

5. The method according to claim 4, characterized in that, Based on the aforementioned association order, the query result information determined by each first sub-model is obtained sequentially, including: Based on the association order, determine the first sub-model in the association order; The first query information is input into the first sub-model, and the query result information determined by the first sub-model is obtained; According to the association order between the first sub-model and other first sub-models, the query result information determined by the previous first sub-model is used as the input parameter to input the next first sub-model in turn, and the query result information of each first sub-model is obtained in turn.

6. The method according to claim 5, characterized in that, The first query information is input into the first sub-model, and the query result information determined by the first sub-model is obtained, including: Based on the semantic recognition result of the first query information by the main model, the first query information is formatted. The formatted first query information is used as input parameter into the first sub-model. Obtain the query result information determined by the first sub-model.

7. The method according to claim 6, characterized in that, Following the association order between the first sub-model and other first sub-models, the query result information determined by the previous first sub-model is sequentially input into the next first sub-model as an input parameter, and the query result information of each first sub-model is sequentially obtained, including: According to the association order between the first sub-model and other first sub-models, the formatted processing result of the query result information determined by the previous first sub-model is used as the input parameter to input the next first sub-model, and the query result information of each first sub-model is determined in turn.

8. The method according to any one of claims 1-7, characterized in that, All query results determined by the first sub-model are returned to the first community as the first response information, including: Perform compliance verification on all query result information determined by the first sub-model; After the verification is passed, all the query results determined by the first sub-model are returned to the first community as the first response information.

9. The method according to any one of claims 1-7, characterized in that, Also includes: Obtain the second query information input by the second user within the first community; Determine the correlation between the first query information and the second query information; Based on the aforementioned relationship, obtain the second response information and return the second response information to the first community.

10. A community operation device based on a large model, characterized in that, include: The first acquisition module is used to acquire the first query information input by the first user in the first community. The first determining module is used to determine at least one query dimension corresponding to the first query information based on the main model; The main model is obtained by training the model using the first training data, which is historical query information under different query scenarios. The second determining module is used to determine the first sub-models associated with the at least one query dimension and the association order among all the first sub-models; each first sub-model is obtained by training the model with second training data under a query dimension, the second training data being historical question-answer pair information under the corresponding query dimension, and different first sub-models are used to determine the query result information of the first query information under the corresponding query dimension. The second acquisition module is used to sequentially acquire the query result information determined by each first sub-model according to the association order; The output module is used to return all the query results determined by the first sub-model to the first community as the first response information.

11. The apparatus according to claim 10, characterized in that, The first determining module, based on the main model, determines at least one query dimension corresponding to the first query information, for the purpose of: Input the first query information into the main model; Obtain at least one query dimension determined by the main model after performing semantic recognition on the first query information.

12. The apparatus according to claim 10, characterized in that, The first determining module, based on the main model, determines at least one query dimension corresponding to the first query information, for the purpose of: Perform compliance verification on the first query information; After the verification is passed, at least one query dimension corresponding to the first query information is determined based on the master model.

13. The apparatus according to claim 11, characterized in that, The second determining module determines the first sub-model associated with each of the at least one query dimension and the association order among all the first sub-models, for the purpose of: Based on the semantic recognition result of the first query information by the main model, the first query scenario corresponding to the first query information is determined, and the first query scenario corresponds to the at least one query dimension; Obtain pre-configured mapping information for the first query scenario; the mapping information includes the second sub-models associated with all query dimensions corresponding to the first query scenario, and the association order between all the second sub-models, all of which include all the first sub-models; Based on the mapping information, determine the first sub-model associated with each of the at least one query dimension and the association order among all the first sub-models.

14. The apparatus according to claim 13, characterized in that, The second acquisition module, according to the association order, sequentially acquires the query result information determined by each first sub-model, for the purpose of: Based on the association order, determine the first sub-model in the association order; The first query information is input into the first sub-model, and the query result information determined by the first sub-model is obtained; According to the association order between the first sub-model and other first sub-models, the query result information determined by the previous first sub-model is used as the input parameter to input the next first sub-model in turn, and the query result information of each first sub-model is obtained in turn.

15. The apparatus according to claim 14, characterized in that, The second acquisition module inputs the first query information into the first sub-model and acquires the query result information determined by the first sub-model, for the purpose of: Based on the semantic recognition result of the first query information by the main model, the first query information is formatted. The formatted first query information is used as input parameter into the first sub-model. Obtain the query result information determined by the first sub-model.

16. The apparatus according to claim 15, characterized in that, The second acquisition module, following the association order between the first sub-model and other sub-models, sequentially inputs the query result information determined by the previous sub-model as input parameters into the next sub-model, and sequentially acquires the query result information of each sub-model for: According to the association order between the first sub-model and other first sub-models, the formatted processing result of the query result information determined by the previous first sub-model is used as the input parameter to input the next first sub-model, and the query result information of each first sub-model is determined in turn.

17. The apparatus according to any one of claims 10-16, characterized in that, The output module returns all query results determined by the first sub-model as the first response information to the first community, for the purpose of: Perform compliance verification on all query result information determined by the first sub-model; After the verification is passed, all the query results determined by the first sub-model are returned to the first community as the first response information.

18. The apparatus according to any one of claims 10-16, characterized in that, The first acquisition module is also used for: Obtain the second query information input by the second user within the first community; Determine the correlation between the first query information and the second query information; Based on the aforementioned relationship, obtain the second response information and return the second response information to the first community.

19. A computer device, characterized in that, include: A memory and a processor, the memory storing computer program instructions, which, when executed by the processor, are used to implement the method described in any one of claims 1-9.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions, which, when executed, implement the method described in any one of claims 1-9.

Images